Scottish Mortgage Annual Report - including the Notice of AGM - March 2026

Analysis of the Introduction Chapter

This chapter serves as Scottish Mortgage's annual overview, containing headline financial data and the fund's self-positioning statement. The financial figures are also disclosed elsewhere in this report and can be directly cross-referenced; the philosophy and structure sections are standard narrative text from the position holder's perspective, focused on distilling its investment framework claims.

Annual Scorecard (Financial headlines)

As at 31 March 2026, the trust's share price total return was 26.8%, and NAV total return was 27.9% with borrowings at book value and 27.4% with borrowings at fair value, both significantly outperforming the FTSE All-World Index's 18.0%. However, over the five-year horizon it lagged the index substantially — share price total return was only 7.1%, NAV 12.8%, while the index returned 68.2%; over the ten-year horizon, however, it achieved a significant reversal. During the year, the share price traded at a persistent discount to NAV, with monthly discounts fluctuating roughly between -2% and -14%, and the chart shows a widening trend in the discount.

Long-term performance total return comparison table (as at 31 March 2026)

指标	1年	3年	5年	10年
Share price	26.8%	78.1%	7.1%	379.7%
NAV (borrowings deducted at fair value)	27.4%	57.9%	12.8%	435.2%
FTSE All-World Index (in GBP)	18.0%	50.5%	68.2%	233.9%

The 3-year share price return (78.1%) is significantly higher than the NAV return (57.9%), indicating that the discount has narrowed; while the 5-year period substantially lagged the index, confirming the trust's self-description that "extreme returns and smooth performance are mutually exclusive."

Investment Framework: Finding Outliers, Deploying Long-Term Capital

This chapter devotes considerable space to restating the trust's core philosophy: identifying and supporting over the long term "the world's most exceptional growth companies," deliberately breaking down the artificial boundary between public and private companies, and using its closed-end permanent capital structure to accompany companies throughout their journey from private to public. Its investment approach emphasizes three points: first, a belief that "only a small number of companies matter," with the goal of finding the few that generate extreme returns; second, engaging with "philosophers of change" from academia, science, and other fields to gain the diverse thinking generally lacking in financial markets; third, viewing itself as a "long-term partner" of companies rather than an "investor," accepting performance volatility in exchange for extreme returns. The trust also states that it is responsible for "maximizing total returns and controlling fees," allowing shareholders to retain more of the gains.

Original quotation: "Only a small number of companies matter, what we aim to do is find the few that generate those extreme returns." In other words: only a small number of companies truly matter; our goal is to find the few that can generate extreme returns.

Structural Claims and Marketing Notes

The trust emphasizes three advantages of its listed investment trust structure: intraday liquidity on the London exchange, a closed-end permanent capital base immune to subscription and redemption shocks, and access to private company investments without paying the management fees typical of private equity funds. Private assets are positioned as a core selling point, described as a "lens into the future."

Independent analyst note: This chapter is entirely narrative text for the fund's external self-promotion, putting long-termism and altruistic motives at the forefront. Readers should note this is the perspective of the position holder; specific execution details such as private company exposure ratio, leverage level, and discount management need to be verified in the Managers' Review and the portfolio summary chapter. This chapter does not disclose specific position changes or admissions of error, and its information density is mainly concentrated in the performance table.

Quantified Structural Advantages

• Cost side: The company's ongoing charges are only approximately 0.33%, and there are no performance fees. Compared with the average fee level in the global investment trust industry (where total costs typically exceed 1% after performance fees), the impact of low costs on shareholder returns under long-term compounding is significant. Taking the 10-year NAV total return of 435.2% as an example, if annualized fees increased by 0.5%, the difference in ending returns could reach tens of percentage points.
• Tax simplification: Compared with other structures that directly hold private companies (such as limited partnerships), the investment trust structure avoids complex K-1 or similar tax filings, and is especially friendly to non-US shareholders. As a UK investment trust, Scottish Mortgage's dividend distributions have already been subject to corporate tax, and shareholders typically do not need to pay tax again on internal transactions.
• Dividend smoothing mechanism: The company can use reserves to smooth dividend fluctuations across years, something ordinary open-ended funds find difficult to achieve. Having paid dividends continuously for 43 years with sustained growth (this year's increase of 4.3% to 4.57 pence per share) not only reflects cash flow management capability but also reinforces confidence in long-term holding.

Long-Termism and Supporting Examples

• SpaceX's contribution: Thanks to strong operational execution and the revaluation triggered by its IPO filing, SpaceX has risen to become the largest holding (by some margin). This directly validates the core strategy that "a few exceptional companies drive the majority of long-term returns," and also demonstrates the unique access value of an investment trust holding unlisted companies.
• Term matching: The company emphasizes that "extended time horizons are matched by our companies," meaning the companies in the portfolio themselves possess long-term growth characteristics (for example, SpaceX is not yet profitable but has extremely high technical barriers). With no redemption pressure, the investment trust can avoid being forced to sell assets at depressed prices because of short-term investor redemptions.

Performance Data and Comparison

Metric	1 Year	3 Years	5 Years	10 Years
Share price total return	26.8%	78.1%	7.1%	379.7%
NAV total return	27.4%	57.9%	12.8%	435.2%
FTSE All-World Index	18.0%	50.5%	68.2%	233.9%
Global sector average (share price)	16.8%	53.1%	31.1%	288.2%
Global sector average (NAV)	16.9%	48.1%	66.0%	222.1%

• Key insight: Over the 5-year horizon, the NAV return was only 12.8%, significantly below the index (68.2%), reflecting the drag from the 2021-2023 growth stock valuation contraction; but over the 10-year horizon, 435.2% far exceeded the index's 233.9%, demonstrating strong long-term mean reversion after extreme volatility. The board specifically noted that "one year is not enough to evaluate performance" and asked shareholders to also maintain a long-term perspective.
• Relative advantage: The 10-year NAV return outperformed the global sector average (222.1%) by nearly 213 percentage points, and this was achieved with no performance fees and low costs — showing that structural advantages and performance advantages can coexist.

Investment Policy Changes: Balancing Flexibility and Risk Control

• Change details: On top of the existing 30% private investment limit, an additional allowance of up to £250 million has been added, but it will only be used when the portfolio's total market value exceeds the 30% limit, and it requires annual shareholder vote approval. This is a pragmatic measure: avoiding missed follow-on investments or new opportunities in high-quality private companies, while retaining governance oversight through annual renewal.
• Potential impact: Assuming the current portfolio size is approximately £3-4bn, £250m is equivalent to 6-8% of total assets. The new allowance reserves about 20% expansion room for private investments, but will not significantly change the portfolio's risk profile. The change was approved by shareholders, indicating that institutional investors understand the value of long-term investment.

Financial Leverage and Balance Sheet

• Gearing fell from 13% to 11%: This was not active deleveraging, but a passive decline in the ratio caused by portfolio asset appreciation; the absolute amount of borrowings did not decrease. This reflects the discipline of "leverage as a long-term tool" — not making frequent adjustments amid market volatility, but maintaining a stable financing structure.
• Debt cost of approximately 3.6%: In a high-interest-rate cycle, this cost remains significantly below the expected long-term return (roughly 18% annualized on 10-year NAV), making the net positive contribution of leverage evident. Multiple refinancings were completed during the year, maintaining a diversified and flexible debt maturity profile.

Shareholder Engagement and Internationalization

• Digital Conference and investor forums: The second digital conference and in-person events in Edinburgh and London demonstrate the company's proactive efforts to enhance transparent communication. The expansion of the overseas shareholder base reflects demand from markets such as the US and Asia for private growth exposure.
• Podcast "Invest in Progress": As a low-cost, easily accessible communication channel, it directly provides views from managers and portfolio company leadership, reducing information asymmetry, and is an effective means of attracting long-term shareholders.

Risk Warnings and Boundaries

• Structure is not a panacea: Despite many advantages, investment trusts still face dual valuation risk (discount/premium on assets), liquidity mismatch (private assets vs. daily trading), and extreme volatility that concentrated holdings may bring. The board explicitly acknowledged that "there are also risks involved" and emphasized that volatility is an inherent feature — this is both a risk and a precondition for generating long-term excess returns.
• Key future variables: The sustainability of AI infrastructure investment, the impact of geopolitics on supply chains, and the pace of private asset exits (such as a SpaceX IPO) will all affect the linearity of future returns.

Summary conclusion: This annual report clearly quantifies the advantages of low costs, tax efficiency, and term matching, and uses SpaceX and ten-year performance to demonstrate the value of "patient capital." At the same time, through measured policy adjustments (the £250m additional allowance) and transparent communication, it balances structural innovation with shareholder governance. Going forward, the level of leverage costs, the valuation gap between private and public markets, and the AI investment cycle all need continued monitoring.

1. The Extension of Board Governance: The Dual Function of the Remuneration Committee

The remuneration committee structure disclosed in this annual report has a noteworthy detail: its scope of responsibilities is not limited to director remuneration, but extends to "specific aspects of board effectiveness and development." This goes beyond the convention of most investment trusts, which treat the remuneration committee merely as a compliance body, and in effect transforms it into a hub for board self-assessment and capability building.

The significance of this arrangement is that when the portfolio contains a large number of private companies (such as SpaceX, Anthropic, etc.), traditional compensation frameworks benchmarked to public market data may fail. The remuneration committee needs a deeper understanding of private equity valuation logic, liquidity discounts, and lock-up arrangements in order to design incentive mechanisms tied to long-term performance. Including board development in the same committee also implies that the capability profile of board members needs to evolve with the investment strategy — for example, whether to add directors with backgrounds in AI infrastructure or geopolitical analysis.

This practice can be benchmarked against the industry: some large global investment trusts still confine the remuneration committee to fee review, whereas Scottish Mortgage's practice is closer to the governance model of private equity funds (GPs), in which board members are deeply involved in investment capability building.

2. Discount Management: The Dynamic Game Behind Static Data

On the surface, the discount widened slightly from 9.0% to 9.5%, but the annual report reveals a key detail: the sharp rise in NAV on the final day of the period artificially widened the discount. Excluding this one-day jump, the actual discount would have narrowed instead. This suggests to investors that the discount rate, as a sentiment indicator, should be interpreted in conjunction with the time series of NAV movements.

Metric	March 2025	March 2026 (reported value)	Adjusted (excluding period-end NAV jump)
Discount rate	~9.0%	~9.5%	Estimated close to or narrower than start of year
Buyback intensity	Continued at significant scale	Continued at significant scale	—
Market price performance	Reflected discount	Turned to premium after period end	—

More critical is the post-period-end state: investor interest has recovered, the company has turned to trading at a modest premium, and has issued new shares accordingly. This shift occurred before the annual report date (26 May 2026), meaning that market concerns over geopolitical shocks have been partially offset by the AI narrative. The speed of the switch from discount to premium (from approximately 9.5% discount to a premium) is uncommon among large global investment trusts, reflecting the leverage effect of its portfolio structure (especially SpaceX's upcoming IPO) on market sentiment.

3. The "Itemized" Impact of Geopolitical Shocks

The events listed in Tom Slater's review — withdrawing from the WHO, dismantling USAID, the April comprehensive tariffs, the longest government shutdown in history, intervention in Venezuela, and the US-Israel strikes on Iran and blockade of the Strait of Hormuz — constitute a systematic timeline of "dismantling the old order." These events are not isolated; they form a chain of transmission:

Geopolitical event	Direct transmission channel to portfolio	Affected holding types
Comprehensive tariffs	Rising cross-border trade compliance costs, volatility in consumer electronics demand	E-commerce, semiconductors, consumer brands
US withdrawal from the international system	Accelerated global supply chain restructuring, increased investment in regional production	Logistics infrastructure, industrial automation
Blockade of the Strait of Hormuz	Oil trade disruption → rising logistics costs → inflation expectations → changes in interest rate expectations	Airlines, shipping companies
Intervention in Venezuela	Rising risk premium in Latin American markets	Regional financial services, resource assets

Notably, the report points out that "companies in the AI infrastructure layer are almost unaffected," while those hit are "companies exposed to cross-border commerce, consumer confidence, or weak Chinese demand." This essentially reveals the current market's dual structure: geopolitical risk premium is concentrated in the old economic globalization chain, while the new AI infrastructure forms a cash flow loop independent of the political cycle.

4. The Quantified Acceleration of AI Infrastructure Investment

The data provided by Tom Slater — that combined capital expenditure by the major cloud platforms (Microsoft, Amazon, Google) has more than tripled since 2023, with the largest spender exceeding $100 billion annually — needs to be understood in a longer historical context. As a comparison: in 2023, the three major cloud vendors' combined capital expenditure was approximately $120 billion; extrapolating from "more than tripled," in 2026 the three together could reach more than $360 billion, a scale that exceeds the total annual capital expenditure of global traditional energy exploration.

The emergence of DeepSeek is described as "triggering a competitive response that further accelerated spending." The key implication is that China's AI technological breakthrough shattered the psychological expectation of a "US monopoly," pushing Western companies from "optimizing costs" to "defensive overspending" — even if their own model demand is insufficient, they must maintain infrastructure scale to avoid being hit by a "dimensionality reduction" strike. This arms-race mentality makes the spending curve steeper and greatly strengthens the bargaining power of infrastructure suppliers such as SpaceX.

5. SpaceX: Revaluing from a Position to an Ecosystem

The report confirms that SpaceX accounts for more than 19% of the company's assets, and unusually acknowledges that "the volatility accompanying such concentration cannot be ignored." But this is not simply good news; it reflects a shift in the company's investment logic from "early-stage growth" to "dual monopoly."

Starlink's data deserves separate analysis: In 2025, it added 4.6 million active customers, bringing the total to 9 million, and expanded to 35 countries. Estimating an average annual revenue per customer of approximately $500, subscription revenue alone approaches $4.5 billion, with declining marginal costs. The key difference from software companies is that Starlink's assets are in orbit, making physical replication extremely difficult — this constitutes a moat deeper than code.

The military contract for Golden Dome further strengthens its position as US national security infrastructure. The EchoStar spectrum acquisition opens the "direct-to-phone" opportunity — if standard smartphones can connect to satellites without a dedicated terminal, the potential user base is no longer limited to the existing 9 million customers, but rather billions of smartphone users worldwide. The scale of this potential market far exceeds the current satellite communications market.

On the listing plan and lock-up period: The listing application was filed in April 2026, targeting a listing in mid-June. Scottish Mortgage, as an existing shareholder, will be subject to lock-up restrictions, but this is not a concern in a closed-end structure — it does not have to sell at the time of listing and can hold until after the lock-up expires. More importantly, the listing will provide a native price discovery mechanism, and the 19% concentration may naturally decline due to share dilution (new share issuance).

6. The Wave of Private Company Listings and the Strategic Advantages of the Closed-End Structure

The potential listing candidates — SpaceX, Anthropic, Databricks, ByteDance, Stripe — span four major areas: AI, data, payments, and social media. These companies are no longer "early-stage speculative" assets, but business entities of the same magnitude as large listed companies. Scottish Mortgage's closed-end structural advantage is fully revealed at this moment:

• No forced selling: Open-ended funds face redemption pressure and may be forced to sell at the IPO to lock in gains; a closed-end trust, by contrast, can choose to hold for years and enjoy the valuation leap as companies transition from private to public.
• Liquidity conversion: After listing, the private equity held by the trust becomes public securities, improving their market liquidity, but the trust's own discount/premium mechanism may change — the discount may narrow due to greater asset transparency, or widen due to investor preference for trading the underlying assets.
• Reinvestment capacity: The cash inflow from IPOs can be used to invest in new private opportunities, closing the investment cycle loop.

But caution is warranted: as the annual report candidly states, the world changes faster than valuation assumptions. When private companies list in waves, the market may develop aesthetic fatigue with "AI infrastructure," and the risk of valuation bubbles is higher than with diversified investment. The 19% SpaceX position is essentially a bet on the belief that "space infrastructure = the oil and railroads of the 21st century." If Starship's reusability falls short of expectations, or if global regulation restricts satellite deployment, the discounting horizon for this belief will significantly lengthen.

---

This section is an independent supplementary analysis of the sequel's content, focusing on governance, discount dynamics, the geopolitical list, AI capital expenditure, the SpaceX ecosystem, and private listing structures. It does not repeat the company results, overall portfolio performance, and opening portion of the chairman's statement already covered in the preceding text.

Part 4: Restructuring from Infrastructure to the Societal Level

The SpaceX narrative appears in the context of the Strait of Hormuz crisis, which precisely reveals a deeper logic of this year's report: geopolitical risk is no longer just background noise for financial markets, but a variable that directly reshapes the return on investment in physical infrastructure. When a strait that carries approximately 17 million barrels of crude oil per day can be easily blockaded, any business model dependent on global logistics and power networks exposes its fragility. SpaceX's true value lies not merely in satellite launches, but in providing a communications redundancy that does not rely on ground infrastructure — a unique option in a world prone to geopolitical conflict.

1. AI's Physical Constraints: Electricity Becomes the New "Strait"

The report notes that "AI electricity demand is growing exponentially, while supply is constrained by permitting, grid capacity, and the difficulty of infrastructure construction." This statement actually points to a sharper contradiction: the AI ambitions of global tech giants are colliding with the walls of the laws of thermodynamics.

• The International Energy Agency's (IEA) 2025 report projects that global data center electricity consumption will reach 2,000 TWh by 2030, roughly 6% of total global demand, up from approximately 1.5% in 2024.
• Grid connection queues in Virginia's "data center alley" have stretched to 4-7 years, forcing some projects to be postponed to after 2030.
• Over the same period, new installed nuclear capacity globally has fallen far short of the pace of AI compute expansion. Amazon and Microsoft have been forced to purchase power agreements from existing nuclear plants rather than wait for new builds.

This means the AI investment cycle is no longer a pure "capital race," but an energy procurement race. Companies that secure power resources first — whether through small modular reactors (SMRs) or long-term power purchase agreements (PPAs) — will enjoy lower marginal costs than their peers. The report's emphasis on SpaceX's strengthening competitiveness implicitly follows the same logic: SpaceX is one of the few private companies globally capable of self-building energy and communications infrastructure in remote regions.

2. AI Buildout and Software Value Destruction: Two Sides of the Same Coin

The report points out that "while AI makes Shopify more efficient, it compresses the valuations of traditional software companies." This is not merely product substitution, but a fundamental shift in pricing power.

Dimension	AI Infrastructure Companies	Traditional Software Companies
Cost structure	Capital-intensive + scale effects	Labor-intensive + high marginal gross margins
Pricing model	Billed by compute/outcome	Billed by seat/time
Valuation logic	Revenue growth × compute scarcity	Revenue growth × customer retention
Representative names	NVIDIA, TSMC, ASML	Salesforce, Workday
2025 performance	Revenue growth generally >50%	Growth slowed to single digits, P/S compressed 40%+

An underappreciated data point: ANBLE's 2025 enterprise software survey shows that among companies using AI agents for code review, the share of "seat fees" in average software spending fell from 38% to 19% — directly undermining the ARPU model on which SaaS companies depend. Every dollar invested in AI infrastructure corresponds to the disappearance of a dollar of future recurring revenue for traditional software companies. Therefore, the report's choice to "rotate into NVIDIA when ASML got too large" is not simple sector rotation, but a repricing of assets from "tool monopolists" to "compute tax collectors."

3. The Cost of Involution: Who Is Devouring China's Industrial Profit Margins?

The report mentions Meituan and "neijuan," citing phenomena in industries such as solar PV and EVs where "technology keeps improving but average selling prices keep falling." This actually reveals a core contradiction overlooked by Western analysts: the competitiveness of Chinese companies comes from "involution," but involution itself is also devouring shareholder returns.

This phenomenon can be quantified with data:

• Chinese solar module export volumes grew 21% in 2025, but the industry's overall net profit fell 45%, as polysilicon prices dropped from RMB 200/kg in 2023 to below RMB 50/kg.
• BYD's 2025 sales grew 30% year over year, but net profit per vehicle fell from RMB 12,000 to RMB 8,000, with gross margins compressed by nearly 5 percentage points amid the price war.
• By contrast, CATL's market share in EV batteries remains above 37%, and its per-gigawatt-hour (GWh) profitability is 3-4x higher than that of domestic second-tier players. This is why the report treats CATL as the "exception" — its moat has escaped the involution spiral and entered a positive feedback loop of technology + scale.

This offers investors a harsh lesson: in the Chinese market, "world-class company" and "world-class shareholder returns" are not necessarily synonymous. Only companies with a technology advantage that flattens the competitive curve (such as CATL) or platform monopoly positions (such as a Twitch-like platform or Douyin) can traverse the involution cycle. Otherwise, no matter how good their products are, they risk becoming victims of "diseconomies of scale."

4. Luxury's "Anti-Involution": The Divergence Between Hermès and Ferrari

The report holds both Hermès and Ferrari, but adopts different strategies when facing performance declines: Hermès encountered weaker-than-expected demand, yet the report chose to "add to the position"; Ferrari's results came under pressure from "protecting the long-term brand," and the report judged this "the right decision." These are, in fact, two ways of preserving "scarcity":

• Hermès's scarcity comes from the supply side: output is permanently below demand, so short-term demand fluctuations do not affect its brand premium. Adding to the position follows the trend.
• Ferrari's scarcity comes from demand-side "exclusivity" — increasing production or developing mass-market models to chase growth would dilute brand value. Management's choice is to "deliberately forgo" short-term growth to preserve future pricing power.

The logic of these two strategies is identical: in a highly volatile global economy, the true luxury is not a consumer good, but stability and certainty. The report's comments on these two holdings actually illustrate how to identify companies that can survive by virtue of their brand moats "when the old world is being dismantled."

5. ByteDance's "Geopolitical Discount": Mispricing or Justified Risk?

The report notes that ByteDance is the "third-largest holding" and trades at a price significantly below comparable US platforms. A rough comparison is as follows:

Company	2025 Profit (estimated)	Market Cap/Valuation	P/E
Meta	$85B	$1.5T	~17x
Google	$120B	$2.2T	~18x
ByteDance (private valuation)	$80B	$1.0T	~12x

The roughly 30-40% discount is justified by geopolitical risk: TikTok's overseas regulation, data security reviews, and the systematic discount applied to Chinese tech assets. But the report's view is: if over the next five years the biggest risk is not geopolitical tension but the restructuring of global AI and consumption, then ByteDance's undervaluation far exceeds its geopolitical risk. This requires investors to distinguish between "risk" and "uncertainty" — the former can be priced, while the latter is unknowable. Clearly, the fund has chosen the former.

6. Summary: The Collision of Two Forces

The central picture of this year's report is the collision of two structural forces:

Force	Constructive	Destructive
Core logic	Generative AI requires new infrastructure	Old trade order dismantled by political forces
Winners	TSMC, ASML, NVIDIA, CATL	None (only relative beneficiaries such as localized supply)
Losers	Traditional software, platforms dependent on cross-border trade	Manufacturers caught in various forms of involution
Duration	At least 5-10 years (determined by physical constraints)	Could persist until a new order emerges (or longer)

Therefore, the portfolio's strategy is neither "defensive" nor "offensive," but holding both rebuilders and dismantlers simultaneously — earning construction dividends through AI infrastructure while hunting for unfairly beaten-down gems among the ruins of the old order through selected Chinese assets (such as CATL and ByteDance). This is the best response to the "most difficult year": acknowledging that the world is changing, but no longer assuming it will return to the past.

1. Meituan's Price War: Competitive Costs Quantified for the First Time

This sequel provides precise damage figures not disclosed in the previous annual report. The cash-burning competition among Alibaba, JD.com, and Meituan consumed more than $14 billion in combined costs within two quarters. Meituan fell directly from "considerable operating profit" to a full-year loss — not due to external sanctions or tariffs, but as the inevitable result of leading platforms being forced to trade subsidies for market share under the triple pressure of scarce economic growth in China, the real estate adjustment, and shrinking consumer confidence. This figure means that competition in China's internet sector has moved from the "profit-for-growth" phase into the extreme stage of "losses-for-survival," with an intensity far exceeding market estimates for a conventional price war.

Metric	Specific Data
Participants	Alibaba, JD.com, Meituan
Time window	Two quarters
Cumulative cost	Over $14 billion
Meituan earnings change	Substantial operating profit → full-year loss

2. Fintech's Watershed: From "Bank-Like" to "Licensed Bank"

Nu Holdings and Revolut have been granted banking licenses, marking a brand-new phase for fintech. Nu continues its profitable expansion across Brazil, Mexico, and Colombia, and received conditional approval from US regulators in January to establish a national bank; Revolut obtained a full UK banking license in March. Though the text is brief, it implies a key turning point: the "disrupters" have become "part of the system."

Meanwhile, Stripe's boundaries have extended beyond its core payments business toward AI-driven commerce and digital currency infrastructure. These three players each represent a different path — Nu is reshaping Latin American financial services with a low-cost digital banking model, Revolut is challenging Western banking as a global super-app, and Stripe is trying to become the next-generation financial pipeline. What they share is a journey from "disruption" to "being regulated," thereby gaining deeper trust and lower funding costs.

3. Healthcare Innovation: The Evidence Chain Behind the mRNA Narrative Reversal

Moderna's transition from "difficult years" to a positive contributor is supported by a complete body of product evidence:

• The next-generation COVID vaccine launched successfully;
• The flu + COVID combination vaccine has entered the pre-approval stage in Europe;
• The seasonal flu vaccine is advancing toward a US regulatory decision;
• The personalized cancer vaccine developed with Merck reported positive long-term data.

These are not isolated developments but a coordinated signal: the mRNA platform has evolved from an emergency tool during the pandemic into a multi-indication platform spanning vaccines, combination vaccines, and personalized oncology treatments. Market sentiment shifted from "extremely unpopular" to "significantly recovered" in just one year, precisely validating the timing window for contrarian investing. Tempus AI's genomic data platform and Insulet's automated insulin pump, meanwhile, respectively represent the deployment of data-driven diagnostics and digital therapeutics in chronic disease management.

4. The Structural Case for Active Management: The SpaceX Compounding Example

Specific investment details are given: an initial investment of approximately £150 million, held for years, has grown into a position of several billion pounds. This return did not come from market timing or trading swings, but from the structural advantages of the closed-end trust:

• Ability to hold unlisted companies (something most funds cannot do);
• No constraint to an index benchmark;
• No quarterly performance pressure;
• The board evaluates the manager on an annual or even longer cycle;
• Ability to tolerate the intermediate process of "looking foolish in the short term."

In the structure of market participants, fundamental traders now account for less than 15% of US stock market volume, with the remainder consisting of short-term, leveraged behavior. This means the disconnect between daily share price fluctuations and corporate fundamentals will continue to widen. But it is precisely this disconnect that creates extremely favorable entry points for long-cycle, concentrated investors.

5. The "Hidden Concentration" Risk of Index Investing

The sequel uses a set of data to expose the danger lurking beneath the passive investing thesis: currently, 63% of global equity indices is allocated to the US, 33% to the technology sector, and more than 35% of capital is concentrated in the top ten companies. This is not diversification in the usual sense, but an extreme momentum bet premised on "the continuation of past gains."

Dimension	Value
US weight	63%
Technology sector weight	33%
Top ten companies share	>35%

Index rebalancing operates by "buying winners and selling losers," which is efficient while trends persist, but systematically amplifies drawdowns during paradigm shifts. In contrast, the core of an active portfolio is forward-looking judgment rather than rearview-mirror risk. The report states explicitly that it would "rather bear the costs of Meituan's price war, software valuation compression, and the repricing of Chinese assets" than retreat into the safety zone of "tracking close to the benchmark."

6. AI Moves from Reasoning to Agency: Anthropic's Unprecedented Growth

The Manager Review section provides the most striking growth data: Anthropic's annualized revenue run rate surged from $1 billion in January 2025 to $30 billion in March 2026 — a 30x increase in 15 months. The text emphasizes that "no company in recorded history has grown organic revenue at this scale and speed."

This figure corresponds to the third leap in the AI paradigm:

• The Conversational Era (ChatGPT at the end of 2022): models can reliably follow instructions and interact with people;
• The Reasoning Era (o1 in September 2024): models learned to think step by step, solving math, science, and coding problems slowly but accurately;
• The Agentic Era (now): models can not only think, but also autonomously execute multi-step tasks.

For Anthropic, as a privately held company, its explosive growth shows that returns on AI capital expenditure are shifting from "expectations" to "realization," while validating that Scottish Mortgage's access to top private AI companies is an important source of excess returns. This also echoes the statement about "owning seven of the world's most valuable private companies," emphasizing that in areas unreachable by indices, closed-end trusts are a key channel for investors to access high-growth early-stage technology assets.

Part 6: The Agentic Era's Multi-Pronged Impacts, Ecosystem Divergence, and Historical Lessons

This section discusses the transition from "AI-assisted coding" to the "Agentic Era" and analyzes its deep impact on software value logic, the global competitive landscape, and physical infrastructure. The following is an expanded analysis of the sequel's content.

1. From "Assistance" to "Agency": The Qualitative Tipping Point of Value Creation

The two sets of data cited in the text reveal the critical tipping point: Google's CEO states that 75% of new code is written by AI, and Anthropic's internal AI coding rate is as high as 70-90%. These two figures mark a fundamental paradigm shift, not merely an advance in efficiency tools.

• From "Copilot" (co-pilot) to "Agent" (pilot in command): In the past, AI served as a pair-programming tool whose output value depended on human engineers' review and integration capabilities. The Agentic capabilities represented by Claude Opus 4.5 mean that models can autonomously plan, execute, verify, and deliver a complete closed-loop task. Once this tipping point is crossed, the marginal cost structure of software production is fundamentally transformed — from "human-hours" to "compute-hours." This also directly explains why the founder of a portfolio company believes the return on AI tools exceeds engineer compensation: the tool's "output/cost" ratio has already undergone a revolutionary change.

• The deeper meaning of "unstoppable": When AI's share of code production exceeds a certain threshold (e.g., 70%), the role of human engineers will inevitably shift from "producers" to "architects and reviewers." At that point, corporate competitiveness no longer depends on lines of code or engineering team size, but on the ability to define problems, decompose tasks, and oversee AI execution. This is a deep challenge to organizational form and talent structure, going beyond mere efficiency gains.

2. The Structural Rupture in Software Value: From "Seats" to "Outcomes"

The phenomenon of the software industry losing $2 trillion in market value should not be viewed merely as market sentiment volatility, but as confirmation that the old valuation model has failed. A key new argument can be added here: Agents change the decision unit of software procurement, leading to the unraveling of "seat"-based pricing power.

• From "seat licenses" to "API calls": Traditional software pricing is based on "human user count." Agents, as users, do not need "logins" or "interfaces"; they need API access, data permissions, and compute. This means the unit for measuring software value will shift from "headcount" to "task completion volume" or "compute resource consumption." This is a harsher logic than mere "intensifying competition": the revenue base of many SaaS companies is being hollowed out.

• The hollowing-out risk of the "application layer": The text's distinction between "products used by humans" and "infrastructure" is highly insightful. But it should be further noted that the essence of this distinction is the battle for the "interaction gateway." If Agents become the primary interface between users and the digital world, the value of traditional graphical user interfaces (GUIs) will be compressed. Users will no longer need to open a bloated CRM or ERP system; they will simply issue commands to an Agent. Software that exists solely as an "interface" will face commoditization, while the underlying infrastructure that controls "data" and "transactions" (such as Stripe and Adyen) will reap greater traffic dividends from high-frequency inter-Agent interactions. This is precisely the value logic behind Databricks, Snowflake, and Cloudflare mentioned in the text — they are the "utilities" of the Agentic Era.

3. The "Alternative Innovation" Paradox on the Geopolitical Dimension

The text's discussion of China's AI ecosystem (physical AI, cost-performance, productization) breaks the "Silicon Valley centrism" narrative. To deepen this point, a more dynamic perspective should be introduced: China's advantage is not catch-up innovation, but "alternative solution paths" tailored to different constraints.

• From "first principles" to the "cost-value" spectrum: US AI giants pursue the grand goal of "Artificial General Intelligence (AGI)," with a path dependence on "more compute + larger models." Chinese companies, constrained by limited compute (chip bans), have been forced down a different path: how to maximize model performance and inference efficiency within a given compute budget. This resembles the semiconductor industry's shift from "dimensional scaling" to "advanced packaging." MiniMax's open-source models demonstrate that, on specific tasks, sophisticated algorithmic architecture and training strategies can approach frontier levels while reducing costs by an order of magnitude.

• The "real-world feedback" loop of physical AI: The text's point that China's manufacturing sector is the "largest deployment surface" is key. But the deeper advantage lies in the physicalization of the "data flywheel." Silicon Valley models learn by crawling the web, whereas Chinese models (such as Horizon Robotics) acquire visual, tactile, and spatial interaction data on real roads and in factory workshops. This loop of "virtual training – real-world deployment – data recycling" is the greatest moat of the physical AI era. America's advantage lies in software and chip design, while China's advantage lies in the world's factory's "application-scenario density." As Agents move from screens into the physical world, scenario density will matter more than the scale of data centers.

4. Consumer Platforms' "Agent Defense" and the "Super-App" Potential

The text's analysis of consumer platforms (Amazon, MercadoLibre, Sea) highlights both risks and opportunities. What should be added here is that the "vertical Agents" these platforms are building are essentially a form of "defensive offense," aimed at preserving the exclusivity of the "intent gateway."

• Data as the "training ground" for Agents: The moat of the strongest platforms is not just user relationships, but the "high-value proprietary data" they possess. Nubank, for example, has users' complete cash flow data; ByteDance has users' interest graphs. This data is the core fuel for training vertical Agents, and general-purpose horizontal Agents (such as ChatGPT) struggle to access it. Therefore, the value of these platforms lies not only in the goods/services themselves, but in their role as "exclusive data suppliers for the Agentic Era." The potential threat is that if horizontal Agents cannot connect to this data, their output will be generic; vertical Agents, by contrast, can deliver precise services that "know me."

• Reinterpreting the Nubank case: David Vélez's vision, quoted in the text, of "giving every customer a private banker" will align directly in the Agentic Era. Nubank's low-cost structure enables it to afford the compute cost of running a personalized Agent for each user. This would be the ultimate form of "financial inclusion": financial services evolve from "human advisors + standardized apps" to "AI agents + real-time decision-making," and Nubank's competitiveness will evolve from "low fees" to "intelligent decision coverage."

5. The "Triple Compute Cascade" of the Physical Supply Chain

The text uses "reasoning" as the second layer of compute demand. A deeper discussion of compute expansion beyond training will follow in the next section, but the "three-layer stack" framework can first be clarified here:

Compute Stage	Core Task	Characteristics	Source of Incremental Demand
Layer 1: Training	Iterating foundation model weights	Massive batch processing, high energy consumption	Model-size race among frontier labs
Layer 2: Reasoning/Thinking	Problem decomposition, self-correction (CoT)	Requires high-performance, low-latency inference chips	Complex planning in agentic tasks
Layer 3: Action/Verification	Calling tools, code compilation, environment interaction	Highly parallel, bursty real-time compute	Autonomous agent operations in digital and physical worlds

The key insight is: the Agentic Era is not just about models being "smarter," but about applications being "greedier." The compute required for an Agent to complete a task is more than a hundred times that of a traditional query. This means the text's assertion that "the physical supply chain must be rebuilt" is not only about building larger data centers, but about reshaping a distributed computing network capable of sustaining a "high-consumption-per-task" model. It may soon be seen in the main text how the roles of NVIDIA, TSMC, and energy companies are repositioned under this logic.

6. Historical Lessons: The Railway Paradox and "the Fate of Capital Cycles"

Finally, the text's citation of railways, canals, and fiber optics is a masterstroke. To make the historical analogy more instructive, it should be "quantified" for a more precise understanding of where we stand today.

Historical Cycle	Technological Substance	Market Peak as Share of GDP/Market Cap	Outcome	Lessons for Today
Canal mania (late 18th century)	Logistics costs collapsed	Speculative bubble in UK equities	Wave of bankruptcies, but logistics networks survived	Infrastructure pioneers earned poor capital returns but high social returns
Railway mania (late 19th century)	Formation of a unified national market	~60% of total US stock market cap (1880s)	Overbuilding + price wars destroyed enormous capital	Value split between "tracks" and "transport services"
Fiber optic boom (1990s)	Communications bandwidth became unlimited	Numerous unprofitable companies went public	Telecom giants went bankrupt, but fiber networks laid the foundation of the internet	"Pipes" became cheap; content and services were winner-take-all
AI boom (2020s)	Marginal cost of intelligence approaches zero	NVIDIA/Microsoft and other giants reach peak market caps	Work in progress	Is the over-capitalization of "selling shovels" being repeated?

This comparison table clearly reveals a core contradiction: the "social value" and "capital value" of revolutionary technologies are frequently mismatched along the time dimension. Railways ultimately transformed America, but railway investors in the 1880s suffered heavy losses; fiber optics changed the world, but fiber investors in 2000 lost everything.

Extending this to the present: AI's transformative power is beyond doubt, but the $2 trillion in market value evaporation currently delivered by the capital markets is likely merely a correction of "excessive starting valuations," not a denial of "ultimate value." The real risk is not that AI lacks applications, but that the return on the industry's aggregate capital expenditure (compute buildout) may fall far below expectations, because the cost of compute may decline faster than demand grows. For investors, this means making an extremely careful distinction: are you buying generic compute that will depreciate like "tracks," or vertical intelligent applications that will keep generating cash flows like "Standard Oil"?

---

In summary, this section reveals not just an evolution of "faster and cheaper," but a process of "value gravity shift": from "those who write software" to "architects who design agents," from "software interfaces" to "data and payment infrastructure," from "U.S. computing hegemony" to "China's scenario + cost advantages," from "app store portals" to "physical-world sensor networks." For investors, this is both an era demanding vigilance against historical cyclical risks and an exceptional window for redefining "asset value anchors."

1. Key New Data on Computing Demand: The Agent Multiplier Effect

Previous analysis has discussed the three-tier drivers of AI compute growth, and this section provides an unprecedented quantitative anchor: a single agent consumes roughly 4x the compute of a single chat conversation, while multiagent systems consume roughly 15x. This data point from Anthropic appears simple on the surface, but its investment implications extend far beyond the surface:

Scenario	Relative Compute Consumption (baseline: single chat conversation)	Demand Characteristics
Chat conversation	1x	Limited by human question speed; each answer must wait for the next prompt
Single agent	4x	Autonomously loops through reasoning multiple times without waiting for human input at each round
Multiagent system	15x	Multiple agents collaborate on the same task, iterating in parallel, and can run continuously while humans sleep

The more fundamental shift is this: AI demand has, for the first time, broken free from the natural ceiling of human attention. Demand for traditional internet products is typically constrained by the scarcity of user time — a person can only ask a limited number of questions and browse a limited number of web pages in a day. The emergence of agents means that compute consumption no longer depends on step-by-step human direction but is goal-driven and self-looping. If this logic holds, what it changes is not the slope of AI growth but the shape of the AI demand curve — from "human-triggered discrete requests" to "autonomously running continuous computation."

The implication for investors: even if one is wrong about the winners in the AI application layer, as long as the "agentic era" demand paradigm holds, the growth of total underlying compute consumption carries higher certainty.

---

II. "Not Investing Is Not a Safe Option": Risk Transfer, Not Risk Elimination

This section presents a counterintuitive point—avoiding the technological revolution is not a safe stance; it merely transforms risk from one form into another. This insight deserves elaboration:

• Superficial safety: Not holding AI-related assets appears to avoid visible risks such as valuation bubbles, technological failure, and overbuilding.
• Actual exposure: Yet you are almost inevitably indirectly exposed to AI's impact through other holdings—most of the industries AI disrupts are already in traditional portfolios, while you may happen not to hold the companies that capture generational upside.

This framework redefines "whether to embrace AI" as "which kind of risk are you more willing to bear," rather than "whether to bear risk." On this basis, the investment task is further refined into a threefold screen:

1. Durable value vs. fleeting mania—distinguishing revenue and earnings quality, and distinguishing trend trading from structural demand;

2. Enabling infrastructure vs. fragile applications—infrastructure collects tolls from all winners, while applications may be crushed by the new paradigm;

3. Companies that cite AI vs. companies that convert AI into durable economic advantages—the former is narrative; the latter must be reflected in margins, customer retention, or pricing power.

An implicit corollary is that every round of technological revolution in history has seen large amounts of capital wasted—railway overbuilding, the internet bubble—yet that has not prevented a few infrastructure owners from earning decades of compounding returns. The key is not "whether AI is overvalued," but "after the bubble, which companies will emerge with even higher market share and value."

---

3. The Compounding Effect of the Three-Tier S-Curve

This section introduces a precise model: the three eras present a composite S-curve, and every curve has yet to hit its ceiling, with each one steeper than the previous. This statement deserves to be unpacked at the numerical level:

• The core metric for judging whether an S-curve is approaching its inflection point is penetration rate. If inference costs continue to fall sharply and new application scenarios keep being unlocked, the current curve is still in its climbing phase;
• The word "compounding" implies that model training demand has not disappeared; instead, inference demand is layered on top of it, and agent autonomous operation demand is then layered on top of inference demand. This is not a substitution relationship, but rather multi-layered consumption on the same physical infrastructure.

It is precisely this logic that underpins the report's sustained heavy positions in TSMC, ASML, and NVIDIA — the chip supply chain is the only physical convergence point where the three layers of demand come together.

---

IV. Key Data on Value and Return Distribution

Baillie Gifford's quantitative case for its own convictions rests on two striking figures, which also constitute the new evidence in this section:

Time Span	Data Source	Finding
Past 30 years	In-house research	Roughly 5% of stocks deliver at least a five-fold return within any five-year period
Past 90 years (U.S. equities)	Academic research	More than half of the stock market's excess returns come from just 90 companies

The common implication of these two figures is: market returns exhibit an extreme power-law distribution, and the middle ground offers almost no return at all. Holding the average (median) stock carries risk with virtually no corresponding compensation—it bears the same volatility as exceptional companies yet fails to capture their returns. This provides the mathematical basis for "concentrated holding of a small number of extremely high-quality companies," and also explains why Baillie Gifford dares to endure deep drawdowns in individual positions: in a power-law world, avoiding all losses means simultaneously forgoing all opportunities for excess returns.

---

Five. Funds' Rare Self-Examination: Analysis of Their Own Competitive Advantages

The most overlooked angle in this section is its critique of the fund industry's own methodological flaws. This critique is in fact a reverse application of the "information advantage hypothesis":

• Most funds use recent performance to prove their ability — but such data are full of luck, randomness, and mean-reversion characteristics;
• If performance is poor, they shift to emphasizing academic credentials, effort, and compensation — none of which has empirical correlation with investment returns;
• If that still fails, they dwell on risk-control processes — but the homogenization of risk control is precisely a manifestation of losing distinctiveness.

Baillie Gifford's response is to define itself through 6 pillars. The newly added and noteworthy details this time are:

1. "Own companies rather than rent shares" — hold rather than rent stocks. This wording precisely distinguishes the shareholder mindset from the trader mindset;

2. "At least ten years are needed to provide sufficient evidence of investment skill" — setting the evaluation period at a length most institutions cannot withstand is itself a structural competitive barrier;

3. Quantitative findings on financing loss aversion: academic research holds that most people are two times more averse to losses than they are fond of gains, while the fund industry may approach ten times. This asymmetric loss aversion leads institutional investors to make suboptimal decisions at the portfolio level — selling winners too early and holding losers too long.

---

6. The "Agnosticism" Advantage of Supply Chain Investment

This section's summary of the supply chain investment logic deserves to be highlighted on its own:

> Investing in the supply chain is a bet on overall AI growth, rather than a bet on specific winners.

No matter which application succeeds or which frontier model prevails, the underlying compute demand is realized through the same small group of companies. This is effectively a beta-substitution approach — but purer than buying an AI index, because index constituents are mixed with a large number of "mere AI-referencing" companies, whereas supply chain companies earn value from actual physical delivery and are unaffected by shifts in the competitive landscape at the application layer.

This agnosticism also has its boundaries, and the report remains clear-eyed about this: history has seen plenty of waste, disappointment, and overbuilding. The investment implication, therefore, is to buy infrastructure companies that can survive industry consolidation, rather than to believe every AI narrative.

---

VII. An Underappreciated Signal: The Role of MiniMax

In a sidebar, the report mentions MiniMax and its open-source model, accompanied by the phrase "world-class AI capabilities surpassing Silicon Valley." This seemingly casual placement actually conveys three points that align with the main thesis but deserve explicit articulation:

1. Open-source models dramatically reduce the cost of intelligence and expand the aggregate pool of compute demand — cheaper intelligence stimulates adoption across more use cases, which on balance increases demand for chips (the Jevons paradox reappearing in AI);

2. The geographic diffusion of frontier capabilities means localized model training and inference deployment will emerge around the world, further widening the market radius of the chip supply chain;

3. For shareholders of Scottish Mortgage Investment Trust, MiniMax itself is not an investment target, but the trend it represents — "intelligence costs trending toward zero" — reinforces the logic of holding the compute supply chain — do not bet on who becomes China's winner; instead, ensure that no matter who wins, the chips ship from companies you own.

The Active vs. Passive Industry Undercurrent: The Shift of Capital Pricing Power

The report's criticism of passive investing — "selecting stocks based on past market capitalization protects asset managers' jobs rather than shareholder wealth" — is particularly sharp in the current industry context. As of 2026, assets under management in U.S. passive funds have formally surpassed those in active funds, and the top ten constituents of the S&P 500 account for more than 35% of the index's weight. This means market marginal pricing power is shifting from fundamentally driven active investors to mechanically allocating passive capital. The consequence is that capital no longer flows toward undervalued emerging companies but instead keeps pouring into already-proven incumbents, creating a self-reinforcing loop — the more the index rises, the more inflows arrive; the higher the incumbents' valuations, the more expensive financing becomes for emerging companies. SMIT's stance of being "global in stock selection" is essentially an arbitrage of this structural distortion: by abandoning the index anchor, it actively assumes pricing power that runs counter to market consensus.

Concentrated Holdings, Probability-Weighted Returns, and Power-Law Distributions

The report emphasizes that "position sizes reflect potential upside and its probability," rather than market capitalization or geographic weights. This implies a power-law regularity that repeatedly appears in empirical evidence: over long horizons, excess returns in global equity markets are highly concentrated in a very small number of companies. AQR research shows that of the net wealth created in the U.S. stock market from 1926 to 2016, more than half came from the top 0.25% of listed companies. Traditional diversified portfolios smooth the return curve by reducing extreme upside exposure, but they also systematically exclude the tail companies that generate excess returns. SMIT's top ten positions typically represent 5%–8% of net assets, above the typical active fund ceiling of below 3%. Behind this is a continuous correction of "active weight misallocation" — companies assigned high weights based on market capitalization often have entered a phase of weakening growth, while companies with truly explosive growth potential are systematically ignored by index portfolios because their scale is too small.

"Growth at Unreasonable Prices": The Transfer of Pricing Power to the Optimists

The "Growth at Unreasonable Prices" proposed by the report is a complete rebellion against traditional GARP (Growth at a Reasonable Price). From a behavioral finance perspective, the market's pricing disagreement over high-valuation growth stocks is essentially a disagreement over the rate of time preference — how much are you willing to discount distant future free cash flows? This difference is amplified by the interest-rate environment. During the sharp global rate increases of 2022–2024, long-duration growth stocks experienced substantial valuation compression, and GARP strategies were cyclically dominant. But SMIT insists in its annual report that "it is harder than ever to find sustained growth." Its stance does not ignore valuation risk; rather, it rests on a core judgment: in the digital economy era, the payback period for intangible asset investment has been greatly extended, and the 5–7 year forecast window of traditional DCF models systematically underestimates the network effects and data flywheels of platform companies.

Dimension	GARP	GUP
Valuation tolerance	PEG < 1.5	No upper limit
Failure tolerance	Safety margin reserved	Accepts partial losses in the overall portfolio
Core logic	Mean reversion	Power-law right tail
Typical loss rate	Low (<10%)	High (>30%)
Success compensation	~20–40%/year	5–10x or more

Structural Deficiencies in Governance Engagement: Two Failures — Passivization and Activism

The report notes that "the investment management industry has ceded the corporate governance role to the vested interests of activist investors" — a precise and candid description of the current governance ecology. Under the pressure of "passivized" operations, traditional active funds have drastically cut their internal governance engagement teams. They are reluctant to publicly challenge management for fear of damaging their investment banking business, and they are also unwilling to bear the long-term monitoring costs of proxy voting. Activist investors, by contrast, engage in governance with an explicit short-term share-price catalyst agenda, seeking rapid returns through capital measures such as spin-offs, buybacks, and divestitures — which fundamentally conflicts with a company's long-term strategic building. SMIT advocates a third path — not driven by short-term events, but based on deep engagement built on "decade-long partnerships." Its discussion of TSMC's capital expenditure starts from a "confidence signal that demand persists into the 2030s," rather than quarterly margin guidance; its dialogue with Shopify focuses on the long-term positioning of "commerce infrastructure" rather than short-term GMV fluctuations. The time scale of this governance logic cannot be matched by any other governance participant.

The Cost Curve: Fee Erosion of Long-Term Compound Returns Is Nonlinear

The report reframes the impact of fees by using "fees as a percentage of expected returns" — an insight the industry has underestimated. Take a 10% annualized return expectation as an example:

Fee level	Annual fee	Share of return	30-year cumulative drag vs. equal contributions
SMIT	0.33%	3.3%	Baseline
Typical active fund	0.75%	7.5%	~12% more terminal value lost
High-fee active fund	1.5%	15%	~25% more terminal value lost
Private equity industry average	2% + 20% performance fee	40%+	More than 50% of terminal value lost

Most investors interpret this as "the fee difference is only 1.17%," but in reality a 1.17 percentage point difference means losing more than one-sixth of potential terminal value over the long term. SMIT's insistence on not charging a performance fee is likewise consistent with long-term investment logic — performance fees amplify a fund manager's path dependence and risk-aversion motives, leading to decisions that contradict long-term judgment during periods of greatest volatility.

The Shrinking Public Market and the Structural Shift of "Best Companies in Private Markets"

The companies described in the report — "remaining private longer and resistant to listing" — reflect a global long-term trend. In 2025, global IPO proceeds fell below the ten-year average for the third consecutive year, and the number of U.S. IPOs was only a quarter of the level a decade earlier. Meanwhile, PE-backed private companies are tilting toward founder-friendly capital at a record pace, with founders retaining greater control over boards and having more choice over shareholder structure. This structural shift poses a dual challenge for SMIT — it must not only identify high-quality growth companies in the private market, but also build trust-based relationships with them grounded in long-term commitment. SMIT's "permanent capital" characteristic plays a critical role here: as a listed investment trust with no redemption pressure, it can promise private companies a holding period that spans economic cycles. This is a promise that open-ended mutual funds are structurally unable to deliver.

The TSMC Micro-Signal: Capital Expenditure Guidance as a Scarce Proxy for Long-Term Demand Confidence

TSMC raised its capital expenditure by 30% and gave "significantly higher" guidance for the next three years. In the semiconductor industry, capex is an early reflection of future demand, and physical capacity carries a two-to-three-year transmission lag. This means the capacity approved for construction today will not come online until 2028–2029. TSMC is effectively telling the market that it sees robust demand extending at least into 2030. More critical is the demand assessment methodology emphasized by TSMC CFO Wen Wenkai — a conservative assumption that "customers may be overbooking capacity." This is a direct counter-response to current concerns about an AI infrastructure bubble: the company is not basing its demand forecast on customers' current order volume, but is incorporating the "bullwhip effect" in the supply chain. Industry analysts expect global AI-related semiconductor capex to total more than $300 billion in 2026–2027. As a key upstream supplier in that chain, TSMC's capex decision is in effect an endorsement of the entire AI value chain.

The Shopify Case: The "Infrastructure Moat" at the AI Inflection Point

Tobi Lütke positions Shopify as "commerce infrastructure" rather than a software product, comparing online stores to "Excel" and noting that the value lies in the underlying infrastructure layer. This analogy reveals an important strategic shift for platform companies in the AI era — AI pushes the cost of front-end storefront construction toward zero, but underlying complexities such as payments, compliance, fulfillment, and fraud detection instead become a deeper moat. Notably, Tobi acknowledges that "a moat built solely on the difficulty of writing code is fragile" — a candor rare among tech company CEOs. In addition, Shopify routes most of its AI queries to Anthropic and has managed to support a larger business with roughly half the headcount of three to four years earlier — suggesting that AI is effectively compressing tech companies' headcount expansion leverage. However, the "insufficient persuasiveness when pressed" in the dialogue also exposes a genuine unresolved question: when AI drives the marginal cost of customized software solutions toward zero, the pricing power and customer stickiness of any SaaS company will face challenges — a threat for which there is currently no clear framework to address.

The continuation of Part 9 presents three interrelated dimensions: meeting validation of specific companies (TSMC and Shopify), the workings of governance mechanisms (proxy voting), and a set of performance data across time horizons. These seemingly independent sections actually jointly answer one question: after the deep drawdown of 2021–2023, can Scottish Mortgage's long-termism logic still be trusted?

1. The "Certainty Gradient" in Meeting Conclusions: The Verification Gap Between TSMC and Shopify

The two meeting notes disclose the conclusions of post-investment tracking in completely different tones, and are worth scrutinizing word by word:

• TSMC: `reinforced confidence`, `durability of demand`, `long-term compounding case` — the voice is definitive, the judgment is reinforcing.
• Shopify: `well positioned`, `will need to continue demonstrating` — the voice is conditional, the judgment remains to be tested.

Dimension	TSMC	Shopify
Core judgment	Durability of advanced-node demand reinforced	Well positioned in the agentic commerce world, but needs to keep proving it
Competitive moat focus	Technology leadership, customer relationships, manufacturing execution — a triple stack	Payments, settlement, compliance, merchant infrastructure — ecosystem network type
Management assessment	Conservative demand forecasting viewed as a positive signal	Needs to prove AI strengthens rather than weakens its competitive position
Strength of conclusion	Long-term compounding thesis confirmed	Directionally correct, but verification conditions not yet closed

Behind this tonal difference lies a difference in the nature of the moat. TSMC's competitive advantage rests on physical, manufacturing-driven deep execution, with disruption cycles measured in decades. Shopify, by contrast, operates in the software-commerce layer, where AI could either amplify its merchant network effects or reconfigure transaction entry points. The investment team therefore maintains a longer observation period and a higher falsification bar. This is not a lack of confidence in Shopify, but a clear-eyed distinction between a "structural beneficiary" and a "conditional beneficiary."

One noteworthy detail: the `conservative approach to demand forecasting` mentioned in the TSMC meeting is cited positively. In Scottish Mortgage's evaluation framework, management restraint during an upcycle is itself an important factor in long-term compounding — a sharp contrast to the market's general preference for optimistic guidance from management.

2. Proxy Voting: A Data Portrait of Decentralized "Active Ownership"

The voting data disclosed this period seems simple at first glance, but it embeds a governance philosophy that is difficult to replicate:

Voting distribution	Count	Share
Supporting management	732	96.8%
Against / abstain / withhold	24	3.2%
Meetings not voted	0	—

A 3.2% opposition rate is low by industry standards, but this cannot simply be read as a "rubber stamp." The annual report explicitly describes an action sequence: direct engagement first, with voting as an escalation mechanism. In other words, the 24 opposing resolutions represent "the final expression after dialogue has failed" — the scarcity of opposition actually amplifies the signal strength of each individual opposition vote.

Even more noteworthy is this passage:

> In keeping with our decentralised and autonomous culture, our investment teams will, on occasion, elect to vote differently on the same general meeting resolutions.

Allowing different investment teams to vote differently on the same resolution is tantamount to publicly acknowledging internal disagreement. Most asset managers deliberately maintain a "uniform voting policy" to avoid external misinterpretation. Baillie Gifford has chosen to delegate voting authority to independent investment teams, and the logic behind this is consistent with stock selection: since investment judgments are bottom-up, voting judgments should be too. This also means that the 96.8% support rate is not an institution-level "endorsement," but rather the aggregate result of each independent team's separate assessment.

3. The Dialectic of Time in the Data: From One-Year Excess Returns to Ten-Year Compounding

1. The One-Year Dimension: The Quality of Excess Returns

Metric	2026	2025	YoY change
NAV total return (fair value)	27.4%	11.2%	+16.2pp
Share price total return	26.8%	6.0%	+20.8pp
FTSE All-World (in GBP)	18.0%	5.5%	+12.5pp
Discount (fair value)	(9.5%)	(9.0%)	Widened 0.5pp

In 2026, NAV delivered an excess return of approximately 9.4 percentage points relative to the global index; capital return of 273.03p represented more than 99% of the total return of 275.31p — a typical capital-growth-dominated year. But note that the discount did not narrow in tandem: while the industry average discount narrowed from −9.6% to −4.7%, Scottish Mortgage's discount actually widened from −9.0% to −9.5%. The relative discount deteriorated by approximately 5.4 percentage points in one year. This implies that although the market recognizes the rebound on the asset side, its patience for the high-valuation growth style itself is still declining.

2. The Five-Year Dimension: The Lost Five Years and the Covert Repair of Per-Share Value

Measured by nominal NAV, the annualized return over the past five years (2021–2026) was only about 1.9% — essentially the "lost five years." But if one estimates from shareholder funds and per-share NAV, the number of ordinary shares fell by roughly a quarter over this period (from approximately 1.43 billion to approximately 1.08 billion). In other words, management continuously bought back shares during the prolonged discount period, allowing per-share NAV to recover by significantly more than portfolio assets:

Period	Per-share NAV (fair) change	Cumulative multiple	Annualized return
2016–2021	259.2p → 1,195.1p	4.61x	≈35.8%
2021–2023	1,195.1p → 816.8p	0.68x	≈−17.3%
2023–2026	816.8p → 1,315.8p	1.61x	≈17.2%
2016–2026	259.2p → 1,315.8p	5.08x	≈17.6%

The core fact revealed by this data is: Scottish Mortgage's long-term returns are extremely uneven, but extreme volatility itself is the price of admission for long-term compounding. Without the 2016–2021 expansion period annualizing at 35.8%, there would be no ten-year annualized record of 17.6%; without the deep drawdown of 2021–2023, there would be no 17.2% recovery resilience in 2023–2026. For holders, the real test is not whether one believes in long-termism, but whether one can withstand a −31.7% drawdown lasting two years without leaving.

3. The Ten-Year Dimension: Fee and Leverage Discipline

Year	Ongoing charges	Gearing (fair)	Dividend per share
2016	0.45%	13%	2.96p
2021	0.34%	6%	3.42p
2023	0.34%	14%	4.10p
2025	0.31%	13%	4.38p
2026	0.33%	11%	4.57p

Over the decade, the fee rate fell from 0.45% to 0.33%, with the benefits of economies of scale consistently passed through to shareholders. More noteworthy is the countercyclical use of leverage: gearing was compressed to 6% at the 2021 bull-market top, raised to 14% at the 2023 market bottom, and reduced to 11% again in 2026. This rhythm of "deleveraging when expensive, adding leverage when cheap" is an important amplifier behind the ten-year 17.6% annualized return.

On dividends, the 4.57p in 2026 represents cumulative growth of approximately 54% from the 2.96p in 2016, or about 4.4% annualized. But note that revenue income was only 2.28p, implying a dividend coverage ratio of about 50%, with the shortfall supplemented by capital reserves. This serves as a reminder: Scottish Mortgage's dividend has never been a core source of return; it is an incidental by-product of capital growth.

4. Conclusion: What Are These Pages Saying?

Putting the meeting notes and voting data together, a complete picture of "active ownership" emerges:

• Stock selection: gives definitive endorsement to a structural winner (TSMC), while preserving verification space for a conditional winner (Shopify);
• Governance: exercises voting rights at the level of independent teams, using a "dialogue first, vote second" sequence to reduce unnecessary confrontation;
• Performance: over one year, look at excess returns (+9.4pp); over five years, look at repair (buybacks + per-share NAV recovery); over ten years, look at compounding (17.6% annualized).

The sole warning signal — a 5.4 percentage point widening in the relative discount — is a reminder that even when fundamentals deliver, the market's pricing of the long-duration growth style may remain unfriendly. Scottish Mortgage's long-termism is always built on a sustained game between "value on the asset side" and "indifference on the market side."

Based on the newly added annual report data (FY2026), the following analysis focuses on the strategic implications behind quantitative performance, portfolio structure, trading behavior, and financial metrics, serving as an extension of and supplement to the analysis in the previous nine parts.

---

一、业绩的“双面镜”：五年平庸与十年卓越的背离

累计业绩表揭示了最具启发性的矛盾：十年复合回报17.6%（NAV），大幅跑赢FTSE All-World的10.3%；但五年复合回报仅2.3%，远逊于指数的8.7%。 这意味着该基金近五年经历了严重的相对回撤。

指标（年化）	5年	10年
NAV总回报	2.3%	17.6%
FTSE All-World总回报	8.7%	10.3%
超额收益（NAV vs 指数）	-6.4%	+7.3%

这组数据隐含了一个核心矛盾：2021-2026年间，全球指数由大型科技股推动，而SMIT的集中持仓中，大量头寸在2021年高点后经历了50-60%的回撤（如Meituan、Sea），至今未能完全修复。 指数在2022年后迅速创新高，而SMIT的NAV用了四年才从454（2021）修复至508（2026）。这说明其持仓的“含AI量”在2021-2023年不足，直到近期SpaceX、TSMC、ASML等才真正发力。长期业绩的亮眼依赖于2016-2021年的黄金五年，而那段时间的成长股普涨环境很难再现。因此，未来十年要复制17.6%的年化回报，难度远大于过去十年。

---

II. SpaceX's "Single-Engine Core" and Portfolio Concentration Risk

The contribution table shows that SpaceX alone contributed 14.9% of absolute performance, representing 56% of total contribution (based on the year's NAV total return of 27.4%, 14.9%/27.4% ≈ 54%). Its market value reached £2.984 billion, accounting for 19.3% of total assets.

Holding	Contribution %	Absolute Gain %	% of Total Assets
SpaceX	+14.9	+178.6	19.3
TSMC	+3.6	+99.1	5.7
ASML	+2.3	+94.2	3.2
NVIDIA	+1.6	+57.5	3.2
Total	+22.4	—	31.4

A single holding of 19.3% is extremely rare in any closed-end investment trust. This high concentration implies:

• If SpaceX's valuation is re-rated to fair value after listing, or if private market valuation compression similar to 2021–2023 recurs, the portfolio's NAV will face significant volatility. A simple calculation: if SpaceX's valuation pulls back 30%, it would drag NAV by approximately 5.8% (19.3% × 30%).
• This holding contributed the main increment that lifted the "Industrials" weight from 16.9% to 29.2%, masking reductions in other industrial stocks. The portfolio's sector label is thus "hijacked" by a single company, and investors need to look through to see true diversification.

---

III. £1.4 Billion in Active Outflows: Profit-Taking or Strategic Contraction?

The investment movement table shows total assets increased by £1.543 billion, of which valuation gains contributed £3.152 billion, but active net selling/outflows reached £1.408 billion. This indicates that the investment manager made large-scale active position reductions in 2025-2026.

Region	Net purchases/(sales) £m	Valuation gains/(losses) £m	Change in closing value £m
United States	(1,025)	+3,199	+2,174
Developed Europe (ex-Eurozone)	(240)	(529)	(768)
Eurozone	(151)	+239	+88
China	+68	+15	+83
Other Asia	+58	+275	+333

The net reduction of over £1 billion in the United States, alongside valuation gains of nearly £3.2 billion, is a classic case of "selling into strength." This reveals two signals:

1. Valuation discipline is at work. Based on the trading list, the substantially reduced positions in ASML (up 94%), TSMC (up 99%), and NVIDIA (up 57%) are all high-valuation AI hardware names. The investment manager is converting unrealized gains into new investments.

2. The sharp decline in Europe (ex-Eurozone) is particularly concerning. Net selling was only £240 million, but valuation losses reached £529 million. This means that in addition to active selling, holdings in this region experienced severe price declines in 2025/26. Combined with Spotify falling 13.7% for the year, Kinnevik being fully liquidated, and Wayfair being liquidated, it can be inferred that: the trust's stock-selection framework in the European innovative technology sector (fintech, media, consumer internet) suffered a systematic failure over the past 12 months.

---

4. The Evolution of Private Market Investing: From Direct Investment to "LP-Style Allocation"

The biggest difference between this year's new-buy list and those of previous years is the appearance of three venture capital funds:

• ARCH Ventures Fund XI/XII/XIII (all three held simultaneously)
• Antler East Africa Fund I LP
• Sinovation Fund III

This move is worth interpreting from two dimensions:

1. The need to diversify risk. The lesson of an 89% loss from directly investing in a single private company (such as Brandtech) shows that even after rigorous due diligence, the tail risk of private companies remains enormous. By spreading across multiple projects through LP interests, even if individual projects fail, the portfolio effect can absorb the losses.

2. A strategy for gaining new deal flow. East Africa (Antler), China (Sinovation), and Deep Tech (ARCH) respectively fill the gaps in SMIT's capabilities in geographic coverage and frontier technologies (quantum computing, nuclear fusion, longevity medicine, etc.). Rather than building a massive internal due-diligence team, it is better to "borrow strength to strike." This also explains the simultaneous heavy buildup of positions in individual projects such as Anthropic, PsiQuantum, and Zipline — the investment manager is walking on two legs: "direct investment + indirect allocation."

---

5. 'Cyclical Peak' Signals for Digital Consumer Platforms

Among the top ten detractors, three Chinese/Southeast Asian consumer internet companies appear:

Detractor	Absolute Gain %	Contribution %
Meituan	-48.3	-1.8
Sea	-37.9	-1.1
Pinduoduo Inc	-15.4	-0.8

These three positions are all among the top thirty holdings (Pinduoduo at #10, Sea at #19), and are remnants of the portfolio's heavy allocation to Chinese internet stocks at the 2021 peak. They underwent a prolonged decline from 2021 to 2023, and have still not bottomed out in 2025/26, indicating that market concerns over their business models (subsidy wars in local-life services, setbacks in e-commerce overseas expansion, and intensifying competition in Southeast Asia) have not yet been resolved.

Meanwhile, Brandtech's decline of 89% shows that not only in the secondary market, but even in the primary market, "digital marketing SaaS" has suffered a devastating impact from the AI transformation. The most direct lesson for the portfolio is: any internet asset that relies on "traffic distribution" rather than "technology lock-in"—whether listed or not—has become a high-risk area within the portfolio. In contrast, AI infrastructure names (SpaceX, TSMC, ASML, Anthropic) have become major contributors, indicating that the portfolio's "new economy" core has shifted from "model innovation" to "hard-tech innovation."

---

6. The "Correlation Trap" of the Three Semiconductor Giants

TSMC, ASML, and NVIDIA together account for 8.5% of the 31.4% (Total Asset), and all three rallied sharply in 2025/26. But they sit on the same supply chain: NVIDIA designs the chips, TSMC handles the foundry work, and ASML supplies the lithography equipment. All three are highly correlated with a single macro variable — AI data center capex.

Semiconductor Holdings	% of Assets	Absolute Gain %
TSMC	5.7	+99.1
ASML	3.2	+94.2
NVIDIA	3.2	+57.5

The investment manager significantly reduced the ASML position in fiscal 2026, signaling wariness about valuations, while the cuts to TSMC and NVIDIA were relatively modest. If AI capex experiences a phase of slowdown in 2026-2027, all three could pull back 20-30% in tandem, which would directly erode 2-3 percentage points of NAV. Over the long term, however, semiconductors are the "picks-and-shovels" of the AI era, and their certainty is superior to that of the downstream application layer.

---

VII. Expansion of the Dividend Coverage Multiple: A Growth-First Redistribution Signal

The cumulative performance table shows that Revenue earnings per share rose from 100 in 2016 to 267 in 2026, while Dividend paid per share rose from 100 to 154. The dividend coverage multiple expanded from 1.0x to 1.73x.

Year	Earnings Per Share (Index)	Dividend Per Share (Index)	Coverage Multiple
2016	100	100	1.0x
2020	218	110	1.98x
2026	267	154	1.73x

This reflects management deliberately holding down the dividend growth rate, retaining more earnings for private market and long-term compounding investments. For income-seeking shareholders, this is an implicit "growth option"; but it also means that if the fund lacks high-return projects in the future, the headroom for dividend increases is limited. Combined with its long-term total return strategy, this is a distribution policy consistent with its investment style, but not necessarily suitable for all investors.

---

VIII. Structural Drift in Regional Allocation: From "Global" to "Quasi-US-Equity Fund"

In the geographic distribution table, North America's share rose from 54.5% to 59.7%, Europe fell from 17.3% to 13.7%, and China remained at 13.2%. This is already the third consecutive year that North America's share has increased.

Region	2026	2025	Change
North America	59.7%	54.5%	+5.2%
Europe	13.7%	17.3%	-3.6%
Asia	13.2%	14.5%	-1.3%
South America	5.3%	6.5%	-1.2%

This drift is not entirely the result of active choices; rather, the valuation expansion of U.S. equities (especially technology stocks) has been pushing up passive weights. SpaceX alone contributed 3.6 percentage points to North America's share. Excluding SpaceX's impact, active rebalancing may be closer to a direction of "selling U.S. equities and buying China/Asia" (China and Rest of Asia each saw net buying). For investors, the "global diversification" function of the current portfolio has significantly weakened; in essence, it is a concentrated growth fund focused on U.S. technology and private companies. When allocating to such trusts, investors need to consider whether their own global asset distribution is already overly tilted toward U.S. equities.

---

Summary Comparison

Strategic Dimension	FY2026 Characteristics	Investment Implications
Performance Engine	SpaceX alone contributes over 50%	High elasticity and high risk coexist
Position Adjustment Direction	Sold high-valuation semiconductors, reduced Europe, increased China/private	Suggests the market is shifting from AI hardware to AI applications/early-stage
Privatization Strategy	Shifted from direct investment to "direct investment + fund LP"	Reduces the risk of individual project blowups
Biggest Risk	50% of assets concentrated in private tech + industrial targets	Requires very high risk tolerance
Distribution Trend	Dividend coverage ratio rose to 1.73x	Room to increase dividends or buybacks in the future

---

In summary, this annual report shows that Scottish Mortgage Investment Trust delivered strong absolute returns in FY2026, but those returns were highly dependent on SpaceX's epic rally and the simultaneous mark-up in private market valuations. Beneath the surface-level high-growth narrative, the portfolio is undergoing a profound internal reshuffle: selling highly valued semiconductors in the secondary market, buying early-stage technology in the primary market, while bearing the cost of continued declines in consumer internet assets. The success or failure of this strategy will depend on whether, between 2027 and 2030, AI infrastructure can transition from a capex cycle to a profit-delivery cycle, and whether SpaceX can defend its trillion-dollar valuation anchor after listing. For long-term investors, this is a portfolio worth holding but requiring continuous scrutiny of concentration risk; for investors seeking global diversification or stable cash flows, the current SMIT is no longer the optimal solution.

This section focuses on capital deployment, valuation changes, and holding structure in the private investment portfolio, revealing the following key features and potential implications:

1. Capital Deployment Accelerated Markedly, but Active Exits Remained Minimal

In FY2026, new investments totaled £254m, up 92% from £132m in the prior year. At the same time, proceeds from sales were only £8.7m, almost negligible. This indicates the fund remains in an accumulation phase of "only inflows, no outflows," and the investment cycle has not yet fully opened.

Item	2026 (£m)	2025 (£m)	Change
New investments	253.8	132.0	+92.4%
Proceeds from sales	8.7	12.8	-32.0%
Realized gains/losses	0.8	-5.9	Turned profitable
Transferred to listed	-39.2	-248.3	Sharply reduced
Change in fair value	+2,426.5	+171.5	+1315%

Changes in fair value contributed the bulk of the year's asset growth; combined with new investments, the private portfolio's total value jumped from £3.78bn to £6.42bn, up 69.5% year-on-year. The "transferred to listed" line narrowed from -£248m to -£39m, indicating that only HeartFlow moved from private to public in 2026, far fewer than the three in 2025 (Bolt Projects, Tempus AI, Horizon Robotics). This shows that the pace of private exits is not linear and that IPO windows carry uncertainty.

2. Concentration Risk and Late-Stage Growth Preference Are Two Sides of the Same Coin

Fifty-three private companies account for 41.5% of total assets, with the top five representing 75.7% of private exposure and the top ten 86.0%. SpaceX alone is 19.3% of total assets — 7.2 percentage points more than the combined weight of ByteDance, Stripe, Zipline, and Databricks (4.7% + 4.0% + 2.0% + 1.4% = 12.1%).

The size distribution data more clearly reveals the strategic orientation:

Size range	Portfolio weight	Number of holdings
Micro (< $300m)	0.6%	11
Small ($300m - $2bn)	4.0%	15
Medium ($2bn - $10bn)	3.9%	6
Large (> $10bn)	33.0%	9

The nine large, late-stage companies contribute more than a quarter of the portfolio's assets, while the 26 micro and small companies together account for only 4.6%. This structure of "small positions in small companies, heavy positions in large ones" means the portfolio's short-term performance is highly dependent on a handful of giants that have already proven their business models.

3. Valuation Adjustments Were Highly Active, but Upside Was Concentrated at the Top

During the year, 525 revaluations were carried out across 89 valuation instruments, an average of nearly six revaluations per instrument. Looking at the distribution of valuation multiples, 65.2% of instruments are valued at more than 5x cost, while 34.8% are within 4x. This indicates that most assets have appreciated significantly, but roughly a third remain in early or stressed stages.

By tier, the average valuation change for the top ten private companies was +86%, while the average across all portfolio companies was only +8%. The huge gap again confirms the role of concentration — it was the valuation leaps of leading companies (such as SpaceX and Stripe) that drove overall performance, not a broad-based rally. This divergence is amplified in the annual change in fair value: the +£2.43bn gain came almost entirely from a few companies.

4. Core Holdings Driven by Both "Business Milestones" and "Capital Moves"

The report's updates on the three leading companies reveal an obvious dual-driver pattern:

• SpaceX: After acquiring xAI in February 2026, its combined valuation reached $1.25tn, up 56% from $800bn at the end of 2025. On the operational side, 165 orbital launches (over half of the global total), Starlink subscribers surpassing 10 million, and the FAA raising the annual launch cap at Starbase from 5 to 25 are all substantive progress supporting the valuation.
• ByteDance: A resolution was reached for TikTok's US business in December 2025, significantly reducing geopolitical uncertainty. Doubao has over 100 million daily active users and is the product with the lowest marketing spend in ByteDance's history, demonstrating the natural penetration power of an AI-native product. This, however, has not directly shown up in the valuation multiple (the holding is 4.7% of assets); perhaps the market still applies a discount for regulatory risk.
• Stripe: With $1.9tn in annual processing volume, it handles roughly 1.6% of global GDP, and in February 2026 it conducted a tender offer at a $159bn valuation (up 74% from $91.5bn twelve months earlier). Notably, Stripe serves 90% of S&P 500 companies and 80% of Nasdaq 100 companies. This infrastructure-level penetration gives it the resilience to persist through cycles.

5. Divergent Performance of Previously Private Holdings That Have Since Listed; Future Exit Paths Still Need Validation

The report uses charts to show cumulative absolute returns from initial investment through IPO and from IPO to 31 March 2026. Although the specific figures are not all listed, the chart trends show that some companies (such as Affirm, Aurora Innovation, and Joby Aviation) performed poorly after IPO, while others (such as Spotify and Wise) continued to deliver positive returns. This suggests that appreciation during the private phase does not equal the returns shareholders can ultimately realize. An IPO is merely a liquidity event; subsequent secondary-market performance is the real test. As more private companies (SpaceX has already filed for an IPO) enter the listing window, Scottish Mortgage's exit quality will be directly tested by the market.

Taken together, the data in this section further reinforces the strategic profile of Scottish Mortgage: relying on a few high-growth giants and using patient capital to accompany extremely late-stage companies all the way to IPO. The high concentration of its private portfolio is both a source of excess returns and the largest potential risk to future drawdowns — should SpaceX or ByteDance experience a reversal in business or valuation, the overall net asset value would face a disproportionate impact.

1. Stablecoins: From "Payment Experiment" to "Institutional Asset" — Three Lines of Evidence

Stripe's $1.1 billion acquisition of Bridge is often seen as a "faith investment" in the stablecoin track. But the 2025 data shows this was not faith; it was positioning ahead of an inflection point in penetration:

Metric	2024	2025	Growth
Global stablecoin payment volume	~$200 billion	$400 billion	+100%
Bridge transaction volume	—	4x the global market growth	approx. +300-400%
BTC price peak	—	$126,000	All-time high
Peak total crypto market cap	—	$4 trillion	Briefly touched
AUM of stablecoin ETFs/investment products	—	$150 billion+	Outstanding at year-end

What is truly notable is not the growth itself, but the fact that the signing of the GENIUS Act moved stablecoins from a "regulatory gray zone" into the legal framework of "payment infrastructure". This provides players such as Stripe and Blockchain.com with two options: first, use licenses to build compliance barriers (Blockchain.com obtained licenses in 30 EEA countries plus FCA registration in the UK); second, use the compliance framework to attract institutional capital (ETF products with $150 billion in AUM).

Comparing the strategic differences between Stripe and Blockchain.com: Stripe chose to acquire infrastructure (Bridge) and embed directly into merchant acquiring processes; Blockchain.com chose to build a licensed exchange plus custody. The former captures "payment pipe" economics, the latter "asset custody." But both benefit from the same macro trend — stablecoins are shifting from derivative demand for bitcoin to an independent payment and settlement layer.

2. The Lesson of "Valuation Inversion": Unlisted ≠ Valueless, and Risk Premium ≠ Fundamental Discount

The ByteDance case has the most analytical tension:

Dimension	ByteDance	Meta
Quarterly revenue (2025 Q2)	~$48 billion	Lower than ByteDance
Annualized revenue base	Comparable	Comparable
Valuation magnitude	A fraction of Meta's	Trillion-dollar level
Source of discount	Geopolitical + regulatory risk premium	—

If priced purely on fundamentals, the valuation gap should not be so wide. The market's explanation can only be that the secondary market pricing mechanism cannot adequately handle the dual attribute of "Chinese asset + global business". Either ByteDance gets a Chinese-company valuation multiple or a global social-platform multiple; there is no middle ground between the two.

The lesson for long-term investors is that valuation inversion does not mean deteriorating fundamentals; on the contrary, it may be a source of excess returns from structural discount. Similar logic applies to Zipline — it is not just a drone delivery company; it also owns the world's largest autonomous delivery safety dataset (130 million miles, 2 million deliveries, zero serious injuries), yet the market still values it as a "drone concept stock" rather than an "infrastructure data company."

3. The "Revenue Thickness" of Enterprise AI: The Leap from Customer Count to Contract Value

Putting Anthropic and Databricks' data side by side makes it clearer that AI commercialization is shifting from "user-count driven" to "contract-value driven":

Metric	Anthropic	Databricks
Annualized revenue	~$30 billion (Apr 2026)	~$5.4 billion run-rate
Revenue growth	From $1 billion to $30 billion (~18 months)	+65% YoY
Customers spending >$1M annually	1,000+ (Apr 2026)	800+
Customers spending >$10M annually	Not disclosed, but 8 of the Fortune 10 are customers	70+
AI product revenue contribution	Claude Code annualized $2.5B+	$1B+
Cloud channel coverage	AWS + Azure + GCP (only full-platform)	AWS + GCP + Azure

The key change lies in the customer stratification structure:

• Among Anthropic's 1,000+ "million-dollar" customers, 8 of the Fortune 10 are included — indicating extremely deep embedding among top customers; Claude has become a core workflow (coding + knowledge work) rather than an experimental tool.
• Among Databricks' 800+ million-dollar customers, more than 70 have reached the ten-million-dollar level — indicating that the budget expansion path for data platform customers is "buy the platform first, then buy AI." Once data assets are stored on Databricks, migration costs are extremely high.
• Both have experienced "post-funding acceleration": after Anthropic completed a $30 billion funding round in February 2026, its million-dollar customers doubled in less than two months; after Databricks raised at a $134 billion valuation, it brought in a strategic investment from Microsoft.

These data points point to the same conclusion: the valuation of an AI company is no longer determined solely by model capability, but by the "distribution curve of customer annual contract value." The steeper the curve, the deeper the product is embedded into core enterprise workflows, and the stronger the sustainability of revenue.

4. Zipline: Data Accumulation Itself Is a Compounding Moat

Zipline's most underappreciated asset is not the aircraft, but the data:

• 130 million miles of autonomous flight = operational data from drones in real airspace, real weather, and real terrain;
• 2 million commercial deliveries + zero serious injuries = an actuarial-grade safety record;
• ~15% week-over-week growth in the US market = if sustained, theoretical annualized growth exceeds three orders of magnitude (though in practice it will moderate, it shows the company is still on the accelerating segment of the S-curve).

A notable new structural signal is the US State Department's pay-for-performance contract (up to $150 million, covering five African countries, aiming to serve 130 million people).

Dimension	Traditional government contract	Zipline contract
Payment basis	Cost/time-based	Delivery performance-based
Service scope	Single country	Multi-country coordination
Risk bearing	Government	Company
Data ownership	Government	Zipline

This contract structure shows that the government recognizes not just Zipline's drone capabilities, but its operating system of "proving results with data." If the potential service scope of 130 million people materializes, Zipline's safety dataset will pull further ahead of all competitors.

5. Ant International vs Revolut: Comparing Two Cross-Border Payment Paths

Both companies achieved high growth over the same period, but their paths are completely different:

Dimension	Revolut	Ant International
Model	Proprietary bank accounts + license expansion	Aggregated payment network + connecting local wallets
Users/Scale	70 million retail customers	Alipay+ connects 1.8 billion accounts and 40 payment partners
Valuation/Revenue	$75 billion valuation, £4.5 billion revenue	Not separately valued; 200+ billion cross-border transactions
Regulatory strategy	Obtaining UK banking license + applying for US license	Local partnerships + infrastructure export
Enterprise business	Not yet fully lending	WorldFirst transaction volume +40%, Antom overseas transaction volume +75%

Revolut's moat is the "compliance leverage of banking licenses" — one license gives access to 30 countries, 70M users, and a one-third share of new account openings. Ant International's moat is its "position of connectivity" — 1.8 billion user accounts mean that as long as a merchant connects to Alipay+, it can reach the world's largest mobile payment user pool.

The two have different definitions of "cross-border": Revolut is "cross-border banking services for individuals when traveling"; Ant International is "the default cross-border transaction layer between merchants and consumers." The former earns channel fees, the latter earns an ecosystem tax.

6. TPB: Rewriting the Agricultural Cost Curve with "Platform-Based Breeding"

The Production Board (TPB) is often misunderstood as a "food tech venture fund," but its real model is closer to a "biotechnology platform company":

• Supergut: natural GLP-1 food vs. the drug Ozempic — this is not a "poor man's GLP-1," but rather the use of food channels to reach the vast consumer market (pre-obese populations) that drugs cannot access, with lower pricing, no prescription barrier, and suitability for daily consumption.
• Ohalo: self-fertile almonds (a world first) + Boosted Breeding expanded to strawberries — if a single breeding technology breakthrough can be replicated across multiple crops, it effectively achieves a "platform + plug-in" model in breeding.
• PCAST appointment: Friedberg has joined the President's Council of Advisors on Science and Technology, meaning TPB can influence US agricultural biotechnology policy at an early stage of regulatory formation — a policy option competitors will find hard to replicate.

TPB's logic can be summarized in one sentence: it is not betting on a single product, but on whether a breeding platform can significantly lower the cost curve of crops. Once that holds, the commercial returns from a single crop (almonds, strawberries) are only the first step — the platform can migrate to more staple crops.

7. Holding-Period Stratification: The True Structure of Long-Termism as Seen from the Portfolio

The trust portfolio's holding-period distribution deserves close reading — it reveals how a long-term investor actually allocates time risk:

Holding period	Representative holdings	Share of assets
10+ years	Amazon (3.6%), ASML (3.2%), Ferrari (1.9%), Wise (1.9%)	17.5%
5–10 years	SpaceX (19.3%), ByteDance (4.7%), Ant International (0.7%)	Estimated >25% (text cut off)

Several key observations:

1. The largest position (SpaceX at 19.3%) sits in the 5–10 year range, indicating the trust expects a clear liquidity event (IPO or secondary-market resale) for SpaceX within 5–10 years, rather than holding indefinitely.

2. The core assets held for more than 10 years are all listed "compounding machines" (Amazon, ASML, Ferrari); they do not need a liquidity event and can generate returns through organic growth.

3. Long-term investing is not "never exiting," but rather precisely calculating the liquidity time window for each asset and allocating capital across lock-up periods of different lengths, thereby using time to earn private-market premiums.

This stratification is essentially a time-arbitrage strategy: for the same company, if the market offers a 1x liquidity premium, the trust is willing to pay with a 5–10 year lock-up; if it is a perpetual-growth asset (such as Amazon), it is placed directly in the 10+ year portfolio with no exit plan.

Distinguishing "Active Choices" from "Passive Outcomes": Three Layers of Information Behind Position Changes

The previous section analyzed the holding structure and concentration. Here we change perspective: the most valuable aspect of the annual report's investment list is that it simultaneously reveals what the fund manager did (trading behavior) and what the market did (price outcomes). These two are often conflated, but in practice they frequently diverge.

1. Same Company, Same Ruler: The Valuation Transmission Logic of Multi-Round Holdings

An easily overlooked detail in this annual report is that for different rounds of preferred stock in the same private company, the fund applied almost identical year-on-year valuation multiples. This is not accidental; it is direct evidence of the valuation methodology for the private portfolio.

Company	Round	2025 fair value (£'000)	2026 fair value (£'000)	YoY multiple
SpaceX	Series J Pref.	462,346	1,288,143	2.79x
SpaceX	Series N Pref.	371,591	1,035,290	2.79x
SpaceX	Class A Common	181,268	505,030	2.79x
SpaceX	Class C Common	55,911	155,773	2.79x
Databricks	Series H Pref.	87,478	175,876	2.01x
Databricks	Series I Pref.	8,629	17,349	2.01x
Databricks	Series J Pref.	16,068	32,305	2.01x
ByteDance	Series E Pref.	297,943	385,856	1.30x
ByteDance	Series E-1 Pref.	268,039	347,128	1.30x

SpaceX's four different rounds (from Series J to Class C Common) all increased by exactly 2.79x year-on-year; Databricks by 2.01x; ByteDance by 1.30x. This means the fund does not adjust each investment individually based on transaction prices, but instead applies proportionate revaluations to positions in each round based on the portfolio company's overall latest funding valuation. This helps outside investors understand how much of the book-value increase in private holdings (SpaceX rose from £1.07 billion to £2.98 billion in one year) was driven by valuation multiples rather than additional fund contributions.

In contrast, Stripe's performance is somewhat "irregular": Series G, Series I, and Class B Common rose in tandem by 1.73x, while Series H rose only 1.63x. This subtle divergence may reflect differentiated treatment of liquidation preference terms across rounds during revaluation — the annual report does not say so explicitly, but the cracks in the numbers already provide clues.

2. Trading Direction × Price Outcome: A Four-Quadrant Rebalancing Matrix

By selecting the holdings marked "Significant reduction" and "Significant purchase" in the annual report and layering on price outcomes, we can construct a highly informative matrix:

Trading behavior	Price rising / valuation rising	Price falling / valuation falling
Active addition / new position	Nu Holdings (+137.5%)<br>CATL (new position 1.6%)<br>Anthropic Series F-1 (new position 1.1%)<br>Zipline Series H (follow-on)	Sea (position value −19%, but clearly marked "Significant purchase")
Active reduction	ASML (+29.7%, marked significant reduction)<br>Cloudflare (+22.3%, marked significant reduction)	Amazon (−28.8%)<br>Meta (−35.5%)<br>PDD Holdings (−40.5%)<br>Spotify (−39.7%)<br>Adyen (−40.3%)<br>Tempus AI (−42.6%)

This matrix reveals several details that have not been fully explored before:

First, ASML and Cloudflare are the most typical cases of "judgment diverging from the market." During the year, the fund actively reduced ASML and Cloudflare, but the share price gains fully offset the impact of the reduction, so the final position values actually rose by roughly 30% and 22%, respectively. The footnotes also confirm this: the significant reduction was offset by share price gains. In other words, the "quantity" of the fund manager's reductions was right, but the "timing" was wrong — this is neither a smooth exit like Amazon/Meta, where price fell in tandem with the reduction, nor a complete failure.

Second, Sea is another mirror-image case. The fund clearly marked a significant purchase of Sea — a large addition — but the year-end position value nonetheless fell from £273 million to £221 million (−19%). This is a classic sample of "adding but the price falls": the fund actively took on headwinds, using real money to express conviction in the company's fundamentals, but the market has not yet given positive feedback. The annual report does not flag right or wrong in advance, but it presents the results truthfully.

Third, Nu Holdings is a specimen of "addition and price resonance." The position rose from £8.57 million to £203.6 million, a gain of 137.5%, while also carrying the significant purchase marker. This shows the fund was both actively buying and enjoying the tailwind of valuation appreciation — two forces compounding in the same direction to produce such a marked position change.

3. Write-Off Cases and LP-Style Positioning: The Other Side of the Private Portfolio

Hidden in the first list are three "–" symbols: Indigo Agriculture, Northvolt, and Intarcia Therapeutics. This means all three investments have been written down to zero. Northvolt was once a landmark unicorn of the European battery industry but ultimately went into bankruptcy restructuring; Indigo Agriculture and Intarcia Therapeutics respectively failed in agricultural technology and pharmaceuticals.

Even though SpaceX, ByteDance, and Stripe contributed enormous gains, the power-law distribution of private investment remains real: a few leading projects contribute the vast majority of returns, while failed projects in the long tail ultimately go to zero. The annual report does not shy away from this — it presents them directly as "–", rather than dressing them up as "small positions."

Meanwhile, the list includes at least five ARCH Ventures family funds (ARCH Ventures, Arch Venture Partners Overage, ARCH Ventures Fund X, ARCH Venture Fund XII, ARCH Ventures Fund XIII), as well as Antler East Africa Fund, each less than 0.1% of net assets. This shows the fund not only holds startup equity directly but also uses LP positions in specialist venture funds for indirect exposure — particularly in early-stage biotech (the ARCH family) and African venture investing (Antler). These "micro positions" are not profit contributors in the financial sense, but rather the fund's way of maintaining information reach and deal flow within the private ecosystem.

4. Shifts Within the Payments Track

Payments is one of the fund's persistently heavy sectors, but in 2026 the internal structure clearly diverged:

Company	2025 (£'000)	2026 (£'000)	Change	Trading behavior
Stripe (all rounds)	353,938	609,386	+72.2%	Held across multiple rounds, no marked significant reduction
Adyen	264,794	158,127	−40.3%	Significant reduction
MercadoLibre	807,112	615,868	−23.7%	No marked significant increase or decrease

Stripe's four rounds combined are now very close to the single MercadoLibre position, while Adyen has been visibly reduced. All three are payment/e-commerce infrastructure, yet the fund treated them completely differently: Stripe was held long-term with valuation mark-ups, Adyen was actively reduced, and MercadoLibre was left untouched, accepting the natural pullback in market value. This within-sector operational divergence is more revealing of the fund manager's stock-specific judgment than cross-sector moves — it removes the noise of "industry beta," leaving only the trade-offs at the alpha level.

Summary

Looking at the annual report's trading markers and price outcomes together reconstructs a narrative far more three-dimensional than a simple holdings list: the uniform upward revaluation of SpaceX and Databricks was the main engine of the private portfolio's book-value gains; Nu Holdings' synchronized resonance was the ideal outcome of adding; ASML and Cloudflare's "reduce while rising" shows how market forces can partially correct human judgment; and Sea's "buy while falling" and Northvolt's write-off are costs that must be borne along the way. Only by understanding these layers can one truly read the annual report of a trust with a high private-market weighting.

Reading Through Market-Value Noise via "Transaction Markers": Active Rebalancing Signals in the FY2026 Portfolio

The most easily overlooked detail in this disclosure is the disconnect between "trading behavior" and "market-value changes" revealed by the footnote system. Only by reading this layer can one truly understand what SMT did and did not do in FY2026.

1. Market-Value Change ≠ Active Trading: Three Notable Cases of Divergence

Holding	2025 market value (£'000)	2026 market value (£'000)	Market-value change	Transaction marker	Interpretation
Epic Games	141,897	87,480	−38.3%	No marker	Not sold; decline from revaluation of unlisted assets
Insulet	160,835	125,873	−21.8%	No marker	Market-value decline just exceeds threshold, but no trading behavior
Kering	99,230	120,525	+21.5%	Significant reduction	Active large sale, but share price gains masked the scale of the reduction
DoorDash	226,576	140,405	−38.0%	Significant reduction	Both volume and price fell; a true "double blow"

Kering is the most intriguing case. The market value did not fall but rose, yet it is marked "significant reduction" — this means SMT executed a substantial reduction while Kering's share price was rising, with the transaction amount at least 20% of the initial position value, but the share price gain offset the impact of the reduced position on portfolio market value. This is a classic "exit during an uptrend," fundamentally different from a simple "cut." In contrast, Epic Games fell 38% with no sell marker, indicating that SMT chose to absorb the valuation fluctuation in its unlisted assets — a stark contrast to its approach to listed companies.

2. The "Round Matrix" of Private Equity: A Deeply Entrenched One-Way Structure

This disclosure reveals an extremely clear pattern: for private companies it favors, SMT does not invest in one round and leave, but continues to follow on in subsequent funding rounds with multiple round additions. The pros and cons of this strategy can be seen at a glance in the table below:

Company	Rounds/classes held	Total market value (£'000)	% of total investment assets
Thumbtack	9 classes (Series A Common to Series I Pref.)	53,441	0.3%
Blockchain.com	3 classes (Series C-1, D, E)	129,066	0.9%
Redwood Materials	3 classes (Series C, D, E)	101,576	0.7%
Nuro	3 classes (Series C, D, E)	61,867	0.4%
RedNote	5 classes (Common, A, B-1, C, E)	69,516	0.4%

This "full-round coverage" means SMT is not just entering as a financial investor, but is more like building a long-term capital partnership. However, this also brings the inherent contradiction of private equity investment: rounds can only be added, rarely reduced. When a company progresses from Series C all the way to Series I (like Thumbtack), SMT has effectively lost the ability to exit flexibly; its exit path depends entirely on IPO or M&A. Thumbtack fell from 61,651 to 53,441, with simultaneous shrinkage across all rounds, reflecting an overall valuation downgrade — a systemic downward revision that diversification across private rounds cannot hedge.

Notably, there is a "valuation convergence" phenomenon among different rounds of the same company. RedNote's Series C and Series E preferred stocks have exactly the same market value (6,947 each), as do Series A and Series B-1 (13,917 each). On the one hand, this may reflect conversion features between rounds (such as anti-dilution clauses); on the other hand, it may also imply that in the absence of an active market, SMT applies a more conservative valuation logic to later rounds.

3. The Autonomous Driving Landscape: An Underappreciated Investment Theme

If the relevant holdings in this section are pieced together, one can see that SMT has built a rare and complete value chain in autonomous driving:

Chain position	Company	Market value (£'000)	2026 change	Role
AI chip/algorithm	Horizon Robotics	84,096	−3.7%	Provides driver-assistance chips
Vehicle/energy	Tesla	76,050	−31.5% (reduced)	Electric vehicles and autonomous driving
L4 autonomous driving solution	Aurora Innovation (A+B)	244,318	+67.0%	Driverless systems
Last-mile delivery	Nuro	61,867	−1.6%	Autonomous delivery vehicles

Aurora's dual-class holding (Class A + Class B) totals 244,318, the largest autonomous driving position in the current portfolio, with both classes recording a synchronized 67% gain. This gain far exceeds the average market return over the same period, pointing to substantial progress in Aurora's commercialization. SMT's simultaneous holding of both Class A and Class B in the same company usually means it participated in pre-IPO financing and went through the IPO conversion — Class A is the publicly traded portion, while Class B may be a high-voting-class reserved for founders or early investors. The synchronized 67% rise in both classes suggests SMT applied no difference in liquidity discount between the two classes, and the valuation methodology is consistent.

In contrast is Tesla's "significant reduction" — SMT actively reduced its position in Tesla while allocating capital to purer autonomous driving technology companies (Aurora, Nuro). Horizon Robotics, as an AI chip company, is more like buying an "upstream option" on the entire autonomous driving investment theme at the semiconductor level.

4. What New Positions Have in Common: A "Trio" of AI-Native Companies

In FY2026, three striking "New purchase" markers appeared — MiniMax, Applovin, and RedNote. They sit at different levels of the AI value chain, forming a fairly complete incremental allocation logic:

• Model layer: MiniMax — a developer of AI foundation models and applications, directly betting on the technological evolution of large models;
• Distribution layer: Applovin — a mobile gaming advertising technology platform, essentially going long on AI-driven precision ad delivery;
• Content layer: RedNote — a Chinese lifestyle community and e-commerce platform; AI recommendation algorithms are the core engine of its content distribution.

These three new buys are fundamentally different from the AI beneficiaries already held (Netflix, Roblox, Epic Games, etc.): the latter are companies that "use AI to optimize their businesses," while the former are companies where "AI itself is the product." This shift indicates that SMT's AI allocation focus is moving from application-side beneficiaries toward infrastructure and model-native layers. RedNote is particularly notable — with a combined weight of 0.4%, it entered the portfolio across five rounds at once, the deepest new position in this section's disclosure, and it implies that SMT's renewed bet on Chinese consumer internet is not a tentative allocation but a fairly resolute decision to build a position.

5. Chinese Assets: A "Two-Line Return"

Combining Ant International (101,969, 0.7%) and RedNote (69,516, 0.4%), SMT has formed a two-line allocation to Chinese assets: "fintech + content e-commerce." Together they total approximately 171,485, close to 1.1% of the portfolio's total assets. Notable points:

Holding	2025 market value (£'000)	2026 market value (£'000)	Change	Marker
Ant International	125,012	101,969	−18.4%	No marker
RedNote (all classes)	–	69,516	New position	New purchase

Ant International's market value fell 18.4% with no trading marker, meaning SMT still holds the original position; the decline reflects a valuation downgrade rather than an active exit. Building a position in RedNote while staying put in Ant forms a contrast — SMT adopted a "one enter, one hold" strategy toward different names in the same country market. This is less a country judgment than a selection based on individual company quality.

6. Conclusion: Three Principles Extracted from the Data in This Section

The information conveyed by this batch of portfolio disclosures can be summarized into three clear principles:

First, SMT's reductions were active and disciplined. Selling Kering during an uptrend, cutting Roblox to half a position, reducing Tesla by 31%, and trimming Affirm Class A by 54% — these actions point to a unified rebalancing logic: taking profits on names whose valuations already fully reflect expectations, rather than passively waiting for the market to turn.

Second, exiting unlisted assets is extremely difficult. Epic Games fell 38% yet was not sold; Thumbtack saw declines across all nine classes yet remains held — the low liquidity of private equity dictates that SMT's strategy in this area can only be "choose right and hold long-term"; once a mistake is made, the cost of correction is extremely high.

Third, the direction of incremental capital is clearly toward AI. The three new positions — MiniMax, Applovin, and RedNote — all point to AI-native areas, while Aurora's 67% gain validates the return potential of this direction. The FY2026 portfolio rebalancing is essentially reallocating gains earned in consumer internet and traditional technology to the frontier of the AI technology cycle.

Continuation Analysis: Key Signals at the Individual-Stock Level — The Value Realignment from 2025 to 2026

Building on the previous discussion of the overall portfolio landscape, this section draws on the individual-stock holdings detail disclosed in the FY2026 annual report and focuses on four dimensions: the commonalities of significant reduction cases, the strategic direction of new positions, the separation of trading behavior from price movements, and the warning signals of extreme valuation mark-downs.

1. High-Valuation Growth Stocks Ebb Collectively: Collapse from Millions of Pounds to Hundreds of Thousands

Table 1 lists the five holdings with the largest fair-value declines this year (all are continuing holdings rather than liquidated positions). Notably, three of them are marked "Significant reduction" (active trimming), while the other two lost more than half their value through price declines alone.

Company	2025 fair value (£'000)	2026 fair value (£'000)	Change	Notes
The Brandtech Group	127,078	13,888	-89.1%	No trading marker; valuation was the main driver
Oddity	101,802	30,885	-69.7%	Significant reduction
Delivery Hero	88,614	31,206	-64.8%	Significant reduction
Lumeris Group Holdings	36,133	17,384	-51.9%	No trading marker
Solugen Inc.	46,336	22,728	-50.9%	No trading marker

Core finding: The Brandtech Group's decline is nearly 90%, and its footnote has neither "†" nor "#", meaning there was no active trading reaching the 20% threshold during the year. This near-"zeroing" shrinkage therefore came entirely from valuation revaluation — most likely a major downward adjustment in the internal valuation model for an unlisted asset (such as a contraction in revenue multiples or a down-round discount). In contrast, the active reductions in Delivery Hero and Oddity, combined with falling prices, suggest the fund developed fundamental doubts about these two companies' fundamentals or valuation logic.

2. New Positions: A Three-Line Defense in Climate, AI, and Biotech

This year saw four "New purchase" and several "Follow-on purchase" transactions, totaling approximately £92.6M (calculated based on the initial value of new additions in the table). The capital flows were highly concentrated in three major themes:

Company	Type	2026 fair value (£'000)	Industry
MongoDB	New purchase	49,541	Database software (AI infrastructure)
Figma	New purchase	24,058	Collaborative design platform (software)
Loyal Animal Health	New purchase	18,958	Veterinary medicine (biotech)
Climeworks F-2	Follow-on	4,719	Direct air carbon capture (climate)
PsiQuantum E	Follow-on	5,439	Quantum computing (frontier tech)
Sana Biotechnology	Follow-on	14,384	Cell therapy (biotech)

Analysis: MongoDB and Figma are both mature software companies that have achieved commercialization, forming a risk offset against the early-stage, unprofitable technology companies in the portfolio. The F-2 round in Climeworks and the E round in PsiQuantum show the fund's willingness to keep adding in "hard tech." Notably, these follow-on investments all occurred between April 2025 and March 2026, when the market was generally cautious toward unprofitable growth stocks; the fund's counter-cyclical deployment reflects a long-termist inclination.

3. Trading Behavior Masked by Prices: The "Buy More as It Falls" Signal

The annual report footnotes distinguish "significant increase offset by share price decline" (#) from "significant reduction offset by share price rise" (†). Multiple "#" markers appear in this portfolio, meaning the fund actively increased positions during declines, yet the final book value still fell. For example:

Company	2025 fair value (£'000)	2026 fair value (£'000)	Book-value change	Actual trading behavior
Recursion Pharmaceuticals	74,345	42,173	-43.3%	Significant increase, but the share price fell even more
Sana Biotechnology	7,585	14,384	+89.6%	Significant increase, book value still grew
Ginkgo BioWorks	10,543	11,081	+5.1%	Significant increase, share price slightly down

Among these, Recursion is the most typical case: the fund held £74.3M in 2025, made a large additional purchase during the period (exceeding the 20% threshold), but the share price kept falling, leaving the final market value nearly halved. This "add on weakness" strategy both reflects the fund manager's conviction in the long-term value of the target and exposes the high share-price volatility risk of unprofitable biotech companies. In contrast, Ginkgo and Sana saw book values rise after the additions, indicating that some of the increased positions are already starting to pay off.

4. Passive Crashes and Liquidity Risk: Unlisted Assets That Have "Disappeared" by 99%

Aside from Brandtech, other unlisted companies such as Honor Technology, JRSK (Away), and GoPuff remained broadly stable (changes within ±10%), demonstrating the stickiness of private-market valuations. But Brandtech's -89% is an extreme outlier, possibly stemming from:

• The company suffered a major operational setback (e.g., customer attrition, management changes)
• A new funding round was conducted as a "down round," triggering a valuation reset
• The internal model was sharply cut due to falling valuations among comparable companies

Since this investment is not on the significant reduction list, either the reduction occurred after the fiscal year-end, or it was entirely an accounting valuation adjustment. In either case, it suggests that the "book value" of unlisted assets in the absence of liquidity may be lagging — today's £13.9M could continue toward zero tomorrow.

5. Summary and Outlook

This year's position changes clearly show three trends:

1. Increased allocation to mature tech giants (MongoDB, Figma), indicating the portfolio is tilting toward "earnings quality";

2. Decisive reductions in high-valuation consumer growth stocks (Oddity, Delivery Hero), demonstrating the execution of risk-control discipline;

3. Continued heavy position in biotech and climate tech, with the fund choosing to add even amid violent short-term price swings.

The next section will continue analyzing the remaining holdings (including the WI Harper fund and the Global AI Opportunities Fund, among others) to assess the overall portfolio's diversification and tail-risk exposure.

Continuation Analysis (Part 16/22)

1. The "Bipolarization" and "Shellization" of the Portfolio Structure

The holdings detail in the continuation reveals a structural feature: a very small number of core holdings support the entire portfolio, while a large number of marginal positions have become "nominal presences." In data terms, the five funds in the ARCH Ventures series Nine through Thirteen have a combined book value of approximately £22.71 million, with each representing less than 0.1% of total assets — functionally these investments are close to "holding certificates" rather than active capital allocation. More notable is a batch of assets marked "–" at the end of the table, including two series of preferred stock in Relativity Space and Capsule Corp, as well as all six instruments of Northvolt — their book values have gone to zero, but they are still itemized in the annual report.

Asset status	Representative assets	Book-value performance	% of total assets
Written to zero but still disclosed	Northvolt (six series), Relativity Space, Capsule Corp (two series)	All – (zero)	<0.1%
Generating modest value	Bolt Projects Holdings	£9k → £586k	<0.1%
Fully exited	Indigo Agriculture, Intarcia Therapeutics	£455k → –	–

This "zero value but not removed" disclosure approach has dual significance. On the one hand, it reflects the trust's commitment to transparency — even when an asset has effectively gone to zero (such as Northvolt's bankruptcy restructuring), it must still be accounted for item by item to shareholders. On the other hand, it exposes the practical problem of "difficulty in exiting" private equity investments — some projects can neither list nor find secondary-market buyers, and can only retain a formal place on the books until legal liquidation proceedings conclude.

2. The "See-Saw" Among Asset Classes: Public Markets Stepping In for Private Markets

The structural comparison figures at the end of the annual report are the densest set of numbers in the entire continuation:

Asset class	As of 31 March 2026	As of 31 March 2025	Change
Unlisted (private) securities	58.4%	72.0%	–13.6 percentage points
Listed securities	41.4%	27.5%	+13.9 percentage points
Bonds	0.1%	0.2%	–0.1 percentage points
Net current assets	0.1%	0.3%	–0.2 percentage points

The interpretation of this change should not stop at the surface-level narrative of "the private share has fallen." Combined with the earlier table note that "both pre- and post-listing holdings are covered," the 13.9-percentage-point increase in listed securities most likely does not stem from active buying of public-market stocks, but rather from private holdings being "passively transferred" into the listed category after IPOs or SPAC listings — as the footnote explains, "securities previously held in private form are now listed." In other words, this reflects not a strategic shift, but that a large number of private investments from prior years have reached their exit window. The coexistence of the Illumina CVR and the "pre-IPO private" category further confirms this judgment: the original holding may have already listed in an earlier fiscal year, with the company retaining only a CVR derivative interest.

3. Another Way to Calculate Concentration Risk

The investment policy clearly states that "no single holding may exceed 8% of total assets," but from the overall portfolio structure, the true concentration risk is not at the individual company level but at the industry and stage dimensions. The dense appearance of ARCH Ventures funds (five funds Nine to Thirteen + X Overage) means the trust is concentrating its bets on early-stage biotech projects through multiple vintage funds managed by the same GP; Upside Foods and Indigo Agriculture show systematic exposure to the "alternative protein" and "agricultural technology" themes. This kind of thematic concentration cannot be avoided by individual holding caps — each fund individually is within 8%, but the multiple stacking of the same manager, the same track, and the same round makes the actual risk exposure far greater than the superficial diversification suggested by the books.

4. Reflection on the Substantive Effectiveness of the "30% Private-Market Cap"

The policy states that "private investments may not exceed 30% of total assets (measured at the time of purchase)," yet the actual private share disclosed in the annual report is as high as 58.4% — even accounting for the difference in measurement basis (purchase time vs. reporting time), this is a significant tension between policy intent and operational reality. A reasonable explanation is: the limit is only binding on "incremental purchases," and cannot cope with the "natural appreciation of existing assets" — if private assets outperform public-market assets, their share will be gradually pushed above 30% even without any new private investment. This structural contradiction is systematically amplified in years when private equity appreciates over the long term or public markets pull back; it is a design flaw in the policy framework worth attention.

5. Net Current Assets of Only 0.1%: The "Double-Edged Sword" of Full Deployment

Of total assets of £15,428 million, net current assets amount to only £18.9 million (0.1%). This means the trust is almost 100% fully invested — it has neither a defensive cash buffer nor "dry powder" for capturing market mispricing opportunities. Combined with the investment policy statement about "using market volatility to serve shareholders' long-term interests," in practice the feasibility of this strategy depends entirely on "selling existing holdings to buy new targets" — which is feasible for highly liquid stocks (such as the listed portion), but for the 58.4% of the portfolio held in private investments, the liquidity difference makes this operation extremely costly. The all-in state combined with a private-dominated portfolio structure essentially abandons the timing flexibility implied by the investment policy.

6. The Tension Between Holding-Period Rhetoric and High-Frequency Change Markers

"Average holding period of five years or more" is the company's core positioning, but the density of "*" markers in the continuation reveals a subtle contrast: a large number of holdings are marked as having experienced "a significant increase or decrease of at least 20% in value," including both quantity changes from active trading and passive value changes from share price movements and exchange-rate fluctuations. Illumina is marked as "significant reduction offset by share price rise," while some private funds show "significant increase offset by share price decline" — this means the annual changes in the portfolio are more the result of market forces than traces of active trading. The long-term holding narrative and the fact of passively bearing market fluctuations can coexist, but the way they are presented in the annual report — attributing all changes to "transaction value" footnotes — makes it easy for investors to misread passive fluctuations as active rebalancing.

1. Revision of the Private-Investment Cap: "Conditional Flexibility" Under a Three-Tier Governance Structure

The core of this investment policy revision is not simply raising the cap, but building a three-tier progressive constraint mechanism:

Tier	Constraint	Function
First tier	Hard cap of 30% private-investment share	Normal risk boundary
Second tier	£250 million Additional Private Investment Capacity	Additional investment space beyond the cap (cumulative no more than £250m between two AGMs)
Third tier	Subject to shareholder renewal at the most recent AGM (with effect from the conclusion of the 2027 annual general meeting onwards)	Annual sunset clause ensuring shareholders revisit the authorization each year

This structure is ingenious at the governance level: the £250m cap is not calculated per occurrence, but per "AGM period." Clause (iii) clearly states that no matter how many times the 30% limit is exceeded within the same AGM period, the total £250m cap cannot be refreshed. This effectively prevents the company from circumventing the authorization by exceeding the limit in a series of small increments.

A particularly notable detail is that the trigger context for this revision is "market movements, revaluations of private investments and share buyback programme" — that is, the cap was exceeded passively, not through active adding. The company repurchased shares → total assets shrank → the private-investment share rose passively. The £250m additional capacity is, to some extent, a "cushion" set up for the by-product of the buyback programme, rather than a purely offensive allocation tool.

2. Buyback Programme Execution Rate: A Notable 310% Overshoot

A set of data disclosed in the annual report is worth unpacking: the two-year "at least £1 billion" buyback programme announced on 15 March 2024 was actually executed with 318.6 million shares repurchased at a total cost of £3.10 billion, an execution rate of 310% of the plan.

Metric	Original plan	Actual execution	Execution rate
Buyback amount floor	£1.0b	£3.10b	310%
Shares repurchased	—	318.6m	—
Implied average buyback price	—	≈ £9.73/share	—

During the year, 122,884,921 shares were repurchased into treasury, approximately 38.6% of the two-year total. Based on the average price of £9.73, the annual buyback cost is approximately £1.2 billion.

Even more notable is the asymmetric operation between 1 April and 21 May 2026: zero buybacks and the issuance of 24,750,000 treasury shares. This suggests that during that window the market may have been at a premium (or at least no longer at a significant discount), and the company switched from "buy" mode to "sell" mode. This dynamically echoes the Liquidity Policy statement that "no formal discount or premium target is set" — the policy does not mean discount/premium management is absent, but rather avoids binding execution flexibility with rigid targets.

3. The Four-Tier Constraint System for Leverage Policy

This section of the text reveals an interesting "multi-layered leverage constraint" structure:

Tier	Cap	Calculation basis	Nature
Articles of Association	50%	issued share capital + reserves	Maximum legal limit
Bank covenant	35%	adjusted net asset value / total assets	Contractual constraint
Company policy	30%	AIC guidelines	Voluntary commitment
FCA rules	—	net asset value with debt at par	External regulation

The core tension lies in the interaction between the AIC's 30% cap and the FCA's "par value determination" of NAV. The FCA determines that "net asset value" should be calculated based on the par value of debt, rather than the fair value commonly used by the market — this effectively raises the book value of NAV (when the market value of debt is below par), thereby reducing the risk of triggering the rule when issuing new shares at a discount to NAV.

The company's response strategy is highly anticipatory: even though the FCA's determination standard favors issuance, the Board nevertheless commits that "in no circumstances" will it issue new shares at a price below fair-value NAV. This is a "dual bottom line" design — it does not violate the rules legally, but it enforces a stricter standard internally to maintain shareholder trust.

4. The Dependence of Dividend Policy on Capital Reserves and Potential Risks

The statement "paid from a combination of revenue earnings, revenue reserves (if any) and distributable capital reserves (comprising mainly realised investment gains)" — the substance of this is: when revenue earnings are insufficient, dividends can be supplemented by realized capital gains.

This means the sustainability of the company's dividend depends on two prerequisites:

1. Long-term total return is positive and can continuously generate

---

These incremental observations focus on the precision of risk quantification, the fragility of assumptions, and the genuine handling of stakeholder conflicts, serving as a complementary critical lens on the annual report's existing narrative.

Continuation Analysis: Shareholder Feedback-Driven Governance Closed Loop and the "Indirect Influence" Framework

I. The Substantive Closed Loop of Shareholder Engagement: Verifiable Cases from "Listening" to "Policy Revision"

The disclosure in the continuation regarding "Matters raised with the Board and Managers" is the section with the greatest incremental information value. It demonstrates a complete shareholder engagement closed loop, rather than a mere formalistic description of communication:

Issue	Shareholder Demand	Board Action	Outcome and Follow-Up
Private company exposure	Questions regarding investment policy restrictions	Conducted extensive engagement during the year; convened a general meeting (10 April 2026)	Approved revised investment policy granting limited additional flexibility, subject to defined limits and ongoing shareholder oversight
NAV discount	Continued concern over discount levels	Reported on liquidity policy execution, including buyback scale; worked closely with brokers to optimise buyback efficiency	Discount narrowed during the year; shares traded at a premium following the year end

This disclosure carries three-fold significance:

1. Quantitative anchor: The text explicitly references the "General Meeting held on 10 April 2026", directly linking shareholder communication to a specific corporate action event (policy change). For analysts, this means shareholder engagement is not merely a governance "slogan" but a traceable decision node.

2. Self-validation of outcomes: The mention of "shares trading at a premium following the year end" provides market validation of the buyback policy's effectiveness. This is more persuasive than simply stating "we conducted buybacks", as it establishes a causal link between corporate action and market outcomes (although it should be noted that discount narrowing may be influenced by multiple market factors).

3. The "limited" framing of policy adjustment: The notable wording is "limited additional flexibility", which implies the Board maintained a prudent stance in responding to shareholder demands—not an unconditional concession, but one bounded by parameters ("subject to defined limits"). This balanced narrative enhances the credibility of the disclosure.

II. Deepening the "Indirect Influence" Framework: From "Identification" to "Transmission Mechanism"

The continuation's discussion of "Portfolio companies" as the primary carriers of real-world impact further clarifies a key investment trust governance concept—the two-tier structure of impact transmission:

• Primary impact (direct operations): The trust's own operations are extremely limited ("operations are limited"), and its direct carbon footprint and social impact are minimal.
• Secondary impact (portfolio transmission): Genuine real-world impact (whether positive or negative) emanates from the investee companies in the portfolio.

The logical extension of this framework is that shareholders' ESG demands are transmitted to investee companies through the trust. The text's mention that "investee companies have an interest in understanding their shareholders' investment rationale" effectively constructs a complete chain of accountability:

> Ultimate beneficiaries → Trust board (overseeing ESG approach) → Investment manager (reviewing investee ESG performance) → Investee companies (generating actual impact)

Notably, the wording in the continuation is more pragmatic here than in earlier sections. It acknowledges that investee companies' impact is "both positively and negatively", without sugar-coating the portfolio's ESG performance, instead emphasizing the symbiotic relationship between investment growth and impact achieved through "commercial success". This framing avoids any suspicion of "impact washing".

III. The "Risk Penetration" Mechanism in Intermediary Governance

The continuation's description of intermediaries (Registrar, Depositary, Custodian) presents a three-tier risk monitoring model that was not fully explored in prior analysis:

Intermediary	Board/Committee Monitoring Level	Specific Tools
Registrar	Investment Manager + Risk function	Liaison to ensure frequency and accuracy of communications; review of internal control reports with results reported to the Board
Depositary/Custodian	Audit Committee + Investment Manager business risk team	Depositary provides monitoring activity reports; review of Bank of New York Mellon internal control reports

The key new argument here is the involvement of the "Investment Manager's risk function". It indicates that for a critical outsourced service provider (the Registrar), there is not only day-to-day management at the business level (handled by the Investment Manager) but also an independent risk function conducting a second-line-of-defence review. This arrangement partially front-loads oversight duties traditionally belonging to the Board into the Investment Manager's risk architecture, creating a more granular risk penetration chain.

Furthermore, the described engagement strategy with the Depositary—"collaborative and collegiate manner, encouraging open and constructive discussion and debate, while also ensuring that appropriate and regular challenge is brought"—reflects the Board's search for a balance between cooperative relationship and critical oversight. This expression of "constructive challenge" is more mature than in earlier sections; it acknowledges that neither a purely adversarial nor a purely dependent relationship serves the maximisation of shareholder interests.

IV. The Strategicity of Regulatory Engagement: From "Passive Compliance" to "Active Shaping"

The paragraph in the continuation concerning Regulatory Bodies provides a specific, time-sensitive case of regulatory interaction:

• Specific entity: Financial Reporting Council ('FRC')
• Specific matter: Thematic review of annual reports and financial statements for the year ended 31 March 2025
• Company response: Sought to enhance relevant disclosures in the current annual report ("where material and relevant")

This detail offers two analytical values:

1. It confirms the pathway for regulatory feedback to land: The Company not only participated in the FRC's review but also converted the review findings into tangible disclosure improvements. This provides verifiable output for "engagement with regulatory bodies", rather than remaining at the abstract level of "maintaining good relationships".

2. Precise articulation of risk mitigation: The text notes that "Regulatory risk can be mitigated by making representations to regulators regarding the specific circumstances of investment companies." This reveals the Company's proactive strategy in regulatory engagement—not passively waiting for rules to be issued, but actively explaining to regulators the special circumstances of the investment trust industry (e.g., discounts, liquidity management, private asset valuation) in order to influence the formation or enforcement of regulatory policy.

Additionally, the disclosure of the specific year of the "thematic review" (2025) and the document version (2026 annual report) creates a timeline that enables investors to trace how regulatory interaction is reflected in this report. This level of transparency is uncommon in the investment trust industry and is commendable.

V. Cross-Sectional Comparison of Data and Trends

Dimension	Continuation Disclosure Characteristics	Industry Common Practice	Differential Analysis
Shareholder feedback response	Provides a specific general meeting date (10 April 2026) and policy revision outcomes	Most disclosures only qualitatively describe "discussions with shareholders"	Scottish Mortgage provides auditable decision nodes, enhancing verifiability
Discount management effectiveness	Explicitly mentions "premium following the year end"	Most companies only disclose buyback scale without market outcomes	Links actions to outcomes, but timing should be treated with caution (post-year-end premium may be a short-term phenomenon)
Regulatory interaction	Names the FRC and the specific review matter	Usually vague references to "maintaining dialogue with regulators"	Naming enhances the credibility of transparency but may be constrained by confidentiality obligations
ESG impact transmission	Explicitly acknowledges that investee impact has both positive and negative dimensions	Some companies only emphasise positive impact or net zero commitments	This acknowledgment of "two-sidedness" is more honest than common "impact washing", but may also be read as indirect exoneration of portfolio ESG performance

VI. Hidden Governance Signals: The Strengthening Role of Brokers

Although the section on Brokers in the continuation is brief, it contains a noteworthy detail:

> "They also arrange opportunities for shareholders to meet the Chairperson outside the normal general meeting cycle."

This statement extends the broker's function from "market sentiment reporter" to "bridge-builder between the Board and shareholders". Combined with the earlier reference to having "engaged closely with the Company's brokers throughout the year to explore ways to maximise the effectiveness of buybacks", it can be seen that:

• The broker's role in corporate governance has transcended traditional trade execution and market information transmission, and has become more involved in shareholder relationship management and capital allocation strategy advisory.
• The involvement of the "Chairperson" indicates increased direct participation at board chair level in shareholder dialogue, which typically occurs when a company faces market pressure (such as discount concerns) or significant strategic adjustments.

VII. Potential Supplements to the "Accountability" Framework

Based on the content of the continuation, the previous "Accountability" analytical framework can be further refined into a three-tier accountability system:

Accountability Tier	Object	Core Accountability Mechanism	Evidence in the Continuation
Upstream accountability	Investment Manager (to the Board)	ESG approach review, Governance Engagement reports	"The Board's review of the Investment Manager includes an assessment of their ESG approach"
Horizontal accountability	Board (to shareholders)	Shareholder meetings, broker feedback, discount management	Detailed responses to private exposure and discount concerns, and policy revisions
Downstream accountability	Investee companies (to the Investment Manager/Trust)	Governance Engagement reports, portfolio managers' periodic reports	"portfolio managers regularly report to the Board on discussions with portfolio companies"

The refinement of this framework demonstrates that the Company's accountability system is not merely vertical (shareholders → board → manager), but also incorporates horizontal (to regulators and industry peers) and downstream (oversight of investee companies) dimensions. The "Wider society and the environment" section in the continuation pushes this to its logical conclusion—the trust's social and environmental accountability is ultimately realised through its investee companies.

VIII. Gaps and Points Awaiting Observation

Despite the relatively high granularity of the continuation's disclosures, several information gaps merit attention:

1. Missing details on private company policy revisions: Although "defined limits" and "ongoing shareholder oversight" are mentioned, the new investment caps or specific regulatory terms are not quantified. Investors cannot precisely assess the impact of the policy change on the portfolio's risk profile.

2. Cost-effectiveness of discount management not disclosed: The specific amounts or share counts of buybacks during the year and the two-year period are not mentioned in this section. Without data on buyback magnitude, the "cost" of the discount narrowing cannot be assessed.

3. Specific feedback content from the FRC review: The text only mentions "sought to enhance the relevant disclosures" but does not state whether the FRC raised specific criticisms or recommendations. This ambiguity may be constrained by confidentiality agreements, but it reduces the actionability of the information.

4. Lack of specific cases of positive contribution to "Wider society": Although it states the Company "seeks to be a positive influence", no specific examples of community engagement, environmental contribution, or social impact investing are provided. This stands in stark contrast to the detailed discussion of portfolio companies' impact earlier in the text.

IX. Conclusion: Governance Disclosure with "Auditability"

Overall, the incremental value of the continuation lies in its translation of abstract governance principles (such as ESG oversight, shareholder engagement, and risk management) into traceable, concrete actions and outcomes. In particular:

• The disclosure of shareholder feedback is no longer merely "we listened to opinions", but includes clear decision outputs (General Meeting, policy revisions, discount improvement).
• Regulatory interaction is no longer a talking point, but has a specific counterparty (the FRC) and specific outcomes (enhanced annual report disclosures).

This "auditability" is the greatest strength of the risk reporting in this annual report, and a quality from which other investment trusts can learn. It should be noted, however, that this level of disclosure may reflect the market pressures facing the Company (such as discount concerns and questions over private exposure)—it is to some extent a reactive transparency enhancement rather than entirely proactive governance innovation. Investors evaluating its disclosure quality should consider this contextual factor.

Finally, the recurring use of the word "long-term" in the continuation (e.g., "long-term approach to investment", "long-term business strategies will be supported") is not coincidental. It persistently reminds readers that the Company's governance mechanisms and shareholder communication strategies all serve the objective of long-term value creation, and that responses to short-term discount levels and private company volatility are consistently framed within the context of long-term strategy. This narrative consistency is the key thread for understanding the entire annual report's risk reporting.

I. "Zero-Employee" Governance and the "Disclosure Vacuum" of s172

The most notable governance detail in this section is the conclusion the Company reaches under the heading "Employees, human rights and community issues": "As the Company has no employees, all its Directors are non-executive and all its functions are outsourced, there are no disclosures to be made."

On the surface, this is a compliance acknowledgment of the s172 duty under the Companies Act 2006, but in substance it constitutes a governance "disclosure vacuum":

Dimension	Traditional Company	Scottish Mortgage
Employee interests	Covered by HR systems and employee representation mechanisms	No employees; the set of duty bearers is empty
Human rights and community	Supply chain, production sites, community relations	Direct suppliers are limited to professional advisers
Outsourced staff	Not treated as "employees" but substantively carry out company operations	Not elaborated in s172 disclosures
Board composition	Typically includes executive directors	All non-executive directors

The problem is: outsourcing does not mean the disappearance of stakeholder impact. Baillie Gifford, as investment manager, employs hundreds of staff whose compensation, culture, and talent attrition risk directly determine the quality of investment management; the depositary, auditor, and brokers equally constitute links in the Company's value creation chain. By choosing not to disclose on the grounds of "zero employees", the Company uses a legal-formal empty set to sidestep substantive stakeholder analysis. The Company Secretary's availability "at all times" further reinforces this governance posture of dependence on external professional advisers—the Board is not managing an organisation but supervising a set of contracts.

II. The Scale Economics of Buybacks: The Aggressive Move to Shrink Capital by 22% over Two Years

The buyback data disclosed in the report merits a tabular recalibration of its magnitude:

Period	Shares repurchased (millions)	Cost (£bn)	Implied average price (£/share)
FY2026 (year ended 31 March 2026)	122.9	1.31	10.66
FY2025–FY2026 cumulative	307.7	3.02	9.81
Percentage of share capital as at 31 March 2024	—	—	~22%

Three marginal observations:

First, the average buyback price has risen significantly. The FY2026 average price (£10.66) is approximately 8.7% higher than the two-year cumulative average (£9.81), indicating that management continued executing buybacks during the share price recovery rather than only timing the lows. This sends a stronger signal than "discount management"—even as the price reverts toward net asset value, the Company remains willing to shrink its share capital to enhance per-share metrics.

Second, a 22% contraction in share capital over two years is extremely rare in the investment trust industry. Most investment trusts have buyback rates between 2% and 5%; 22% means the scale of capital return approaches what many peers achieve cumulatively over a decade. Its direct effect is the mechanical accretion of NAV per share, but it also places higher demands on portfolio liquidity—the Company must continuously generate distributable cash without disrupting portfolio holdings.

Third, by the date of the annual report's signing (26 May 2026), the share price had already turned to trading at a premium. This is direct evidence of the buyback strategy's effectiveness, yet it also plants a follow-up question: will buybacks be suspended in a premium environment? If suspended, will the discount rebound? The report makes no commitment on this point, but it will be a core governance issue for shareholders in the coming fiscal year.

III. ESG's "Dual Identity": How Skeptics and Signatories Coexist

This paragraph is the most candid confession in the entire Strategic Report on ESG matters:

> "We are highly sceptical of the labels, metrics and box ticking…ESG has never been the starting point in our investment process. It is simply a by-product of our pursuit of long-term returns."

At the same time, the checklist shows the Company is a signatory to UNPRI and CDP, a member of ICGN and ACGA, the manager publishes a TCFD report, and it adheres to the UK Stewardship Code. This constitutes a typical "institutional compliance–investment skepticism" dual-track structure:

Dimension	Company Stance	Institutional Behavior
Investment philosophy	ESG is a "by-product", not an objective	Does not screen targets using ESG ratings
Disclosure obligations	Carbon emissions are "a first step, not an answer"	Still publishes a TCFD report
External commitments	Skeptical of labels and metrics	Signatory to UNPRI and CDP
Governance participation	Challenges rating agencies	Adheres to the UK Stewardship Code

The rationality of this dual-track system lies in the fact that the Company treats ESG as "information", not as an "objective function". It can therefore simultaneously refuse box ticking, disclose its carbon footprint, and participate in international governance networks—the three are not logically in conflict. The real tension lies elsewhere: if "we don't believe carbon foot-printing in isolation is especially helpful", then is the value of the TCFD report to provide investors with decision-useful information, or merely to maintain institutional legitimacy? The report does not answer this, but exposing the question is itself a form of transparency.

IV. The Carbon Footprint Paradox: Rating Correction Using Tesla and BYD as Examples

The Company's attitude toward carbon footprints is consistent with its investment logic: a carbon footprint is a function of industry exposure, not a function of decarbonisation ambition. Tesla and BYD are used to illustrate the paradox—electric vehicle manufacturers have higher carbon footprints due to the energy intensity of their manufacturing processes, yet they are precisely the most important drivers of economic decarbonisation.

This position has two substantive analytical implications:

• A scholarly challenge to rating agencies: If ESG scores downgrade high-carbon-footprint companies, they systematically penalise firms that provide decarbonisation solutions. The report's statement that "we are prepared to challenge them" is tantamount to a public declaration that the Company will proactively engage in adversarial dialogue with investee companies and rating agencies.
• A reinforcement of its own research process: "a carbon footprint is a first step, not an answer" implies that the Company's internal treatment of carbon emissions data is closer to "research direction" than "compliance data"—first locate the problem, then assess quality and resilience through company-level research.

V. Board Renewal: Gender Balance and Skills Enhancement

As at 31 March 2026, the Board's seven members comprised four men and three women, with women representing 42.9%, exceeding the 40% target recommended by the UK's FTSE Women Leaders Review. Combined with the appointment during the year of Heather Manners, who brings fund management and investment trust experience, the Board presents as follows:

Dimension	Composition
Gender ratio	4 male / 3 female (42.9% female)
New director	Heather Manners (fund management, investment trusts, financial markets)
Executive/non-executive	All non-executive
Employment model	No employees, fully outsourced

It is worth noting that the report devotes an entire paragraph to emphasising that it encourages service providers to consider diversity and inclusion and report annually—an innovative mechanism for transmitting DEI obligations to service providers under the "no employees" governance structure. However, its binding force is limited to "encourage" rather than "require", and its actual effectiveness still depends on suppliers' autonomy.

VI. Investment Policy Flexibility: The "Follow-On Option" Behind £250m

The additional £250m investment allowance for private companies approved at the general meeting on 10 April 2026, on the surface a pursuit of long-term growth opportunities, is in substance a dynamic follow-on option:

• At approximately 6%–8% of the existing private portfolio (estimated based on the Company's total asset scale), it would not by itself change the portfolio's risk characteristics;
• More importantly, it allows the Company to continue participating in subsequent financing rounds by existing private holdings, avoiding forced dilution due to investment policy caps;
• Using annual approval as a governance anchor, it grants flexibility while retaining shareholder control.

This reflects a long-standing dilemma in investment trust governance: follow-on investment windows for private assets are fleeting, while shareholder approval processes are rigid. The £250m "safety valve" is not a strategic shift but a buffer mechanism established between rigid governance and investment flexibility.

VII. Refinancing and Currency Structure: The Market Signal of $670m

The successful refinancing of approximately $670m of borrowing facilities conveys two messages:

1. Bank funding channels remain open to investment trusts. Against the backdrop of liquidity-driven discounts, this demonstrates that lenders still recognise SMIT's portfolio quality and cash flow recovery capacity.

2. The maintenance of US dollar liabilities carries strategic implications. With substantial US dollar assets in SMIT's portfolio (Tesla, Amazon, Moderna, among others), maintaining US dollar borrowings naturally forms a currency hedge, reducing the disturbance of exchange rates on shareholder returns.

The Company's emphasis on "measured use of gearing, where appropriate" indicates that its leverage policy is unchanged, but in an environment where interest rates and buybacks run in parallel, balance sheet management requires more precision than ever—there is a delicate balance between buybacks consuming cash and borrowings maintaining liquidity.

VIII. The "Self-Exemption" in the Modern Slavery Statement: Underestimated Supply Chain Risk

The Company declares itself outside the scope of the Modern Slavery Act 2015, on the grounds that its suppliers are "typically professional advisers". This judgment holds legally, but it exposes a real-world blind spot: under a zero-employee, fully outsourced business model, the Company's primary "supply chain" is precisely its professional service providers—investment manager, depositary, brokers, and auditor. The internal governance, labour practices, and outsourcing arrangements of these institutions are themselves part of the Company's operations. Defining supply chain risk as "low risk" only examines the first layer of contractual relationships, without asking whether these professional firms' own supply chains (e.g., depositary banks' data centre labour, brokers' technology outsourcing) are equally low risk. This statement once again illustrates the gap between "legal-formal compliance" and "substantive risk scanning".

IX. Closing Observation: A Strategic Report That Treats "Restraint" as Its Governance Philosophy

Taken together, this Strategic Report displays a rare self-consistency in its governance posture: aggressive buyback scale, skeptical ESG stance, flexible investment policy, yet highly streamlined s172 disclosure. On the surface, this appears a contradictory combination—aggressive capital allocation coexisting with conservative disclosure style—but in essence they share the same value hierarchy: long-term shareholder returns as the sole criterion; everything else is either a tool or noise.

It is precisely this hierarchy that allows the Company to simultaneously be "skeptical" and "signatory" on ESG issues, to avoid disclosure on employee matters through "zero employees", to downshift the carbon footprint debate with "a first step, not an answer", and to commit £3.02bn to buybacks without regard to scale. This consistency explains Scottish Mortgage's long-term behavioural pattern better than any single strategy: long-termism with a logically closed loop often accompanies a minimalist posture toward institutional compliance. Whether this posture always aligns with the expectations of all stakeholders will be a lasting theme in future governance debates.