Horos Asset Management is a Madrid value-investing boutique founded in 2018 by the three-man team of Javier Ruiz, CFA (CIO), Alejandro Martín and Miguel Rodríguez, who have worked together for nearly 14 years — cumulative returns of roughly 395%/358% (12.3%/11.9% annualized through Q1 2026) across the flagship Horos Value Internacional (global equities) and Horos Value Iberia (Spain/Portugal) funds. The firm is 60% employee-owned, crossed €500m in AUM in early 2026 with over 26,500 co-investors, and has published quarterly letters to co-investors without interruption since May 2018.
This is a quarterly letter from investment firm Horos to its partners. The key message: the recent stock market surge looks exciting, but the author is worried. He thinks AI and semiconductor stocks are overheated—SpaceX just went public with a huge valuation despite losing money, and some chip stocks rose thousands of percent in a year. At the same time, long-term interest rates are high, making borrowing costlier. For everyday investors, the takeaway is to stay calm and avoid chasing hype. The letter is worth reading because it uses data to explain the risks lurking beneath the market's surface.
The author maintains a 'cautious' stance on global markets: the strong equity rally in Q2 masks an unsustainable capital cycle and market frenzy, with persistently high long-term interest rates and the AI infrastructure investment wave forming dual pressures.
The three Horos funds returned 4.9%, 6.6%, and 2.1% in the quarter. The author stresses that short-term results carry little significance; over 14 years, the International strategy has compounded at 12.4% annually (420% cumulative) and the Iberian strategy at 12.2% (388% cumulative).
At the start of the letter, the author notes that the big-tech data center investment cycle highlighted in previous letters continues, its impact having now spread across the entire industry chain, and that SpaceX completed the largest IPO in history. Even so, the author remains concerned about the "turmoil beneath the surface" that the market is ignoring, and therefore stays cautious toward certain markets and companies. This stance has caused the portfolio to "decouple" from the performance of many indices—which could turn out better or worse—but the author says he will not abandon his investment principles, and the goal remains to achieve satisfactory and sustainable long-term returns.
Performance comparison: this quarter, Horos Value Internacional returned 4.9%, Horos Value Iberia 6.6%, and Horos Patrimonio 2.1%. The author says these short-term results are "not meaningful." The team's 14-year cumulative returns: International strategy 420% (12.4% annualized), Iberian strategy 388% (12.2% annualized). A footnote explains that the cumulative performance includes results achieved while the team was at its previous asset management firm (International strategy from May 31, 2012; Iberian strategy from September 30, 2012; until joining Horos on May 22, 2018).
The author's overall stance on global markets is [cautious]: he believes the second-quarter equity rally masks an unsustainable capital cycle and clear market euphoria, as evidenced by SpaceX's historic IPO and record gains in foundries and memory chips.
The author draws an analogy to Simon & Garfunkel's "Bridge over Troubled Water": an index is like a bridge that lets people cross the river smoothly without seeing the rapids beneath. He notes that the de-escalation of the Iran conflict and severe bottlenecks in the semiconductor industry have reignited global equities, particularly benefiting the U.S. and several Asian markets. The author's exact words: "these sharp gains once more conceal capital cycles that could prove unsustainable, along with clear euphoria and complacency in financial markets." He also argues that the largest IPO in history (SpaceX) and the record gains in foundry and memory chip makers are all manifestations of market complacency.
Market performance data:
| Index | Quarterly Change | First-Half Change |
|---|---|---|
| S&P 500 | +15% | Over 10% |
| Nasdaq 100 | Nearly 28% | Over 20% |
| European Stoxx 600 | Nearly 12% | +11% |
| Germany DAX | — | +2% |
| France CAC 40 | — | +6% |
| Italy MIB | — | +18% |
| Spain Ibex 35 | — | Nearly 15% |
| Japan Nikkei 225 | — | +40% |
| India Sensex | — | -9% |
| Hong Kong Hang Seng | — | -9% |
| South Korea KOSPI | — | Over 100% |
| Taiwan TAIEX | — | +60% |
Note: the Nasdaq 100 posted its strongest monthly performance in 23 years in April; the Hang Seng Index was dragged down by falling Chinese real estate investment (-16.2% year-over-year in January-May) and weak consumption. The author notes that the performance of other financial assets may already hint that "some sections of the bridge are beginning to crack."
Portfolio activity this quarter: the International strategy added three stocks and sold two; the Iberian strategy bought two stocks and sold one; the capital-preservation strategy adjusted bond positions in four companies.
As is customary, the author summarized the second-quarter changes:
The author believes long-term interest rates are being pushed up by fiscal deficits and the AI data center investment wave, forcing central banks to turn hawkish; U.S. and UK 30-year government bond yields are around 5% or higher, while Japan's 30-year yield is about 4% and at a record high.
The author says concerns about entrenched inflation have driven major central banks to act on two fronts: the European Central Bank raised rates for the first time in nearly three years, and new Federal Reserve Chair Kevin Warsh has also struck a hawkish tone, vowing that "the Committee will achieve price stability." In the author's view, the upward pressure on energy prices from the Iran conflict may ease as it gets priced into oil, but two structural factors will keep rates elevated and above inflation (positive real rates): first, persistent public deficits in many countries require continued bond issuance, draining savings from the system and pushing up financing costs; second, U.S. tech giants that once injected large amounts of cash into the financial markets each year (buybacks, dividends, bond purchases) are now instead absorbing savings into AI data centers and infrastructure investment, adding further upward pressure on rates.
High rates are hitting long-duration assets: the U.S. 30-year Treasury yield is around 5%; the UK 30-year is comfortably above 5% amid budget problems and yet another prime ministerial resignation on Downing Street; Japan's 30-year yield is around 4%, having briefly broken above that level in May to a record high, and the yen's persistent depreciation also reflects market doubts about the sustainability of Japan's debt. The author also cautions that the divergence in U.S.-Japan monetary policy could fuel carry trades, with funds flowing out of Japan injecting liquidity into global markets and amplifying existing volatility. This is precisely why the author emphasizes maintaining [caution] and focusing on margin of safety.
The capital cycle framework introduced in the letter takes a rare form in the current AI infrastructure investment wave: the historical pricing weight of the supply side has been temporarily stripped away, and explosive demand-side growth has become the dominant variable. Traditionally, cycle turning points are triggered by signals of supply expansion—high prices → high capex → capacity release → falling prices. What makes this cycle different is that the supply response chain has been stretched extremely long, and every link faces hard physical constraints that cannot be accelerated.
The mismatch in chip-manufacturing capacity build times is central to understanding the bottleneck. A leading-edge fab typically takes three to four years from groundbreaking to volume production, while an advanced memory line (especially the TSV packaging step in HBM) also requires more than two years. This means that even if all capacity investment decisions were launched immediately today, supply elasticity in 2026 would still be technically rigid. More critically, the capex plans of the hyperscalers are themselves being revised upward year after year—if the roughly $700 billion in combined investment by the four giants in 2026 continues to roll into 2027, the existing capacity gap will not converge but may widen further.
This "rolling gap" creates unprecedented narrative space in the capital markets: the visibility of demand growth (AI inference/training compute consumption) coexists with the unverifiability of supply-demand balance (the real utilization of data centers), providing a continuous source of positive feedback for every pro-cyclical valuation model.
Understanding the economic significance of the current bottleneck requires distinguishing between cyclical bottlenecks and structural bottlenecks:
| Comparison Dimension | 2017-2018 Memory Super-Cycle | 2024-2026 AI Bottleneck Cycle |
|---|---|---|
| Driver | Smartphone memory upgrades + server replacements | Rigid demand from AI clusters for HBM / high-capacity NAND |
| Supply response | Capacity expansion effective within 12-18 months | Fab/packaging capacity truly comes online in 3-4 years |
| Demand elasticity | Fragmented downstream, relatively strong substitution | Concentrated hyperscaler procurement, low price elasticity |
| Government intervention | Limited | Deep involvement via industrial policy in the U.S., Japan, EU, and South Korea |
The letter cites DRAM prices rising roughly 650% since 2023, with most of the gain concentrated in the past year—a figure that in itself is the market pricing in the "degree of bottleneck." But caution is warranted: memory chips are a highly cyclical commodity, and 2000, 2010, and 2018 all saw prices collapse after sharp run-ups. The difference from history is that the long-term growth curve of AI demand—at least at the narrative level—has led market participants to believe that this time demand will prove significantly more persistent than in previous cycles. The letter's question of whether this "investment cycle is completely different" is precisely the optimistic excuse the market cannot falsify.
SanDisk's nearly 4,500% gain over the past year is an extreme data point worth pondering. In any historical period of the semiconductor industry, a mature memory company rising 46x in a single year means the market has priced in every possible piece of good news—years of future demand growth, its own capacity expansion, the failure of competitors—all at once, with no discount. Such extreme moves usually point to two possible explanations:
1. The market is assigning option value to a company with a very small market cap as a "key bottleneck gatekeeper," with an implied perpetual growth rate in its share price far exceeding industry fundamentals.
2. Concentrated inflows from leveraged capital and passive index funds are creating a positive-feedback spiral of "buying as it rises, rising as it is bought."
Whichever explanation applies, a 4,500% gain exceeds what fundamental analysis can cover; it is more a phenomenon of market microstructure. The same logic applies to the entire semiconductor sector: semiconductors now account for nearly 20% of the S&P 500's weight—four times the level in 2020—and this weight is higher than the peak of the 2000 tech bubble (at the time, the technology sector was about 34% of the S&P 500, but that included some non-semiconductor names on similar themes; semiconductors alone were less than 10%). This concentration in index weight means index volatility will increasingly be driven by technological progress, geopolitical conflicts, or demand data from a single industry chain rather than by broad economic fundamentals. For passive investors, this is a deep source of fragility.
The letter closes by mentioning Elon Musk's attempt to attract investment while the current boom lasts—a detail worth developing further. Insider retreat during an euphoric phase is often a signal that the capital cycle is entering maturity. Looking back at the 2000 internet bubble, insiders and executives at Cisco, Intel, Microsoft, and others exercised stock options and sold on a massive scale in the first half of 2000 while retail money was still pouring in; in late 2017, several NVIDIA directors reduced their stakes after the stock broke above $200, and the shares subsequently fell more than 50% within ten months in 2018.
When the founder with the greatest information advantage chooses to sell shares at the current valuation—rather than continuing to add capex—it means he believes the market's pricing has exceeded his own estimate of the company's intrinsic value. Meanwhile, the "picks-and-shovels" logic in the capital markets is being extrapolated without limit: spreading from chip design to foundries, to memory manufacturing, and even to disk-drive makers (Western Digital, Seagate). When every link in the bottleneck chain receives the same valuation premium, this "indiscriminate pricing" has in effect departed from the economic meaning of scarcity.
Based on the data and logic in the letter, the fragility of the current AI capital cycle is concentrated in four dimensions:
| Risk Factor | Trigger Mechanism | Affected Links |
|---|---|---|
| Hyperscaler capex reduction | AI model monetization revenue far below expectations; CFO-led capital-return reviews tighten | Chip designers, foundries, memory makers |
| Taiwan Strait geopolitical escalation | A potential conflict disrupts TSMC capacity or forces production relocation | Global tech supply chain; Taiwan GDP growth (current 13.7% annualized growth is highly concentrated in TSMC) |
| Accelerated capacity release from Chinese foundries/memory | Mature-node supply floods the market, price competition re-emerges | Second-tier foundries, conventional memory chips |
| Fed tightening beyond expectations | Rising rates pressure long-dated valuations; high-duration growth stocks suffer | Valuation benchmark across the industry |
Among these, the sustainability of cloud providers' capex is the most critical and the hardest to assess. The $700 billion investment equals 70% of total global oil and natural gas industry capex—an amount far beyond what operations require, and one that depends on continuous debt and equity financing. If the incremental revenue from AI commercialization (API calls, subscriptions, improvements in ad conversion rates) fails to materialize at an accelerating pace over the next four quarters, the market's patience will be severely tested.
But until the bottleneck truly eases or demand is falsified, momentum in the capital markets may itself become a self-fulfilling prophecy: capital flows into semiconductors → share prices rise → companies obtain low-cost capital through equity issuance → capacity investment increases → earnings grow → valuation multiples stay high. This positive-feedback mechanism needs a new negative-feedback variable to be broken—whether that is interest rates staying higher for longer, insiders engaging in sustained large-scale cash-outs reaching a critical point, or an actual cost-saving announcement from one of the leading cloud providers. Until that moment arrives, the market will continue to concentrate capital into this chain and keep expanding its influence on the indices, exactly as the letter depicts.
The most exquisite manifestation of Musk's ability to "bend reality" is not that he pushed SpaceX's or Tesla's share prices to record highs—that is merely the result. The real magic lies in his having persuaded a group of the world's most skeptical institutional investors to accept three TAM calculation frameworks that can hardly be validated by traditional financial logic, and to treat them as a legitimate basis for valuation. As the previous section alleged, SpaceX's group-wide free cash flow in 2025 was approximately -$14 billion, yet its post-IPO market capitalization was set at around $2 trillion. The bridge between the two is not current earnings, not stable growth in a discounted cash flow model, but a series of neatly "packaged" TAM figures.
Aggregating the data SpaceX disclosed in IPO documents and investor communications gives a clear view of the scale of its TAM system:
| Business Line | 2025 Revenue (approx.) | 2025 Operating Result (approx.) | Claimed TAM | TAM/Revenue Multiple |
|---|---|---|---|---|
| Starlink connectivity | $11.5B | EBIT +$4.4B | $1.6T | ≈139x |
| Space (rockets + Starship + Starshield) | $4.0B | EBIT loss | $370B | ≈92x |
| xAI / AI business | $3.2B | EBIT -$6.3B | $26.5T | ≈8,281x |
| Total | ~$18.7B | Operating loss + FCF -$14B | ~$28.47T | ≈1,522x |
The comparison is even more striking: the claimed TAM of $28.47 trillion is 1.4x China's 2025 GDP (approximately $20 trillion) and close to 95% of U.S. GDP (approximately $30 trillion). A company with less than $20 billion in 2025 revenue claims to operate in an addressable market equivalent to the total output of the world's second-largest economy.
In the xAI TAM breakdown, the biggest "leap of faith" falls in the enterprise applications category—SpaceX carves out a $22.7 trillion potential market for it. The basis for this figure is not any software market report, but the "total global digital economy" (including AI, cloud, cybersecurity, IoT, robotics, etc.)—equivalent to treating all digital economic activity on Earth as SpaceX's future capture. This is like a coffee chain treating "global consumption of all liquid beverages" as its TAM, or an automaker counting "total human travel distance" as its potential market. Methodologically, it swaps "market size" for "total customer spending," without considering any product substitution, regulatory barriers, competitive landscape, or customer willingness to pay. In investment parlance, it defines the "T" (Total) of TAM while essentially discarding the "A" (Addressable) and the "M" (Market).
The more critical question is: SpaceX's current AI revenue is only $3.2 billion, corresponding to a 0.01% penetration rate. Even under the most optimistic path—assuming its AI revenue doubles every year, it would reach only $3.3 trillion in revenue a decade from now (itself an unprecedented hyper-growth rate)—its enterprise applications TAM would still leave a gap of nearly $20 trillion to be filled by a market that has yet to be defined. This "gap" is not growth headroom; it is an unverifiable "dream chip" suspended in narrative space.
Starlink's TAM similarly suffers a systematic mismatch between "mathematical feasibility" and "real-world feasibility." SpaceX claims a fixed-internet TAM of $870 billion, built on the core assumption of 1.8 billion households globally × an average ARPU of $31 per household per month. But in reality:
This is less "TAM" than graffiti drawn on the borders of an "imagined market": it is not a set of reachable addresses, but a romantic vision that "everyone needs high-speed connectivity."
To give the above criticism a frame of reference, consider how the big tech companies measure their own "market space":
| Company | Business | FY2025 Revenue (approx.) | TAM or Market Space in Filings/IR | TAM/Revenue Multiple |
|---|---|---|---|---|
| Microsoft | Azure + AI + software | $280B+ (total revenue) | Does not disclose a global tech-spending TAM; only cites the "digital transformation" market as a long-term driver | ≈10x (implied) |
| Alphabet / Google | Cloud + advertising + AI | $380B+ | Does not directly disclose a cloud/AI TAM; market researchers estimate its addressable search/cloud market at ~$1-2T | <5x |
| NVIDIA | AI chips + systems | $150B+ | At GTC 2024, cited the "global data center accelerated computing" market at ~$1T (2030 outlook) | <7x |
| Amazon | AWS + retail | $700B+ | Does not emphasize TAM; uses the "total market opportunity" concept | <5x |
| SpaceX | Space + connectivity + AI | $18.7B | Claims TAM of $28.47T | ≈1,522x |
Even NVIDIA—the biggest beneficiary of the AI era—has not crammed "all global computing" into its TAM simply because GPU demand is exploding. It has not claimed an addressable market of $26 trillion. By comparison, SpaceX's TAM logic diverges sharply from rigorous investor relations practice and looks more like a kind of "quantitative rhetoric."
It must be acknowledged that this TAM narrative is not purely absurd. Soros's "reflexivity" theory finds its most extreme expression here: the market buys Musk's story, the higher valuation brings cheap capital, SpaceX uses that capital to hire top talent, mass-produce rockets, and deploy satellites, thereby physically expanding its real market share. In this sense, the TAM narrative has a kinetic force that "manufactures reality"—it lets the company raise money it could never otherwise have raised, and the capital injection in turn makes some chapters of the TAM story (Starship's progress, Starlink's user growth) no longer purely a dream.
However, the fatal weakness of this mechanism is that valuation has no breathing valve. When SpaceX's market capitalization rests on expectations of an exponential realization of a distant TAM, the company is in effect never required to achieve any "sustainable" balance at its current cash burn rate. A -$14 billion free cash flow in 2025 is not a problem—as long as the TAM is $28 trillion, the market is willing to wait. But the moment the macro environment tightens, the AI capex boom cools, or any major financing event raises doubts about profitability, these two opposing projections—"massive cash burn" and "infinite TAM"—will reverse with equal speed. At that point, the label of "dream stock" will turn from praise into an alarm.
If the TAM is so sloppy, why do institutional investors still accept it? It is worth setting aside easy mockery and going one level deeper.
First, TAM's function is no longer "market capacity" but "narrative legitimacy." When SpaceX tells investors that it is "in the midst of artificial intelligence—the largest economic paradigm shift in human history," rather than "we are a rocket company," its valuation anchor leaps from the space industry itself (a finite market) to the collective imagination of the entire technology sector (an infinite market). The TAM figure is merely the "terminological packaging" of this transition—it gives fund managers an "objective framework" they can present in compliance documents and before investment committees, even if its internal logic is riddled with holes.
Second, Musk possesses a rare "discount rate for reality." Ordinary entrepreneurs discount the future with products; Musk discounts the future with vision. His use of the "reality distortion field" is not about making employees believe he can build rockets, but about making the financial markets believe that "a rocket company can become a digital-economy titan." Jim Chanos's comment—"you can always pick a space, whether it is energy, batteries, robotics, or autonomous driving; and this company just happens to be run by the smartest man in the universe"—captures precisely both the absurdity and the effectiveness of this discounting process.
Third, the real "reality distortion" lies not in the precision of pricing, but in the unfalsifiability of the narrative. When Musk extends SpaceX's AI TAM to $26.5 trillion, that figure cannot, in essence, be toppled—because it rests on the unmeasurable premise that "AI will take over the digital economy in the future," with no ceiling and no anchor. It makes the "current 0.01% penetration rate" look like a conservative assumption rather than an absurd excuse. This is the narrative essence of "bend reality": it transforms an unfillable valuation gap into evidence of "enormous future growth headroom."
This allows us to revisit the opening quote—Sebastian Mallaby's mention that "Hassabis faces the same dilemma"—for in fact the entire AI/tech industry has fallen into a similar narrative contest: every company claims that "its own addressable market is unprecedentedly vast," and the larger the claimed scale, the easier it is to mask immaturity in unit economics. But SpaceX is the furthest and most successful example, because it has not only created a "dream company" valuation, but has turned that valuation into a template that other companies now vie to imitate.
From a financial logic standpoint, SpaceX's $28.47 trillion TAM cannot be absorbed by any serious valuation framework; from a behavioral finance standpoint, it is a carefully constructed "hope ticket," designed to make every skeptic concede that "everything has a possible market." The reality Musk has "bent" through TAM is not the physical laws of the material world, but the conventional law of the capital markets regarding "what constitutes a reasonable valuation."
In the sections that follow, this article will examine another critical question from a deeper level: when the TAM narrative myth is reverse-validated by the market—that is, when earnings falsification occurs or capital enthusiasm fades—in what manner will the boundaries of "bend reality" collapse? At that turning point, SpaceX's year-to-date 2026 data will provide an extremely valuable natural-experiment window.
The brilliance of the SpaceX section lies in how the author uses a set of seemingly simple, in fact brutal reverse calculations to puncture the gap between "dream" and "verifiable reality." The figures in the text are worth further extrapolation:
What does it mean to go from $18 billion to $400 billion? If SpaceX is to achieve roughly $400 billion in revenue by 2030, it would rank among the top five highest-revenue companies in the world—far above ExxonMobil's level (approximately $340 billion in 2024). That means SpaceX would have to leap in just five years from a company the size of a mid-sized semiconductor firm to a scale larger than Apple (approximately $391 billion). In history, no hardware-centric company (automotive, aerospace, energy, or technology—without exception) has ever achieved this speed of transition. Even the most aggressive software companies, such as Salesforce, took more than a decade to grow from $1 billion to $200 billion in revenue. SpaceX's implied compound growth rate (approximately 85% CAGR) is not without precedent in tech history—a few e-commerce companies in the internet bubble era achieved it over short periods—but the common fate of those cases is that they never simultaneously maintained a 30% cash conversion rate at such scale.
The deeper issue lies in the subtlety of the 30% FCF margin assumption. This ratio looks modest at first glance (SaaS companies commonly achieve 40%+), but for a company simultaneously operating rocket launches, satellite manufacturing, and a global ground-station network, it is extremely optimistic. SpaceX's cost structure is highly fixed—while the marginal cost per launch has fallen sharply with Falcon 9 reuse, the R&D on Starship, the commitment to the Mars program, and the construction of the Starship factory all mean capex cannot converge before 2030. In other words, this set of assumptions is not a "conservative baseline," but a "Holy Grail scenario" in which every condition must resonate perfectly at the same time.
Bloomstran's sardonic data point deserves another layer of magnification: the absurdity of a TAM larger than the entire annual revenue of the S&P 500. If that logic is pushed to its extreme—a company whose potential market space is 40% larger than the actual economic output of the entire U.S. stock market—then this is no longer a valuation problem but a narrative-structure problem. Here one can introduce the framework from Robert Shiller's Narrative Economics: when an economic narrative ("infinite space," "AI is all-powerful") itself becomes a contagious virus, the degree of its detachment from underlying facts loses its calibration mechanism. Chanos's joke about "infinite space TAM" hits the nail on the head: an unfalsifiable TAM strips any valuation of its anchor. But being unfalsifiable does not mean unrealizable—it only means that investment decisions have slid entirely from "probabilistic thinking" into an "imagination contest."
The IPO design details revealed in the text are more critical than they appear on the surface. Floating only about 5% of the share capital is essentially a strategy of separating "flow pricing" from "stock valuation"—using extremely thin trading volume and extremely high marginal prices to establish a theoretical value for all existing shares. This is not new at the operational level, but SpaceX has taken the technique to its extreme:
| Comparison Dimension | Typical Traditional IPO | SpaceX Case (~5% Float) |
|---|---|---|
| Float ratio | 15-25% | ~5%, extreme scarcity pricing |
| Index inclusion strategy | Passively included after meeting rules | Pushes for rule changes to actively capture index-linked fund flows |
| Post-listing capital operations | Must wait for refinancing/M&A | Immediately issues $25B in bonds |
| M&A currency | Cash or new share issuance | Stock-for-stock acquisition using highly valued shares (Cursor) |
For Musk, a high share price is itself a strategic asset. Using SpaceX stock to acquire Cursor for $60 billion—the largest startup acquisition in history—amounts to buying real assets with "market imagination" rather than "cash flow." This is a subtle signal: if management itself believes the stock is overvalued, then the rational choice is to use that overvalued stock to exchange for real, cash-generating assets. It inevitably recalls the 2021 SPAC boom, when many tech companies acquired unlisted assets with high-priced shares, in essence "monetizing the premium."
Equally noteworthy is the detail of the Nasdaq rule change. A company that has not yet satisfied traditional listing conditions (or time requirements) pushing for index rules to make way for it is itself a case of a market institution being captured by a giant enterprise (regulatory capture). And the quote the author cites at the end of the text—"bull markets pay a premium for promises, bear markets discount reality"—precisely reveals the nature of this window: it is a time window based on sentiment, and a natural property of sentiment windows is that they usually close much faster than they open.
BIS places the current AI capital cycle alongside the canal mania, the railway mania, the electrification frenzy, and the internet bubble—a comparative framework that itself contains deep historical logic. The statistic that has never changed: in every mania, the infrastructure was indeed built, but most of the investors' capital was ultimately wiped out.
| Historical Cycle | Core Infrastructure | Assets Retained After the Bubble Burst | Did the Revolutionary Use Hold? |
|---|---|---|---|
| Canal mania (1830s) | Canal network | Some canals still operate today, but most companies went bankrupt | Yes (transport costs collapsed) |
| British railway mania (1840s) | National railway network | Rail network preserved, but most railway companies merged/bankrupted | Yes (mobility revolution) |
| Electrification frenzy (1920s) | Power infrastructure | Grids consolidated into a handful of companies | Yes (foundation of modern industry) |
| Internet bubble (1990s) | Fiber backbone | Large amounts of "dark fiber" sat idle, later absorbed by cloud computing | Yes (birth of the digital economy) |
The core lesson of this historical pattern is: the revolution is real, but capital is not therefore spared from destruction. The reason AI carries greater destructive intensity than previous revolutions is the extreme concentration of its infrastructure investment—the combined capex of just four giants, Alphabet, Microsoft, Meta, and Amazon, already exceeds the GDP of most small countries. The risks that were spread across thousands of listed companies in the bandwidth era are now highly concentrated on the balance sheets of fewer than ten super-majors. Alphabet's $85 billion capital raise (the largest in history) and Oracle's surging debt both show that even inside the giants, internally generated cash flow can no longer keep pace with the rate at which investment devours capital.
There is an important institutional paradox here: BIS—as the "central bank of central banks"—explicitly warned in its official annual report that this cycle could be amplified by financial fragility. The weight of this signal should not be underestimated. The last time BIS issued a comparable systemic warning was in 2007, when the market likewise paid no heed, until the financial system began to collapse. BIS's stance is not anti-technology but anti-leverage—when super-corporations begin financing long-cycle, high-uncertainty infrastructure investment with debt and equity dilution, financial fragility shifts from "technology risk" to "systemic risk." This is the deeper meaning of the text's "financial fragility as an amplifier of the cycle": when Oracle's share price posted its worst week since 2001 on AI financing concerns, the market was already using the rearview mirror to scrutinize the solvency of this cycle.
Tim Cook's "once in a century" remark raises a critical question: why is even Apple—the buyer with the strongest bargaining power in the global supply chain—not exempt from memory chip price increases? This illustrates the nature of the current shortage: it is not a localized, sudden event (such as an earthquake halting one fab), but a systemic reallocation of capacity triggered by the "siphoning effect" of AI data centers on HBM and DRAM. AI data centers are willing to pay prices for HBM that far exceed the premium consumer electronics can absorb, so fabs allocate limited capacity first to the most profitable products, squeezing capacity for conventional DRAM and lower-cost memory and pushing prices higher.
This "the rich get richer" allocation mechanism for raw materials is, in essence, creating a new class divide atop the core of the digital economy—memory. Handset and game-console makers are "passive takers with no alternative supply"; the challenge they face is not a choice but the dilemma of "price increases or shortages." The collective price increases already announced by Samsung, Xiaomi, Nintendo/Sony/Microsoft, and Apple provide the answer: the conversion of supply-side bottlenecks into consumer prices is a transmission chain without obstacles.
But when judging the cycle, one must remain cautious about elasticity. After the 2018 DRAM super-cycle peak, the industry experienced a price collapse of roughly 30%, driven by the combination of demand destruction and pull-in orders triggered by high prices. Whether history repeats now depends on the struggle between two variables: first, whether the structural demand increment from AI can absorb the gap left by shrinking consumer demand; second, when the expansion capacity of the various manufacturers is released in concentration. The iron law of the chip industry over the past fifty years has never changed: high prices are both a reflection of scarcity and a catalyst for future oversupply.
Meta's announcement that it would lease idle data center compute capacity to third parties triggered a plunge in memory stocks because it fundamentally shook the market's underlying belief in the "self-absorption" of AI capex. Previously, hyperscaler capex was viewed as a rigid commitment, and investors took it for granted that it would necessarily translate into continued procurement of memory, GPUs, and other hardware. Meta's move is tantamount to publicly admitting that its own AI compute capacity is redundant and that it needs to monetize it externally to improve utilization. This directly caused SK hynix, Samsung Electronics, and Micron to fall as much as 15% intraday—a single-day market-capitalization wipeout approaching the most extreme trading day of the 2022 semiconductor downturn (when Micron fell about 12% in a single day). More concerning is that this is not an isolated case: in Q4 2025, Microsoft had already begun reselling some Azure cloud compute to startups, and Google also adjusted the usage priority of its edge data centers. Taken together, these moves suggest that the marginal return on AI infrastructure investment may have already passed an inflection point—a variable not yet priced into the market's valuation models.
| Event | Date | Largest Single-Day Drop in Memory Stocks | Market Interpretation | Capex Revision Over Subsequent 12 Months (Historical Reference) |
|---|---|---|---|---|
| Meta compute-rental announcement | July 2026 | 15% (SK hynix, Samsung, Micron) | Demand expectations deteriorate; capital cycle peaks | To be observed; analogous to 2000 when Cisco's capex growth collapsed from 62% to -15% |
| 2000 network equipment demand warning (Cisco order slowdown) | March 2000 | Jabil and other EMS stocks fell 20% in a day | Internet infrastructure investment excessive | Global telecom equipment capex fell 25% |
Current inventory levels at AI hardware makers are highly similar to Cisco's network equipment inventory in 2000—the latter had accumulated roughly 20 weeks of safety stock before the bubble burst, while as of Q2 2026, major memory makers' DRAM inventory has already recovered to 12-14 weeks, nearly twice the level at the 2023 cyclical trough. If Meta's compute-rental model is imitated by other cloud providers, it will directly puncture the illusion of "double procurement" of AI chips and memory (i.e., stocking up for both training and inference at the same time)—exposing demand-side fragility earlier and more truthfully than revenue cyclicality.
BIS lists interest rate risk as one of the final two pressure points, and its logic chain is worth unpacking: blockade of the Strait of Hormuz and damage to Middle East energy infrastructure → supply-chain cost shock and potential second-round inflation → even if oil prices retreat, fiscal deficit monetization will delay the decline in inflation → the long-term rate center shifts upward → discount factors rise for high-valuation growth stocks + corporate financing costs increase → AI investment plans are cut or deferred. This creates a paradox: AI is widely regarded as the largest capital cycle of the next decade, but the continuation of that cycle depends precisely on extremely low long-term interest rates—because most AI project cash flows are concentrated in the distant future, with extremely long duration and high sensitivity to rate changes.
Viewed through data like a Damocles' sword: in Q1 2026, the real yield on the U.S. 30-year Treasury was about 90 basis points above its 2024 low. If it rises another 50 basis points, Goldman Sachs estimates the fair value of unprofitable AI companies in the S&P 500 would be compressed by an additional 18-22% on average. More critically, global private credit has now surpassed $2.3 trillion, with technology and software loans accounting for 28% of the total, and the collateral for these loans is mostly intellectual property and recurring revenue—so once rates jump due to a Middle East conflict or fiscal disorder, default rates will deteriorate non-linearly rather than linearly. This is essentially the same fragility as the internet companies in 2000 that depended on venture debt to survive, except that the leverage has shifted from the equity market to the credit market.
Also worth noting is the "spiral feedback" between fiscal deficits and interest rates. The BIS report points out that structural deficits in major economies are themselves an inflationary force—governments must continuously issue new debt, pushing up term premia and squeezing the availability of capital to the private sector. There is a comparability with the post-war period: from 1945 to 1947, the U.S. deficit ratio fell from 21% to 4%, while long bond yields rose from 2% to 2.7%, but at that time post-war reconstruction demand supported growth. Today, if the deficit ratio stays at 6-7%, combined with the massive financing required for AI investment, the upward move in rates could far exceed the "benign path" markets imagine. This will ultimately cause AI companies valued at more than 25x forward revenue to be the first cut down by the market's knife of "risk-free rate hikes."
Facing the above macro risks, Horos's portfolio adjustments this quarter clearly embody the principle that "margin of safety trumps everything." Its operations can be divided into three categories:
1. Taking Profits and Exiting: Converting the Cyclical Rebound into Cash
The logic behind exiting Acerinox and Aperam is clear—the European stainless steel industry is indeed recovering, and the European Commission's protectionist measures are intensifying, but the shares had already risen 70-100% in a short period, and the market has fully priced in the earnings-recovery expectation. This is the disciplined execution of "buy low, sell high": once the judgment that the industry cycle had bottomed was realized, the margin of safety disappeared with it. Likewise, ending the Talgo position is an admission of falsification—the transaction being blocked by regulators, Avril's train quality problems, and Renfe's fines, three negative factors stacked together, have completely eroded the original investment narrative. The fund reduced its position in Sopra Steria, which rebounded nearly 40% from its April low, but did not fully exit—indicating that it retains exposure to the long-term demand for European IT services while converting some chips back into cash.
2. Contrarian Reallocation: Embracing Undervalued Assets with Structural Advantages
The new position in Suzano is currently the most instructive move to interpret. The several reasons for its depressed share price (heavy capex, falling pulp prices, China's economic slowdown, and rising self-sufficiency rates for woodchips in China) are all cyclical or sentiment-driven, not the loss of a structural competitive advantage. As the world's largest pulp producer by capacity, Suzano sits in the lowest percentile of the industry unit cash cost curve; during a pulp price trough (currently around $700/ton, below the industry average cash cost), it can instead use a low-price strategy to clear out high-cost capacity and gain market share. More importantly, the investment logic has nothing to do with the AI bubble—it is a physical commodity at historically low valuations (enterprise value per ton of capacity is about 0.8x the industry average), and family control reduces agency costs. This forms a natural hedge against the high-valuation components of AI.
3. Adding to "Opaque Discount" Assets: The Synergy Logic of Bolloré, Odet, and Antin
Bolloré and Compagnie de l'Odet are classic holding-company discount trades. Bolloré holds a large portfolio of listed assets (especially its investment in Italian media group Mediaset and private equity platforms), yet its shares have long traded at a 30-40% discount to net asset value. Compagnie de l'Odet is the controlling entity of Bolloré, with two layers of leverage in the structure. In a rising-rate environment, such discounted assets offer an "invisible margin of safety"—even if the value of the underlying assets does not change, the convergence of the discount itself can generate returns. Antin Infrastructure, as an infrastructure asset manager, has a business model (management fees + performance revenue) whose negative correlation with long-term rate levels is far lower than that of AI hardware or software stocks, because infrastructure assets often have inflation-linked revenue provisions, and when rates rise, newly formed funds are typically able to pass higher capital costs on to new investment projects. These three additions point to a common purpose: shifting the portfolio's exposure from "rate-sensitive, high-valuation growth" toward "rate-insensitive, discounted value."
Particularly noteworthy is that Horos did not fully liquidate TGS; it merely reduced the position, keeping a 3.8% stake—because the team believes that under the energy-security theme, demand for seismic data services has "counter-cyclical" properties (an escalation of Middle East conflicts actually increases exploration demand), which is fundamentally different from AI-related semiconductor stocks. Similarly, the logic for adding to Expro Group lies in the high-margin product line brought by the Enhanced Drilling acquisition, not in reliance on macro liquidity.
Taken together, this series of operations displays a typical "defensive portfolio reconstruction": raising cash positions (through profit-taking), adding low-correlation real assets and discounted holding companies, and reducing indirect exposure to AI end-demand, thereby cushioning against the dual risks of "an interest rate shock + a downward revision in AI earnings expectations." This is not a denial of AI's revolutionary significance, but an acknowledgment that in an environment of such extreme valuations and such uncertain monetary conditions, preserving an adequate margin of safety is the key to navigating the cycle.
The cross-shareholding between Bolloré and Compagnie de l'Odet is not a simple two-way investment but an exquisite form of control leverage. According to the follow-up data, Odet controls roughly 73% of Bolloré, while Bolloré in turn holds about 84% of Odet—forming a closed loop. In economic substance, this is equivalent to Bolloré holding about 61% of itself (73% × 84%), and these shares are either eliminated or shown as minority interests in the consolidated statements, producing two consequences:
This structure makes the "control premium" quantifiable: the Bolloré family leverages more than 70% control with very little personal capital (estimated direct ownership of only about 5-10%), at the cost of external shareholders bearing an additional valuation discount. Historical data show that such two-tier holding structures are common in Europe (e.g., Italy's Exor, Sweden's Investor AB), but Bolloré's closed-loop cross-shareholding is its unique defense mechanism—it ensures that even if a hostile acquirer bought up all the free float, it could not gain control, because the voting rights of the Bolloré shares held by Odet are in effect controlled by the family.
The follow-up's mention that "Odet is the main recipient of the extraordinary dividend recently distributed by Bolloré" reveals a frequently overlooked cash flow logic. As an operating holding company, Bolloré's special dividend to shareholders does not flow evenly to family and public shareholders; it is designed so that most of it flows to Odet. This leads to:
Horos Value Iberia reduced stainless steel (Aperam, Acerinox) and exited Talgo, while adding to Alquiber and establishing a new position in Línea Directa. This is not merely portfolio rebalancing; it also implies a judgment about the macro-economic cycle:
| Industry | Reduction / Addition | Core Logic | Cycle Stage |
|---|---|---|---|
| Stainless steel (Aperam/Acerinox) | Reduced | Weak European manufacturing, soft auto and construction demand, nickel price volatility squeezing margins | Cyclical downturn |
| Rail equipment (Talgo) | Exited | Uncertain order conversion, political intervention lowering returns | Structural risk |
| Flexible rental (Alquiber) | Added | During corporate capex contraction, rental penetration rises; the company has counter-cyclical properties | Counter-cyclical growth |
| Direct insurance (Línea Directa) | New position | Inflation peak passed; auto insurance rate increases lag claims costs; strong profit recovery elasticity | Bottom reversal of the cycle |
The difficulties of the stainless steel industry are nothing new, but Acerinox and Aperam look cheap on the surface (P/B both below 1.5x). Yet because of the European carbon border tax and rising energy costs, their replacement costs are actually below book value, creating "value trap" risk. By contrast, Línea Directa's average ROE exceeds 25%, and its share of the Spanish direct auto insurance market is above 30%—a position that allows it to maintain a 5-8 percentage point cost advantage even when the industry-wide combined ratio approaches 100%. The author estimates that if industry rates rise 10-15% over the next two years (driven by parts inflation and higher accident frequency), Línea Directa's underwriting margin could recover from the current 3% to above 6%, corresponding to at least 30% upside in the share price.
Alquiber is a typical case: the Acebes family controls about 76%, the free float is only 24%, and average daily turnover is often below €500,000. This liquidity barrier has kept its share price persistently below intrinsic value—even though its ROIC has exceeded 20% for five consecutive years, with continuous growth and stable dividends, the market still applies a discount of about 30%. For large funds, this type of investment cannot be built into a position; but for Horos Value Iberia (which is small), it instead offers an opportunity to buy a high-quality asset at a reasonable price. The author notes that Alquiber's net debt/EBITDA is only 1.8x, while its fleet utilization remains above 90%, with expected annual expansion of 8-10% over the next three years—enough to support a target of 12% compound annual growth in earnings per share. The existence of the liquidity discount reflects market structure rather than fundamental deterioration, which is precisely the typical prey of a value investor.
Cirsa is another example of a "quality asset suppressed by a majority shareholder's stake." Blackstone controls about 75%, leaving an extremely small free float and little institutional coverage after the IPO. Yet the key to its business model is the scarcity of regulatory franchises: in Spain, Italy, Panama, and elsewhere, gaming license issuance is restricted, and existing operators enjoy barriers to entry. Cirsa's EBITDA margin has remained in the 28-32% range over the long term, and its M&A integration capability has been proven—small acquisitions are typically lifted to the group's average return within 18 months. The author estimates its free cash flow yield per share at close to 8%, while its current EV/EBITDA is only 7.5x, significantly below the 10-12x of European gaming peers such as Flutter and Entain. Blackstone's stake is not a permanent suppressing factor—once it achieves its exit objectives, a valuation re-rating is possible. In addition, the market worries about regulatory tightening, but Cirsa has already expanded its online gaming business into Mexico and Colombia, and this diversification reduces dependence on any single market.
Horos Patrimonio's rebalancing reflects the complementarity between equities and bonds. Selling the Vermilion Energy 2030 and Cirsa 2029 bonds and buying Fertiberia, Serica, and Performance Shipping shows a shift from "recession defense" to "reflation hedge":
| Bond | Initial Yield (EUR hedged) | Credit Rating | Key Drivers | Risk Points |
|---|---|---|---|---|
| Fertiberia 2031 | 7.15% | Unrated (BB- equivalent) | Specialty fertilizer share rising to 57%, EBITDA €89m, debt/EBITDA 2.5x → expected to fall to 1.5x | Natural gas price volatility, weak agricultural demand |
| Serica Energy 2031 | 5.9% | BB- | 10% share of UK natural gas production, 2P reserves of 138 Mboe, expected net cash in 2026 | Crude/gas price collapse, production disruption |
| Performance Shipping | Above 8% | Unrated | Aging tanker fleet capacity, strong spot market | Freight volatility, single-vessel-type risk |
Fertiberia's highlight lies in the huge gap between its asset replacement value and book value: the replacement cost of its production facilities is about €1 billion, while book value is only €360 million—meaning that even under extreme operating difficulty, the liquidation value of its assets provides sufficient cushion for the debt. More importantly, compared with the 2022 peak, the 2024-2025 profit decline already fully reflects price normalization, and once the current new capacity comes on stream, unit costs are expected to fall 10-15%, pushing 2027 EBITDA back to €120 million. A 7.15% yield corresponds to only about a 3.8x EV/EBITDA multiple on 2026 figures—buying near the bottom of the cycle offers an attractive risk-reward profile.
Serica's bond offers a "natural gas undervaluation" view: its assets are entirely on the UK continental shelf, UK natural gas prices have long been above the European benchmark, and the country's production will decline over the next decade, giving existing reserves a lifespan of about 7-8 years. The bond terms allow the company to use its liquidity for acquisitions; if successful, bondholders would benefit from credit improvement. The 5.9% EUR hedged yield is slightly lower than the 7% average for commodity bonds, but given its expected ending net cash, the credit spread more than compensates for the risk.
Placing Bolloré/Odet and the fund rebalancing in the same framework reveals a common logical thread: seeking assets that the market misprices for structural reasons (control design, insufficient liquidity, a large shareholder awaiting exit), while their fundamentals are undergoing positive change. Bolloré releases value by simplifying its structure, and the funds wait for cyclical inflections through rebalancing. Historical data show that discounts on European family-controlled holding companies typically converge 20-40% after structural changes, while small value funds can earn annual excess returns of 4-6% by exploiting liquidity disadvantages. At present, Bolloré is in the re-evaluation phase after the spin-off, and Odet has gained an additional safety cushion from the special dividend and cross-shareholding replenishment—this is precisely the window in which patient capital can harvest returns.
The following supplements the quantitative analysis of the PERSHI investment logic from the perspective of a bondholder, and compares the risk-return characteristics of comparable assets. The focus is on the financial flexibility, contract structure details, and relative yield positioning that the original text has not yet developed.
The original text notes that PERSHI recently disposed of its oldest vessels, leaving the active fleet at an average age of roughly 7–8 years. The deeper implication is that with the older ships gone, dry-docking costs, special survey expenses, and insurance premiums should decline meaningfully over the next two to three years. For medium-range (MR) tankers, annual maintenance costs for vessels over 10 years old typically run 15%–20% higher than for ships aged 5–8 years. Once a vessel exceeds 12 years of age, classification society special surveys double in frequency, which directly affects operating cash flow.
When the three new MR newbuildings are delivered in 2027–2029, the fleet's average age will decline further to roughly 5–6 years. The new vessels also feature lower fuel consumption and better energy-efficiency metrics (EEDI/EEXI), allowing them to maintain a lower break-even point during weak freight markets. This means even if the contract coverage ratio falls below 80% in 2028, PERSHI's break-even level is expected to be 10%–15% below the industry average, thereby reducing bond default risk.
Repsol, as the charterer, carries an investment-grade credit rating (Moody's Baa1 / S&P A-), far above that of the typical independent tanker owner. Two of the three newbuildings are contracted to Repsol under multi-year charters, effectively creating a "quality-credit charter cash flow → passed back to bondholders" transmission chain. In an asset-backed context, this approaches the safety level of a finance lease.
More critically, charters with oil majors like Repsol typically include minimum operational efficiency clauses (such as speed and fuel-consumption guarantees). This means if PERSHI's newbuildings fail to meet technical performance standards, the company could be required to pay compensation or renegotiate charter rates. Conversely, this clause also forces PERSHI to maintain high operating and maintenance standards for its new vessels, indirectly reducing the risk of asset value impairment.
The original text emphasizes that capital expenditure will drop sharply as newbuildings are delivered, but it does not quantify the decline. Using the company's existing 14 vessels (including those under construction) as a baseline and assuming a typical MR newbuilding cost of approximately $45 million per vessel:
Assuming the company's 2026 EBITDA is roughly 30%–35% of revenue and incorporating a 90% contract coverage ratio, a conservative estimate puts EBITDA interest coverage at 4–5 times. By 2028, when coverage falls to roughly 75%–80%, even a 20% decline in freight rates would likely leave interest coverage above 3 times. This provides bondholders with a substantial cushion in a scenario where both oil prices and freight rates weaken simultaneously.
Currently, euro-denominated high-yield bonds (BB/B rated) offer average yields to maturity of about 5%–6%, while dollar-denominated high-yield bonds issued by tanker companies generally yield between 6.5% and 8.5%. PERSHI's 7.1% estimated euro yield actually sits between European high-yield bonds and shipping high-yield dollar bonds. Given that its contract coverage over the next three years is close to 80%–90%, this yield level represents an opportunity with notably generous risk-premium compensation.
That said, it should be noted that 7.1% is the "initial estimated euro yield." If the bond's principal and coupons are dollar-denominated, currency fluctuations could erode returns. Based on current euro/dollar hedging costs (roughly 0.3%–0.5% per annum), the net euro yield after hedging remains around 6.6%–6.8%, still above the average for European corporate bonds of the same rating.
| Comparison Asset | Reference Yield | Core Risk | Contract Coverage |
|---|---|---|---|
| European BB/B corporate bonds | 5.0%–6.0% | Credit risk; no specific asset backing | Not applicable |
| Shipping high-yield USD bonds (Euronav/Frontline) | 6.5%–7.5% | Freight-rate cyclicality; higher leverage | Typically 40%–60% |
| PERSHI bond (estimated euro) | 7.1% | Newbuilding delivery delays; tanker rate volatility | 2026: ~90%; 2027: ~80% |
The three newbuildings are scheduled for delivery in 2027–2029. If the shipyard delays, this would not only push back the start of contracted revenue but could also trigger the restrictions on the use of asset sale proceeds in the existing bond covenants, potentially forcing PERSHI to refinance at a higher cost before the bonds mature. Buying this bond therefore amounts to implicitly selling a "tail-risk option" — although the delayed-compensation clauses (such as default penalties) that the company may receive offer partial coverage, a shipyard bankruptcy (e.g., among mid-sized South Korean yards) could lead to consequences similar to an order cancellation.
The author recommends monitoring the "force majeure clause" and "cross-default risk" in the newbuilding contracts. If PERSHI defaults on other debt obligations, this bond could be subject to accelerated repayment. However, given that PERSHI transformed from a single-hull tanker operator and has no history of debt default, this is a point in its favor.
As the seventh investment, PERSHI's role in the portfolio is not to chase high growth but to provide a high-probability fixed-income cash flow. Its high contract coverage ratio of 90%/80% has a low correlation with the portfolio's other cyclical equities or high-yield bonds. In addition, the phase mismatch between the tanker market and the dry-bulk and container shipping markets reduces overall portfolio volatility. Viewed as the "final piece of the puzzle," PERSHI's inclusion significantly enhances the portfolio's cash-recovery certainty during periods of macroeconomic uncertainty.
Summary: PERSHI's bond investment is not a simple high-yield play, but rather a quasi-fixed-income asset built on threefold logic: fleet renewal, blue-chip charters, and declining capital expenditure. The 7.1% euro-denominated yield is clearly attractive relative to comparable assets, while contract coverage, as a hard metric, provides a strong margin of safety. Going forward, key monitoring should focus on newbuilding delivery progress and the execution of the Repsol charter.
| Instrument | Direction | Author's Stance (One Sentence) | Key Data |
|---|---|---|---|
| SpaceX | Not specified | Author criticizes its TAM narrative and free cash flow deficit, viewing it as a specimen of market mania | Claims a TAM of $28.47 trillion, 2025 FCF of approximately -$14 billion, and an IPO market cap of about $2 trillion |
| SanDisk | Not specified | Author regards the 4,500% one-year gain as an extreme bubble signal beyond the bounds of fundamentals | Gain of approximately 4,500% over the past year |
| Suzano | Added (original: "added"; not specified as a new position) | Buy added in the international strategy | — |
| Bolloré | Added (original: "added"; not specified as a new position) | Buy added in the international strategy | — |
| Compagnie de l'Odet | Added (original: "added"; not specified as a new position) | Buy added in the international strategy | — |
| Acerinox | Exited | Exited from the international strategy (original: "exited") | — |
| Talgo | Exited | Exited from both the international and Iberian strategies (original: "exited/divested") | — |
| Cirsa (equity) | Added (original: "invested in"; not specified as a new position) | Buy added in the Iberian strategy | — |
| Línea Directa | Added (original: "invested in"; not specified as a new position) | Buy added in the Iberian strategy | — |
| Vermilion Energy (bonds) | Exited | Capital-preservation strategy exited the bonds (original: "exit from the bonds") | — |
| Cirsa (bonds) | Exited | Capital-preservation strategy exited the bonds (original: "exit from the bonds") | — |
| Fertiberia (bonds) | Added (original: "additions"; not specified as a new position) | Bonds added in the capital-preservation strategy | — |
| Serica Energy (bonds) | Added (original: "additions"; not specified as a new position) | Bonds added in the capital-preservation strategy | — |
| Performance Shipping (bonds) | Added (original: "additions"; not specified as a new position) | Bonds added in the capital-preservation strategy | — |
| TSMC | Not specified | Risk matrix mentions that a Taiwan Strait conflict could disrupt production capacity | Taiwan's annualized GDP growth of 13.7% is highly concentrated in TSMC |
| NVIDIA | Not specified | Positive example of conservative framing in the TAM comparison | FY2025 revenue of $150B+; TAM of ~$1 trillion (2030 outlook) |
| Microsoft | Not specified | Example of insider selling in historical bubbles; TAM comparison | FY2025 revenue of $280B+ |
| Alphabet / Google | Not specified | Does not emphasize an exaggerated TAM in the TAM comparison | FY2025 revenue of $380B+ |
| Amazon | Not specified | Uses the "total market opportunity" concept in the TAM comparison | FY2025 revenue of $700B+ |
| Apple | Not specified | Compares revenue scale in the thousand-fold narrative | 2024 revenue of approximately $391 billion |
| ExxonMobil | Not specified | Compares revenue scale in the thousand-fold narrative | 2024 revenue of approximately $340 billion |
| Salesforce | Not specified | Compares growth duration in the thousand-fold narrative | Took more than a decade to grow revenue from $1 billion to $20 billion |
| Cisco | Not specified | Example of large-scale insider selling in the 2000 bubble | — |
| Intel | Not specified | Example of large-scale insider selling in the 2000 bubble | — |
| Western Digital | Not specified | One of the extrapolation targets of the "picks-and-shovels" logic | — |
| Seagate | Not specified | One of the extrapolation targets of the "picks-and-shovels" logic | — |
| Toshiba | Not specified | Example of an unexpected supply contraction in the 2017–2018 DRAM cycle | — |
| Micron | Not specified | Example of an unexpected supply contraction in the 2017–2018 DRAM cycle | — |