This piece explains how investor Alex Sacerdote uses a 'technology S-curve' framework to find AI opportunities. He argues AI is turning hardware from a commodity into a premium product—each server component now has pricing power due to surging demand. Key holdings: Anthropic (an AI model company he invested in, revenue went from $100M to $9B, excels at coding), Celestica (makes AI servers costing $200k-$300k each, like airplane parts), and Corning (fiber optics; one Microsoft data center uses enough fiber to circle Earth 4.5 times). He warns traditional software firms (e.g., Salesforce) face disruption, as he's shifted from heavy ownership to shorting them.
Alex Sacerdote, founder of Whale Rock Capital Management, manages over $17 billion in assets, with a hedge fund annualized return of approximately 44% over the past three years. He invests using a technology S-curve framework, with the core thesis of identifying technology S-curves, durable competit
Alex Sacerdote, founder of Whale Rock Capital Management, manages over $17 billion in assets, with a three-year annualized return of approximately 44% for his hedge fund. The main theme of this issue: analyzing AI full-stack investment opportunities through the technology S-curve framework, spanning from chips to models to application layers. The most impactful judgment in the entire piece: AI is shifting the hardware industry from "commoditization" to "de-commoditization" — every server component is gaining pricing power and margin improvement due to surging demand and technological upgrades. This stands in stark contrast to the past 40 years of zero growth and zero innovation in hardware.
Alex Sacerdote believes Anthropic is the highest-conviction position, with the core thesis built on three pillars: the explosion of the code market, differentiation in enterprise positioning, and the scale effect of "escape velocity."
Discovery and Investment Process: After ChatGPT's launch in November 2022, Whale Rock immediately initiated a firm-wide deep-dive research effort. The initial judgment was to "invest in chips and infrastructure first," as compute demand was certain regardless of which layer won. Over the following two to three years, nearly all of the 60+ foundation model companies disappeared, leaving only Anthropic, OpenAI, and Google Gemini—similar to the cloud market's evolution into a triopoly of AWS, Azure, and GCP. The key inflection point was the breakthrough in coding capabilities: early models could only write a snippet of code, whereas Claude Code can now run entirely autonomously. Two top-tier programmers, Andrej Karpathy and Linus Torvalds, have "fully switched"—Karpathy noted that last year the tool could write 20% of code while 80% required manual writing, but after the latest model was released, he "hasn't written a single line of code except in English."
Market Size Estimate: Anthropic insiders once spent $100 per day on tokens, annualizing to $20,000–$30,000. With 20 million programmers globally, the code market alone reaches $500 billion. Whale Rock invested in August 2025 at a $18 billion valuation, when the company's revenue growth from $100 million to $9 billion was "unprecedented."
Competitive Moat: Unlike cloud services (commoditized servers and storage), AI models exhibit substantial differentiation—Anthropic excels in private equity and finance, while Google excels in PDF parsing. Additionally, Anthropic is building a complete product ecosystem around its API (SDK, Claude for Work, orchestration layer, tool sets), similar to how AWS established lock-in early on by continuously launching new products. The team has near-zero turnover, high management quality, and strong code quality.
> Readers should note: Sacerdote holds a position in Anthropic, and his argument reflects a holder's perspective. He acknowledges risks including: model improvement slowdown potentially allowing open-source to catch up, changes in the competitive landscape, and regulatory uncertainty.
Alex Sacerdote’s framework consists of three elements: the technology S-curve + a durable competitive advantage + underestimated earnings potential. The core is to identify the intersection of "exponential unit growth" and "exponential earnings growth."
Trigger conditions for the S-curve: Every technology goes through a long, flat period (the first 10 years for smartphones before the iPhone, the first 20 years for the internet before Netscape, and AI hidden for years before ChatGPT) until all adoption barriers are removed — prices fall to a tipping point, infrastructure is in place, and usage becomes simple. This triggers a "demand tornado."
Buying timing: Andy Grove said, "You cannot trust data at a strategic inflection point"; rely on intuition and anecdotal evidence. Sacerdote uses his "right brain" to observe — seeing children in China playing high-quality games on large-screen phones signaled the arrival of the mobile gaming S-curve; at the Gartner IT Summit, seeing AWS’s presentation hall packed from 9:00 to 11:00 AM indicated surging enterprise demand. Key point: It’s okay to arrive late; missing the first 100% doesn’t matter — if the top of the S-curve is $500 billion, growth can persist for a long time.
Selling signals: When penetration reaches 30%-40%, exponential growth stops, sell-side expectations catch up, and there are no more "big beats." Sacerdote admits he made a mistake with Apple — selling in 2012 when smartphone penetration hit 50%, but Apple continued to compound at 20% through low valuation and ancillary services (App Store 30% commission).
The "height" of the S-curve determines the holding period: AWS’s TAM was initially underestimated — thought to be $600 billion in IT spending with 50% deflation, but later it was found that self-build costs were similar, making the TAM larger. The EV S-curve hit a "big wall" at 10-15% penetration and did not complete its trajectory as expected — requiring continuous tracking and adjustment.
Screening for competitive dynamics: First find the S-curve, then look for players with strong competitive advantages. Moats in the digital world can be stronger than offline — network effects (LinkedIn, Facebook), industry standards (Oracle, Bloomberg), platform effects (AWS), key IP (Qualcomm, ASML), and brand (Google, Amazon without needing ads). But without a moat, even the best S-curve will lead to losses — the list of RIM, Nokia, Motorola, and others is long.
Alex Sacerdote argues that AI is triggering a "hardware renaissance" — an industry that saw zero growth over the past 40 years is now being completely de-commoditized by surging AI demand, granting every component pricing power and margin expansion.
Historical Comparison: Over the past 40 years, data centers have barely changed. Intel x86 became the standard, compute loads grew 25-40% annually, but Moore's Law improved in lockstep, requiring no major innovation. Hardware was fully commoditized — from chips to memory to chassis to networking — with upgrade cycles as long as 7 years (1G to 10G), and rapid commoditization after each innovation.
Changes Brought by AI: AI workloads are growing 10x per year, pushing every hardware component to its physical limits. This creates two effects: 1) massive unit growth; 2) de-commoditization — every component requires continuous innovation, and it is no longer a case of "whoever is cheapest wins."
Specific Examples:
Market Size: The cloud market is approximately $800 billion, while the AI market could reach $3-5 trillion. Current AI infrastructure penetration is only 10 basis points (0.1%), and Sundar Pichai noted that only 10 bps of knowledge workers truly use AI. Sacerdote calls this the "L-curve" — straight up.
> Readers should note: Sacerdote acknowledges the risk in the "hardware renaissance" narrative — if model improvements slow down, open-source catch-up could lead to a "race to the bottom," but chip companies "don't care who wins" (Jensen Huang hopes open source succeeds). Additionally, all components are currently in shortage, which would create a good cycle even under commoditization — but shortages will eventually ease.
Alex Sacerdote argues that AI is disrupting traditional enterprise software. Whale Rock has shifted from a 40-50% software allocation five years ago to a net short position, driven by AI products failing to prove value, budgets being consumed by AI, and the threat of "AI-native" replacements.
The Transition: In April 2023, Whale Rock initially believed software companies could build products using AI APIs and their own data, which was seen as a positive. However, it quickly became apparent that their AI products were "not good enough," "could not be monetized," and "did not drive performance." The firm then sold off nearly all its software positions, turned net short in early 2026, and realized significant gains in Q1.
Four Sources of Pressure:
1. Budget Competition: On CIO priority lists, AI (Anthropic tokens) offers faster ROI, crowding out traditional software budgets.
2. Disappearing Pricing Power: Software companies used to raise prices annually but now face competitive and client pressure that prevents them from doing so.
3. Hiring Freezes: Some companies have significantly cut headcount or frozen hiring, reducing demand for per-seat software licenses.
4. AI-Native Replacements: Within 1-5 years, entirely new AI-native companies could emerge, systematically challenging incumbents. If data advantages are neutralized, replacement could become easy.
"AI Version of the Rule of 40": The traditional Rule of 40 = growth rate + profit margin ≥ 40. Sacerdote's new Rule of 40 = AI revenue share × AI category market share. For example, if a company has 30% AI revenue and 30% market share, it scores 60 — worth watching. The problem is that traditional software companies' AI revenue is only 1-2%, far from the target.
Possible Exceptions: AI could make some software platforms more important — because AI agents need tools to operate, and network-effect tools (e.g., Slack) may become entrenched as AI's "critical repository." CRM could become "headless" (de-interfaced), reduced to a database, but could also be reinforced if AI agents work within it.
> Readers should note: Sacerdote admits this thesis is "half-baked," and traditional software does have stickiness (integrations, workflows, customer habits). However, he believes valuations are high, pressure is mounting, and AI coding tools continue to improve — "We will closely monitor whether they can achieve AI revenue that changes their trajectory."
Alex Sacerdote positions Whale Rock as a "learning machine" – a 10-person team with an average of 10 years of experience, 2,500–3,000 in-person management meetings per year, and 20 years of knowledge compounding. The product structure is merely the output of this machine.
Research Process: Rooted in Philip Fisher’s 1950s "scuttlebutt method" – speaking with suppliers, customers, and competitors to identify key characteristics of leading companies. AI tools help quickly get up to speed on new areas (e.g., ABF substrates, PCBs), but "cannot pick stocks for you." The core value of an analyst is "wisdom" – understanding how changes impact the thesis, rather than acting as a reporter.
Product Structure Evolution: For the first 15 years, only a long/short fund existed. Around the 10-year mark, LPs began requesting long-only products. In 2020, private investments were formally opened (LPs could choose a 15% or 25% allocation). Core insight: Large-cap tech stocks are structurally underweighted – endowments underallocate to large-cap tech because they believe "large caps have no alpha," but Sacerdote argues that large-cap tech stocks do have alpha, as it requires 100 diversified PMs to simultaneously realize that "Google is not a loser but a winner," and Whale Rock can identify this earlier than 95% of PMs.
"Three-Legged Stool" Decision Framework: When Sacerdote likes a name + the analyst likes it + an externally respected investor also likes it, the three supporting points greatly enhance conviction.
| Position | Guest Stance | Key Data |
|---|---|---|
| Anthropic | Highest conviction holding | Valuation at investment $18B; revenue from $100M to $9B; DAU 14-15M; code market TAM $500B |
| NVIDIA | Bullish (already held) | Bought at 4x P/E (2023); per-generation chip/rack power consumption up 50-125% |
| Celestica | Bullish | Cloud Ethernet switch market share 50-60%; AI server unit price $200-300K (previously $5K) |
| Elite Materials | Bullish | PCB layer count increased from 10 to 40 layers; unit CAGR 50-60% |
| Corning | Bullish | Optical fiber; one Microsoft data center uses enough fiber to circle the Earth 4.5 times |
| Delta / Advanced Energy | Bullish | ASP increased 40% annually for 4 consecutive years, with higher margins |
| Apple | Previously held, sold in 2012 (acknowledged mistake) | Sold when smartphone penetration was 50%; subsequently grew at 20% CAGR |
| Tesla | Previously held | Bought in 2019 at 5x P/E (auto S-curve) |
| Amazon (AWS) | Previously held | Judged in 2013 that "the war was won"; 7-year lead; bought as "free AWS" |
| Stripe | Previously held (private) | Bought $100M at $35B valuation in 2020; TPV over $1T |
| Google (Alphabet) | Bullish (large position) | Gemini cannot be ignored; excels at PDF parsing |
| OpenAI | Neutral to slightly positive | Consumer lead; enterprise improving; good coding tools |
| Salesforce | Risk warning | AI revenue only $500-700M out of $40B (1-2%); faces threat from AI-native alternatives |
| Slack | Watching | May become entrenched due to AI agent usage |
| Sierra (Brett Taylor) | Watching (not invested) | AI-native application company; CEO previously led CRM, built Google Maps, served as Facebook CIO |
| AppLovin | Historical success case | Analyst discovered through industry conferences and 200+ client interviews |
1. "S-curve + Competitive Advantage + Underestimated Earnings" Three-Factor Framework (Sacerdote): When the three converge, earnings can go from $1 to $10 — NVIDIA bought at 4x PE, Tesla at 5x PE, Apple at 4x PE, Amazon "got AWS for free." The world does not think exponentially.
2. "Hardware De-commoditization" — AI turns every server component from a commodity into a critical part (Sacerdote): Over the past 40 years, hardware saw zero growth and zero innovation; now workloads grow 10x annually, pushing every component to its physical limits, with ASP and margins rising in tandem. Celestica went from a "disaster industry" to an "aircraft critical parts supplier."
3. "AI Version of the Rule of 40" (Sacerdote): Traditional Rule of 40 = Growth rate + Profit margin ≥ 40. New Rule of 40 = AI revenue share × AI category market share. Traditional software companies have only 1-2% AI revenue, far from the target.
4. "The code market is AI's true unlock" (Sacerdote): 20 million programmers globally, each spending $20,000-30,000 annually on tokens, giving the code market alone a TAM of $500 billion. Karpathy and Torvalds have fully pivoted — "haven't written a single line of code except in English."
5. "The trigger for an S-curve is when all adoption barriers are removed" (Sacerdote): Smartphones needed touchscreens + 3G + a $200 price + an ecosystem; EVs needed a $40,000 price + 300-mile range + a supply chain in place. AI is currently at "10 basis points" penetration — the tinkerer phase of a classic S-curve, about to enter early mainstream.
6. "There is structural alpha in large-cap tech" (Sacerdote): It takes 100 diversified PMs to simultaneously realize "Google is not a loser," and Whale Rock can spot it earlier than 95% of PMs. Endowments underweight due to "no alpha in large caps," but leaders in the digital economy are "typically larger, faster, and winner-take-all."
7. "AI is a quadruple blow to traditional software" (Sacerdote): Budgets are consumed by AI, pricing power disappears, hiring freezes reduce seats, and AI-native alternatives pose a threat. Whale Rock has shifted from a 40-50% software position five years ago to net short.
8. "Three-Legged Stool Decision Framework" (Sacerdote): You like it + Analysts like it + Respected external investors like it = three support points. Philip Fisher said, "Know 10-15 like-minded people and share ideas" — this is a two-way street.