Invest Like the Best · Alex Sacerdote · 2026-08-06
Investing Through Technology Cycles
Alex Sacerdote, founder of Whale Rock Capital Management (~$17B AUM), breaks down his S-curve investment framework for technology cycles — why the AI infrastructure buildout resembles an unprecedented “L-curve,” how his firm identified Anthropic as a high-conviction bet during the foundational model shakeout, and how the decommoditization of hardware is creating structural winners across the supply chain. The conversation also covers the risks facing enterprise software incumbents and how the Whale Rock “learning machine” compounds knowledge across decades.
The talk itself: youtube.com/watch?v=DZt1DDmMNGk
A working investor's version of the S-curve lens, with the AI buildout argued as an L-curve rather than a bubble. Worth the hour for the framework, not for the stock talk.
The argument, in 8 sections
18 quotes, none repeated01The Anthropic Bet: How the Foundational Model Oligopoly Emerged
When ChatGPT launched in November 2022, Whale Rock immediately conducted a deep dive into the AI stack — power at the bottom, chips, clouds, foundational models, and applications on top. They invested first in chips and infrastructure (“we know who the winners are, and no matter who wins above, we know we're going to need tremendous amounts of compute”), then turned their attention to the foundational model layer.
What started as ~60 companies pursuing foundational models collapsed into a three-horse race: OpenAI, Anthropic, and Google. Most startups died; even Amazon and Meta struggled. Anthropic emerged as the “dark horse candidate” by focusing purely on enterprise, while OpenAI won the consumer. The real unlock was code — when Anthropic's models reached agentic-level coding capability in 2025, the market exploded. Sacerdote's team heard that internal Anthropic users were spending $100/day on tokens (~$20-30K/year), which when multiplied across an estimated 20 million coders globally pointed to a half-trillion-dollar market from coding alone.
The investment thesis goes beyond just the API. Anthropic is building a full ecosystem — SDK, orchestration layer, and software harnesses around the API — analogous to how AWS built products around its commodity servers that slowly created lock-in. Models are meaningfully differentiated (Anthropic for finance/PE, Google for PDF ingestion), not pure commodities as many assumed.
Cross-reference: consume/concepts/inside-the-exponential for the parallel with Dario Amodei's framing of being on the hockey-stick part of the AI capability curve.
“We wrote in our letter, 'We made the investment at the 180 valuation.' And we said — I think they were hoping to get to a nine billion. And the numbers were like nothing we'd ever seen before: 100 to a billion on the way to nine.”
“A lot of people have routers that switch in between, which sort of makes it sound like they're commodity, but Anthropic — they're very good for anything that has to do with private equity and finance. Google's very good for ingesting PDF.”
02The S-Curve Investment Framework: Map, Moat, Model
Whale Rock's three-part framework: S-curve (identify the technology adoption wave), competitive advantage (find the winner with sustainable moats), and underappreciated earnings power (invest when the market can't see the exponential). Sacerdote emphasizes that the world doesn't think exponentially: markets fixate on the next quarter and next year, but if you understand the S-curve, moats, and how to model, you can predict 2-4 years out.
Every major technology follows the same pattern — smartphones existed 10 years before the iPhone, the internet was around 20 years before Netscape, AI was hidden inside companies before ChatGPT “took it public.” The S-curve stays flat until barriers to adoption are eliminated, then rockets upward in a “tornado of demand” when “everybody in the world knows they need this right away.”
Sacerdote shared the framework's track record: they bought Nvidia in 2023 at 4x earnings, Tesla in 2019 at 5x earnings, Apple at 4x earnings, and got Amazon for AWS essentially for free. The key insight: the S-curve is a map for the future — what Warren Buffett avoided in tech because he “couldn't predict the future” — and it works when paired with understanding the competitive moats and having conviction in the modeling.
Two models for adoption speed: the radio S-curve (7 years to near-100% penetration) and the dishwasher S-curve (slow, because it needs to be plugged into existing systems). B2B tech adoption often follows the dishwasher model — cloud took decades to penetrate because of security concerns. But AI is different: “you just open up the browser and it's there.”
Cross-reference: consume/concepts/s-curves, consume/concepts/s-curves-arent-smooth, consume/concepts/s-curves-following-s-curves.
“When you get the right part of the S-curve, you get exponential unit growth. If you have a very strong business model, your earnings don't grow linearly, they grow exponentially. You know, the world doesn't think exponentially. Very few people believe you can accurately predict 2, 3, 4 years out.”
This is the entire Whale Rock thesis in one paragraph — S-curves provide a map for predicting exponential outcomes that markets systematically underestimate.
“It's okay to miss the first 100%. Peter Lynch, I started at Fidelity and he loved to mentor the young kids. So I got some time with him. He said, 'White out the chart. It's all about the future.'”
A liberating principle for investors — you don't need to catch the bottom of an S-curve. If the TAM is half a trillion, growth goes on for a long time.
“The dishwasher S-curve is like that because it needs to be plugged into the back end... consumers generally tend to go a lot faster. But what's amazing about AI is you just — at least with consumers or even business — you just open up the browser and it's there.”
03The AI "L-Curve": Why This Time Is Different
Sacerdote's most striking claim: the enterprise AI market is less than 1% penetrated, with only ~10 basis points of knowledge workers truly using AI the way it can be used, and yet the world is already sold out of compute. He calls this an “L-curve” — just straight up — rather than a traditional S-curve.
The progression: from “AI 1.0” (a search engine on steroids) to agentic AI with new primitives (Claude on your computer, skills, bots, corporate-scale deployments). Using the classic technology adoption lifecycle: the current 10 basis points are the tinkerers, about to move into early adopters, then early mainstream — going from ~0.1% to 2-3%, to 5%, to 15% in the next four years. A “light switch went off in the enterprise this year” where everyone realizes they need to move fast.
The implications for infrastructure are staggering: Anthropic has only half the compute it needs today — before the massive adoption wave. Marc Andreessen's prediction: in the next four years, there won't be enough compute.
“The enterprise AI or enterprise application AI market is less than 1% penetrated and we've never seen — you know, we talk about S-curves, we call this an L curve, just straight up.”
Sacerdote makes the case that AI adoption may break the traditional S-curve model entirely — and we're already sold out of compute at near-zero penetration.
“We're at 10 basis points of people really using AI and we're already sold out of all the — there's not enough compute in the world. So Anthropic has half of what they need right now and that's before this huge takeup.”
04Competitive Advantages in the Digital World Are More Powerful Than Offline
Contrary to the common fear that tech companies face constant disruption, Sacerdote argues that competitive advantages in the digital world can be more powerful than those in the offline world. The framework includes:
Network effects: LinkedIn, Facebook, Alibaba — winner-take-most dynamics
Industry standard: Oracle's chokehold on relational databases through DBAs and tuned software
Scale advantage: Anthropic going to $30B in sales or Amazon achieving Walmart-size scale in 5 years instead of 40
Platform: AWS building an ecosystem that competitors couldn't replicate
Critical IP: Qualcomm (every phone), ASML (every chip)
Brand: Google, Amazon, Elon's companies — never had to advertise; the cost-to-acquire vs. lifetime value is the whole business model
First-mover + ecosystem + scale compounding: AWS had a 7-year lead, became a platform, then got 10x the size — no one could invest enough R&D to catch up
Sacerdote notes that being in the right S-curve without a competitive advantage means you still lose — Rim, Palm, Nokia, HTC, LG, Motorola all rode the smartphone wave straight to zero. The foundational model layer followed the same pattern: ~50 companies tried, most fell away, and 2-3 emerged at the top.
For Anthropic specifically: critical IP in code, a strong enterprise brand (“go talk to any CIO and the first thing they'll say is Claude”), escape velocity in scale, and the recursive improvement flywheel where leading in code feeds back into model quality.
Cross-reference: consume/concepts/winners-look-invulnerable-until-they-dont.
“What we've found over the years is some of the competitive advantages within the digital world are more powerful, if not equally or more powerful, than in the offline world.”
A direct rebuttal to the “tech is too fast-changing to invest in” argument — digital moats (network effects, scale, IP) compound faster than physical ones.
“Almost all the companies I mentioned — Apple — they have all of these rolled into one.”
05The Decommoditization of Hardware: A Renaissance After 40 Years of Stasis
For 40 years, nothing changed in the data center. Even the cloud era ran on Intel x86, computing workloads grew at 25-40% annually, and Moore's Law improved at roughly the same rate — so hardware required little innovation and the entire industry commoditized (chips, PCBs, memory, enclosures, networking). A 1-gig-to-10-gig networking upgrade took 7 years and commoditized within the first year.
AI changed everything. Workloads are now growing 10x every year, pushing every aspect of hardware to its physical limits. Sacerdote calls this the “decommoditization of the hardware industry” — not just tremendous unit growth, but innovation requirements that create structural competitive advantages.
Examples across the supply chain:
Celestica: A contract manufacturer (historically a “disaster industry since 1999”) became the sole supplier of Google's TPU server. AI servers are $200-300K liquid-cooled machines vs. $5K throwaway servers — they become critical infrastructure, like a part on a plane. Celestica also has 50-60% share of the cloud Ethernet switch market.
Elite Materials: Produces copper clad laminate for PCBs — AI servers need 40-layer boards vs. 10 for regular servers. Unit growth + rising layer counts + rising ASPs + rising margins = “50% to 60% CAGR just in the units.”
Corning: Their fiber goes into AI data centers — a single Microsoft data center had enough fiber to circle the world 4.5 times. Corning's fiber is thinner, more bendable, higher-margin, and the fastest-growing part of their business. Future upside: when scale-up (GPU-to-GPU) networking moves from copper to fiber, Corning's opportunity 2-3x.
Power supplies: Every Nvidia chip/rack uses 50-125% more power, driving ASPs up 40% annually for years — with higher margins.
The broader picture: if the L-curve thesis is right, the world is already 30% short on DRAM, NAND, PCBs, and other components — and demand hasn't even started in earnest.
“In the old days you would go from 100 gig to 400 to 800 — it would be a 7-year cycle to upgrade. Now they're upgrading every year and that's really hard to do.”
“We see that up and down the supply chain. You find these companies — like Corning, they make the fiber. They've got some ridiculously high share of the fiber. I was reading this Microsoft data center they just built — there's enough fiber to circle the world four and a half times in that one thing.”
06Enterprise Software: Running Into a Jet Engine
Whale Rock once had 40-50% of its portfolio in software. After their April 2023 deep dive, they sold almost all of it and entered 2026 “net net short” — a move that helped significantly in Q1. The thesis: the old way of software is like pen and paper, the new way is like a “jet engine” or “the transporter from Star Trek.”
Four headwinds for enterprise software:
1. Priority displacement: Software has fallen down every CIO's priority list — they're spending on Anthropic tokens because the ROI is faster. 2. Budget pressure: AI spend cannibalizes software budgets. 3. Pricing power evaporating: Software companies that raised prices annually are now nervous to do so. 4. Seat erosion: If companies freeze hiring or cut jobs due to AI, software seats shrink.
Incumbent software companies face a structural challenge: when you're Salesforce with $40B in revenue trying to build a $500-700M AI business, it barely moves the needle. Sacerdote introduces a modified “rule of 40” for AI: multiply your AI revenue percentage by your market share in that category. A company with 30% AI revenue and 30% market share scores 60 — that's “a great place to look.” Enterprise software companies are at 1-2%.
However, Sacerdote acknowledges a counter-narrative forming: AI agents operating inside existing software tools (Slack, CRM, HR systems) could actually solidify incumbents' positions. If agents work through these platforms like a human would, the platforms become more embedded, not less.
Cross-reference: consume/concepts/three-generations — technology transitions appear as threats before they're understood as new platforms.
“The old way of software is like using a pen and paper or it's like a horse and buggy. The new way of software is like a jet engine or frankly like the transporter from Star Trek. It's so revolutionary changing that it feels like it has to be disruptive now.”
Captures the scale of disruption facing enterprise software — this isn't incremental improvement; it's a paradigm shift that may obsolete existing business models.
“The first thing you do with Claude? You plug it into Slack. If that can become a key repository, that will make Slack a permanent fixture within the organization.”
07Risks to the AI Bull Case
Despite his extreme bullishness, Sacerdote outlines specific risks:
Public negativity and regulation: Only 20% of the population is optimistic about AI. Maine banned data centers. Negative regulation is possible, though “the genie is out of the bottle.”
Model improvement plateau: If models stop improving, open source catches up and it becomes a race to the bottom — bad for model companies, though chip companies don't care who wins the tokens.
One or two players faltering: If a major player pulls back on AI investment, excess compute capacity could emerge. However, Sacerdote noted that when Oracle cancelled a deal, Meta stepped in immediately.
DeepSeek/China open source risk: Open source models can get close to the frontier but struggle to leapfrog — the gap from 80% to 85% benchmark performance is a huge unlock that requires massive compute.
“If good enough is good enough, I won't have a business. Now every year he made the graphics a little bit better and people always wanted the best.”
08The Whale Rock Learning Machine: Scuttlebutt at Scale
Sacerdote describes Whale Rock as a “learning machine” — a team of 10 highly experienced individuals who do 2,500-3,000 face-to-face meetings with management teams per year. The research method traces directly to Philip Fisher's 1950s “scuttlebutt approach”: talk to suppliers, customers, competitors — looking for the key characteristics of leading companies and developing conviction.
Sacerdote draws a line between what AI can and cannot do in investing: AI is a great reporter (write notes, review quarters), but can't pick stocks or predict the future. The analyst's edge is wisdom on top — “What does this mean? How does this deal with our thesis? What changed? Don't just be a reporter.”
He also highlights the “tripod” method for building conviction: when Sacerdote likes something, his analyst likes it, and a respected peer investor also likes it — “that's three legs of the stool” that deepens conviction.
Cross-reference: consume/concepts/judgment-human-advantage, consume/concepts/protect-your-thinking.
“The system we use is right out of *Common Stocks and Uncommon Profits* which was written by Philip Fisher in the 1950s. And it's the scuttlebutt approach. It's growth investing — get out there and talk to suppliers, customers, competitors, looking for the key characteristics of these leading companies and really developing conviction in them.”
Also worth keeping
3 more · each links to its minute“We were buying Nvidia in 2023, we were paying four times earnings. When we bought Tesla in 2019 for the car S-curve, we were paying five times earnings. When we were owning Apple, we were paying four times earnings. When we bought Amazon for AWS, we were getting it for free.”
The payoff of the framework — these generational companies were acquired at single-digit P/E multiples because the market couldn't see the exponential.
“For the past 40 years, nothing has changed in the data center. Even with cloud, we're basically Intel x86. ... Now you go to AI and the workloads are growing 10x every year and they're pushing every single aspect of this hardware to the physical limits.”
Explains why hardware is experiencing a structural transformation, not a cyclical uptick — the “decommoditization” thesis in one quote.
“AI is a great reporter. It can't quite pick into the future. ... There better be a really good paragraph on top which is the wisdom. What does this mean? How does this deal with our thesis? What changed? Don't just be a reporter.”
Defines the durable human edge in investing — AI accelerates research but judgment, synthesis, and conviction remain irreplaceable.
What to take away
5 lines- 01The S-curve is your map for the future — but only if paired with moat analysis and earnings modeling
Technology adoption follows predictable patterns, and the critical skill is identifying which company will capture the value (competitive advantage) and when the market is underpricing that outcome (underappreciated earnings power). The framework works because markets systematically underestimate exponential growth.
- 02AI infrastructure is the clearest bet right now, precisely because the winner is uncertain at other layers
Sacerdote's team invested in chips and hardware first because compute demand is guaranteed regardless of which model or application wins. The “decommoditization” thesis means hardware companies that were once cyclical commodity plays now have durable competitive advantages and long-term visibility.
- 03Enterprise software faces an existential reckoning that most incumbents haven't faced
The combination of budget displacement (CIOs spending on tokens instead of SaaS), pricing pressure, seat erosion, and the risk of AI-native competitors makes the software sector structurally challenged. The counter-narrative — that agents operating inside existing tools will embed them further — is plausible but unproven.
- 04The difference between a commodity and a moat in AI is differentiation that compounds
Anthropic stays ahead in coding because market share in code feeds back into model quality. Celestica retains its position because liquid cooling and complex manufacturing are genuinely hard. The “router” argument (switching between models) creates the illusion of commoditization where real differentiation exists.
- 05Conviction comes from triangulation — the scuttlebutt method, your analysts, and trusted peers
The “tripod” (you + analyst + respected peer) plus the compounding knowledge of 20 years of face-to-face meetings creates an edge that AI can augment but not replace. The durable human advantage is the “wisdom paragraph on top” — interpreting what the data means. ---