Summary

Sacerdote’s highest-conviction position is Anthropic, bought in the August 2025 round at the $180 valuation after passing on the $60B round (“the gross margins were negative and frankly we hadn’t seen coding explode”). The revenue ramp — “100 to a billion on the way to 9” — was “like nothing we’d ever seen before,” and the coding math alone underwrites it: people inside Anthropic spending $100/day on tokens annualizes to $20-30K per coder, times 20 million coders worldwide = “a half a trillion dollar market just from coding alone,” on 7-9 month old technology.
The foundational model layer began to look like a three-horse race and somewhat an oligopoly — Anthropic (enterprise), OpenAI (consumer), Gemini — echoing how three clouds came to underpin all of SaaS. Models are not commodities: “there’s tremendous differentiation within” (Anthropic for private equity and finance, Google for ingesting PDFs), and open-source can approach the frontier but “can’t leapfrog it, and then they kind of falter.”
Enterprise AI is “less than 1% penetrated — we call this an L curve, just straight up.” Only ~10 bips of knowledge workers truly use AI (per Sunder), heading to 1-2% or 3%, then 5%, then 15% within four years — yet compute is already sold out, with “Anthropic has half of what they need right now” before the take-up even starts. Mark Andre’s one certainty for the next four years: not enough compute.
The Whale Rock framework — S-curve + competitive advantage + underappreciated earnings power — bought Nvidia in 2023 at 4x earnings, Tesla in 2019 at 5x, and AWS “for free,” because “the world doesn’t think exponentially.” At around 30-40% penetration, exponential growth ends, the sell side catches up, and beats stop.
Whale Rock went from 40-50% of the portfolio in application software to net short software entering this year: AI products weren’t good enough to charge for, software fell down every CIO’s priority list (“they’re spending it on Anthropic tokens because there’s faster ROI there”), annual price hikes are now risky, and job cuts hurt seat counts. Half-baked offset: agents operating inside Slack or CRM could entrench the systems of record.
The infrastructure trade is the “decommoditization of the hardware industry”: AI workloads growing 10x/year push every server component to physical limits — Celestica (bought at 8x earnings as sole Google TPU server supplier, 50-60% of cloud Ethernet switching), 40-layer PCBs, Corning fiber (scale-up moving from copper to fiber “two to three X’s Corning’s opportunity”), power ASPs up 40% for the next four years. “We’re already 30% short” DRAM, NAND, and PCB supply.
Chief risk: if the leaders hit a wall, open source catches up and it might be a race to the bottom — probably won’t be good for model stocks, could be good for chips (“chip companies don’t care who wins”).

Deep dive

1. Anthropic: passed, then invested

When ChatGPT fired the starting gun in November 2022, Whale Rock’s 10-person team did a massive deep dive and sequenced the stack deliberately: chips and infrastructure first, because “no matter who wins above… we know we’re going to need tremendous amounts of compute.” An April 2023 webinar laid out the scenarios for the model layer — winner-take-all, open-source commodity race to zero, or oligopoly of three or four.
Over three years the answer began to emerge: nearly all 60 startups “fell away and died,” Amazon “never really showed up,” Meta “came in strong and then basically their effort faltered and they had to do a total reboot.” Anthropic was the dark horse focused on enterprise while OpenAI had kind of won consumer and Gemini “can never be counted out” — a race that began to look like an oligopoly, rhyming with the three clouds that underpin all of SaaS.
Sacerdote passed on Anthropic’s $60B round — “the gross margins were negative and frankly we hadn’t seen coding explode.” By August 2025 the picture flipped: time with Dario, almost no team turnover, and a revenue ramp “like nothing we’d ever seen before — 100 to a billion on the way to 9.” Whale Rock pitched its way into the investment at the $180 valuation with a 90-page deck built using Claude Code to scour the internet for coding-market feedback, and “punched above our weight in terms of the allocation.”

2. Code is the true unlock

The first generation — Microsoft Copilot at $20/month — could “improve your grammar of coding, maybe find a bug.” Then Anthropic’s mid-2025 release went agentic and the market exploded. Sacerdote’s telling of the flip: Karpathy and Linus Torvalds both reversed — last year’s tools wrote 20% of code with 80% handwritten, and now Karpathy “hasn’t written a line of code except in English.”
The napkin math that anchored the position: people inside Anthropic were spending $100 a day on tokens — $20-30K a year — and with ~20 million coders in the world, “you’ve got a half a trillion dollar market just from coding alone. And mind you, that was on 7, 8, 9 month old technology.”
The strategic kicker is recursive: Anthropic leads in code, feeds that code back into its own model, and “if you look at the pace of their innovation, it’s accelerating” — the setup for a liftoff stage.

3. Models are not commodities — and the moats are stacking

Everyone assumed foundational models would be pure commodity like cloud servers. Wrong, in Sacerdote’s view: “there’s tremendous differentiation within” — Anthropic excels at private equity and finance, Google at ingesting PDFs — and routers switching between models make it merely look commoditized. Open source (the China risk) can get to 80% of the benchmarks, but “going from 80 to 85 is a huge unlock” and without frontier compute “they can’t leapfrog it, and then they kind of falter.”
Beyond the API, Anthropic is building the ecosystem — SDK, orchestration layer, the “harness” of software that gets the most out of the model. The rhyme is AWS in 2013: “people thought it was a commodity server up in a warehouse, big deal,” while Amazon invented products that slowly built lock-in.
Why the leaders hold: critical IP, enterprise brand (“go talk to any CIO and the first thing they’ll say is Claude”), and escape velocity — both Anthropic and OpenAI found ways to raise capital against rivals with huge cash cows, and with 10x sales growth “it looks like they’ve reached escape velocity.”

4. The L-curve: 10 basis points penetrated, compute already gone

The 800 million people using AI today are running “AI 1.0 — a search engine on steroids.” The real curve — skills, true AI bots, corporate builds — is barely started: Sunder’s figure is 10 bips of the world’s knowledge workers. “You’re going to go from 10 bips to 1-2% or 3%, then 5%, then 15% in the next four years.” Enterprise application AI is “less than 1% penetrated… we call this an L curve, just straight up.”
The supply side is the tell: at 10 bips of adoption, “there’s not enough compute in the world. Anthropic has half of what they need right now — and that’s before this huge takeup.” Mark Andre’s one certainty for the next four years: not enough compute.
Why AI adopts faster than cloud ever did: cloud and SaaS were “like the dishwasher — it’s got to be plugged in,” capping growth at 30-50%. With AI “you just open up the browser and it’s there” — hence the backwards-L shape.

5. The framework: exponential earnings bought at single-digit P/Es

The three-part screen — S-curve, competitive advantage, underappreciated long-term earnings power — exploits one behavioral gap: “the world doesn’t think exponentially.” On the right part of the curve with a strong model, earnings compound rather than grow linearly, “and it happens way more than you think.”
The receipts: Nvidia in 2023 at four times earnings, Tesla in 2019 at five times, Apple at four times, “when we bought Amazon for AWS, we were getting it for free.” Few believe you can predict 2-4 years out — “but if you follow and understand the S-curve and you know the moats and you know how to model, you really can.”
Sizing the curve matters as much as spotting it. AWS addressed $600B of IT systems; Whale Rock assumed 50% deflation, then learned it wasn’t deflationary at all — the TAM was far bigger. But curves can stall: EVs “hit a big wall at 10 or 15%” against an expected 40-50%. At around 30-40% penetration, exponential growth ends, when the sell side catches up and beats stop — Apple was sold in 2012 at ~50% US smartphone penetration, after the 50-70%/year gains of the 0-to-50 stretch.

6. Timing the flatline: trust anecdotes, not data

Technologies flatline for a decade-plus before inflecting — smartphones existed 10 years before iPhone, Tesla was public 15 years before going vertical in 2019. The trigger is barriers falling: Jobs got the price from $500-600 to $200 with 3G, touchscreen, “so easy your grandmother could do it”; Elon got the EV to $40K and 300 miles of range. Then comes “the tornado of demand.”
Per Andy Grove, “when you have strategic inflection points, you can’t trust the data” — it’s right-brain pattern recognition (Sacerdote cites The Tao Jones Averages). His examples: a 12-year-old in China playing a serious game on a huge phone told him mobile gaming had arrived; at the Gartner IT Symposium the AWS grand ballroom was packed at 9, 10, and 11 o’clock — “you could actually see the demand exploding before it happened.”
Slope varies by plumbing: a Clayton Christensen collaborator (likely Horace Dediu) charted 100 years of S-curves for the firm — radio hit ~100% in 7 years; the dishwasher crawled because it had to be plumbed in. B2B internet ultimately happened 20 years later with SaaS because the infrastructure wasn’t there; cloud needed the CIA and Capital One to break the security taboo. And it’s fine to be late: “It’s okay to miss the first 100%.” Peter Lynch’s mentoring line: “White out the chart. It’s all about the future.”
The moat taxonomy that decides who captures the curve: network effects, industry standard (Oracle, Bloomberg), scale (“Amazon got a Walmart-size scale advantage in 5 years versus 40”), platform, critical IP (Qualcomm, ASML), brand. Without one, the best S-curve of all time still pays zero — “RIM, Palm, Nokia, LG, Motorola… negative, negative, negative.” The 2013 Robin Hood AWS pitch: “the bulls have no idea what they’re sitting on.”

7. Software: from half the book to net short

Five years ago software was 40-50% of the portfolio. The April 2023 view — big sales forces plus data plus AI APIs would be “amazing for software” — died on contact: “pretty quickly we realized their AI products were not very good… nobody could charge for them.” Whale Rock sold almost all of it and entered this year net short, which “really helped us in the first quarter.”
The four-part bear case: software has fallen down every CIO’s priority list (“they’re spending it on Anthropic tokens because there’s faster ROI there”); that spend squeezes budgets; annual price increases are now risky; and job cuts hurt seats. The framing: old software is “a horse and buggy”; the new way is “a jet engine, or frankly the transporter from Star Trek.”
The bulls’ stickiness argument is “all true” — tablets didn’t kill the PC, companies buy rather than build — but “you can’t imagine a world where in 1-5 years you could have a brand new AI-native company going after each one of these incumbents.” The base problem is arithmetic: Salesforce has $40B of sales and maybe $500-700M of AI ARR.
The half-baked counter Sacerdote is watching: AI may entrench some platforms — “what’s the first thing you do with Claude? You plug it into Slack.” If agents do the work inside CRM, “that will solidify CRM” — though the bear case is being relegated to a headless database.

8. The decommoditization of hardware

For 40 years “nothing changed in the data center”: Intel x86, workloads growing 25-40%/year — matched by Moore’s law — and every component commoditized (one-gig to 10-gig Ethernet took 7 years). Now workloads grow 10x every year, pushing every part to physical limits: “we call it the decommoditization of the hardware industry.” Shawn Maguire’s line from three years ago: “I wish I could come back and be a hardware hedge fund.”
Celestica is the specimen: a “disaster industry since 1999” contract manufacturer that kept its IBM supercomputing talent, turned up as sole supplier of the Google TPU server trading at 8x earnings, and holds 50-60% of cloud Ethernet switching. A liquid-cooled AI server is a $200-300K machine versus a $5K throwaway box — “you become like critical infrastructure, like selling a critical part on a plane. You’ll never get swapped out.”
The pattern repeats down the stack: AI PCBs need 40 layers versus 10 (Whale Rock owns Elite Materials, the copper-clad-laminate leader) — units plus layer count plus ASPs compound into a “35 to 50% topline CAGR for the next four years with rising margins,” and visibility went from “call you next week” to four-year roadmaps. Corning’s fiber (one Microsoft data center holds enough to circle the world 4.5 times) gets its real kicker when scale-up networking moves from copper to fiber — “that two to three X’s Corning’s opportunity.” Power supplies: every Nvidia chip or rack uses 50-125% more power, driving Delta and Advanced Energy ASPs up 40% for the next four years. And “we’re already 30% short” DRAM, NAND, and PCBs — “even if it is a commodity, it’s going to be a great cycle.”

9. Why doesn’t everyone do this — and what breaks the thesis

On giving away the playbook: “My mom said, why do you tell everyone your secret? It’s like — why does the casino teach people how to play blackjack? It’s really hard to do.” Almost nobody covered hardware (“You and Gavin, that’s it”), and pure semi analysts missed Nvidia because they couldn’t see the foundational model layer. The bubble call recurs every six months — “the bear cases are not totally without merit” — but the holistic view sustains conviction. One refinement: his AI-era rule of 40 is % of sales from AI plus % market share in that category, and rate of change matters more than the absolute — Claude once plotted the chart wrong precisely because it missed the rate of change.
The risks he actually worries about: public and political hostility (“I think Maine just banned data centers”; only 20% of people are optimistic about AI) — though “the genie is out of the bottle.” Worse: if Anthropic or OpenAI “hits a wall and stops improving, the open-source models will catch up and it might be a race to the bottom — probably won’t be good for the stocks, could be good for the chip companies.” Jensen’s old graphics-chip line: “If good enough is good enough, I won’t have a business.”
A faltering player could strand compute — though “if AI is so big, somebody else will suck that up,” as when Oracle cancelled a big deal and Meta went right in. On the application layer he’s deliberately absent: apps “always come later,” the model/app boundary is unclear, and the ecosystem is “still kind of unclear and a little bit dangerous” — he’s watching Brett Taylor’s Sierra as the test case, uninvolved.

10. The learning machine: scuttlebutt, privates, and the mega-cap fund

The engine is likely Philip Fisher’s scuttlebutt run at industrial scale: 2,500-3,000 face-to-face management meetings a year (10-15% with privates), knowledge compounded over 20 years. AI helps write notes, “but there better be a really good paragraph on top which is the wisdom… don’t just be a reporter.” The likely AppLovin call — two analysts who tracked it from private, went to the Vegas app-advertising conference, and built the relationship with likely Adam Foroughi — “I don’t see AI doing that.” His conviction test is the tripod: “when I like something, and my analyst likes it, and somebody who I really respect also likes it.”
The Stripe entry shows the privates playbook: diligencing likely Adyen (200 customer calls) revealed “this is Coke and Pepsi,” so he met the likely Collison brothers in 2019, then bought a $100M block from a seller in April 2020 at ~$35B — underwriting take-rate (40-50 bips vs likely Adyen’s 25-30) against a disclosed $550B TPV that “was closer to the 1 trillion.” Sellers like Whale Rock because it holds into the public market, as it did with likely Nubank. Context: the unicorn market is now bigger than Germany’s or the UK’s stock market.
The newest product, the Whale Rock Mega Cap Tech Fund (top-30 global market caps, own the best 12-13), attacks a structural anomaly: endowments are “massively underweight the largest tech companies in the world” on the belief there’s no alpha in large cap. His counter: “it takes 100 diversified PMs to realize Google’s not a loser — can we figure that out before 95% of them do?”

Episode 185 -June 5, 2026 Market Overview: June 9, 2026