Top Tech Fund Discusses AI Investment: Seizing "Exponential Growth" Opportunities When Crossing the S-Curve Inflection Point

Investors who can identify the turning point of the S-curve often enter at undervalued prices before exponential growth launches, and enjoy excess returns over the years. AI is currently at such a turning point.

Alex Sacerdote, founder of Whale Rock Capital—a well-known US hedge fund heavy on tech stocks—made the judgment on the June 9 episode of "Invest Like The Best": AI company penetration is less than 1%, the S-curve is turning into a near-vertical "L-curve"—this is precisely the crucial moment to cross the inflection point and seize exponential growth opportunity.

Scarcity of computing power is the primary signal for crossing the inflection point. Demand is taut across the entire supply chain, from chips to foundational models; traditional enterprise software faces structural replacement. Alex points out that Claude Code is moving from assisted coding to autonomous agents, opening up broader commercialization.

This methodology has been proven: Buying Nvidia at a 4x PE ratio in 2023, Tesla at a 5x PE in 2019, almost zero-cost exposure to AWS in early days. “The world doesn’t understand exponential thinking,” Alex says, “But if you understand S-curves and moats, you can foresee these opportunities.”

Currently, AI investing has shifted from chips to a "three-horse race" between OpenAI, Anthropic, and Gemini, closer to an oligopoly. Risks also exist: model progress slowdown, open source catching up, regulatory pressure remain key variables in this AI cycle. But for investors who grasp the S-curve inflection point, exponential returns are just beginning.

S-Curve Investing: Betting Before Consensus Forms

Whale Rock’s investment framework consists of three core elements: identifying S-curve position, evaluating competitive advantage, and uncovering undervalued long-term profit potential.

Alex stresses, the key to the S-curve is the inflection point—once technology crosses the adoption threshold, growth shifts from linear to exponential, and combined with strong business models, profits often jump from $1 to $10 to $50. "The world doesn’t think exponentially; everyone is too focused on next quarter, next year."

This framework has been repeatedly proven: Pushing Amazon AWS in 2013 when “longs had no idea what they held”; heavy Tesla at 5x PE in 2019 after price and range anxiety were conquered; buying Nvidia at 4x PE in 2023—three moves, all just left of the turning point.

Exit logic is equally clear. When penetration reaches 30%-40%, the exponential growth phase basically ends, sell-side analysts catch up, and upside-surprise disappears. Whale Rock exited Apple when US smartphone penetration hit 50%, after which Apple’s annualized returns dropped from 50%-70% to about 20%.

How to spot the inflection point early? Alex uses “field observation”: At enterprise tech conferences like Gartner IT Symposium, he observes booth traffic of a company. While early investing in Splunk, VMware, and AWS, he saw crowded, standing-room-only scenes. “You can see it with your own eyes before demand explodes.”

Deep Users Only 0.1%, AI Still at Earliest S-Curve Stage

Alex believes, AI is still at the earliest S-curve stage—the global deep user base among knowledge workers is only 0.1%.

This figure comes from Google CEO Sundar Pichai’s estimate. Alex notes that Anthropic has about 14–15 million daily active users, but intense users are still few. “Now is the ‘tryout’ phase, next comes early adopters, then early mainstream users.”

Whale Rock forecasts, in the next four years, AI deep user penetration will jump from 0.1% to 2–5%. The infrastructure layer is now about 10% penetration, enterprise application layer is less than 1%. ‘This isn’t an S-curve, it’s an L-curve—straight up.’

Hard data supports this: global computing power is in severe shortage, Anthropic itself meets only half of its compute needs—and this is before large-scale enterprise procurement arrives. Alex quotes Marc Andreessen: “The one thing certain about the next four years is that compute will always be insufficient.”

Regarding adoption pace, Alex refers to two historical cases: radio achieved near 100% penetration in seven years—a very steep S-curve; dishwashers, due to needing integration with household plumbing, had a much flatter curve. Enterprise software adoption is closer to dishwashers, but AI is unique—users just need to open a browser, which is why its growth curve is nearly vertical.

Below is the full interview (AI-assisted translation, some edits):

Host:
Today’s conversation will deeply explore the technology adoption S-curve, the underlying logic of AI investing, and how to find and hold winners during market transformation. My guest is Whale Rock founder Alex. His entire investment framework is built on understanding tech S-curves, identifying sustainable competitive advantages, and digging into undervalued profit potential. He invested early in Nvidia, Tesla, Amazon’s AWS. Now, his highest conviction position is Anthropic. Welcome, Alex.

Alex:
Thank you, glad to be here.

Host:
You said Anthropic is your highest conviction position. Using this as a starting case, could you talk about how you and investors like you invest in the private market? Why is Anthropic your most certain bet?

Alex:
Sure. When OpenAI's ChatGPT emerged in November 2022, we immediately had our ten-person team study it deeply. Each time a new computing paradigm appears, it spawns a new tech stack, creating new winners and losers atop the old. In this new stack, the base is chips and cloud; the middle is foundational models; the top is applications.

At the time (early 2023), we decided to invest first in chips and infrastructure. Because regardless of who wins at the top, massive compute is always needed. We made deep analyses about this. Over the next two or three years, we gained a clearer picture of how the foundational model layer would evolve. Sixty companies flooded in. We predicted at an April 2023 webinar that the market might be winner-takes-all, might be commoditized (open source leading to price war), or a three-to-four-player oligopoly.

Later, almost all startups disappeared. Large companies like Amazon and Meta also struggled, Meta even had to restart its AI project. Anthropic emerged as a dark horse. They focused on the enterprise market, while OpenAI won the consumer market; Google’s Gemini is strong too. The pattern became a three-legged oligopoly, similar to how cloud (AWS, Azure, GCP) evolved.

We began feeling comfortable with this. Open source has risks, but its compute is limited, it can approach the leaders but can’t achieve qualitative leaps. Scalability laws and feedback loops indicate large scope for model progress. Industry insiders we know all believe the scalability laws will continue to hold.

Host:
What was the pivotal turning point?

Alex:
The biggest catalyst was code. Early models had potential, but it was unclear if they could really replace labor. By 2025, AI coding tools began exploding. First gen like Microsoft Copilot: $20/month, fixing syntax and bugs. But Anthropic’s model released mid-year leapt hugely—could run autonomously (Agentic). We heard that at Anthropic, some spend $100/day on tokens—$20-30k/year. There are 20 million programmers worldwide, just coding is a trillion-dollar market. And this is based on models 7-9 months old.

Another recent breakthrough: Claude Code is almost fully autonomous. Last year Andrej Karpathy and Linus Torvalds said AI could write 20% of code, 80% must be handwritten; now with latest models, this ratio has flipped completely. Karpathy himself barely handwrites code now, only describes it in English.

Host:
Aren’t foundational models commodities? How do they build moats?

Alex:
People generally think AI models are pure commodities, but they actually have huge differentiation. Different training methods create different specialties. Anthropic excels at private equity and finance, Google excels at PDFs—these are key IP. Also, Anthropic isn’t just selling APIs or models; they’re building a complete ecosystem around API—SDKs, collaboration layer (Claude for Co-work), orchestration layer, etc, which they call “Harness” (suite). It’s like AWS in the early days: people thought it was just bulk hosting, but it locked in customers through early product innovation.

Host:
How do you define the present stage?

Alex:
We often talk S-curve, but we call today’s enterprise AI market the L-curve—straight up. Penetration is less than 1%. Maybe 800 million use AI, but that's AI 1.0—a search engine on steroids. Real users, as Sundar says, are just 10 basis points (0.1%) of global knowledge workers. Claude has 14–15 million daily actives, but only a small fraction use AI this way. This is the typical S-curve start: driven by experimenters, then early adopters and mainstream follow. In the next four years, penetration will go from 0.1% to 1–2%, then to 5%, then to 15%. This year in the enterprise, it’s like a light switch—everyone realized action is urgent.

It’s like the 1998 internet—you know you need a website, but it’s hard to build. Now everything is maturing fast. We’ve never seen this before, hence the L-curve. Even infrastructure layer has only 10 basis point effective utilization, but global compute is already tight. Anthropic has only half the compute it needs. Marc Andreessen said, in the next four years, compute will always be insufficient.

Host:
As a former public markets investor (who could buy any stock anytime), how do you get the sizes you want in key private companies like Stripe, Databricks, OpenAI, Anthropic?

Alex:
Take Anthropic—starting from contact with their finance team via an analyst. We missed the $60B round because we didn’t know them, margins were negative, and hadn't seen coding takeoff. But then I spoke to Dario (CEO), listened to his podcast, realized their management team is excellent—focused, dedicated, minimal turnover, high code quality, executing business plans. So we reached out, compiled a 90-page PPT, used Claude Code to gather feedback on coding market and product strengths/weaknesses. They accepted us into this round.

Overall, we do 2000–3000 face-to-face meetings each year with management teams, 10–15% are private companies. Our first private investment was Stripe. At the time, we held another payments company, Audion, so to invest in Audion, we had to understand Stripe in depth. Did lots of diligence, spoke to 200 clients, found Stripe like Coke and Pepsi. In 2019, met the Collison brothers, bought Stripe stock from a VC friend at $35B in April 2020. We knew TPV topped $500B, Audion take rate 25–30 bps, Stripe 40–50 bps, so estimated profitability. Later, it proved higher take rate and TPV, near $1T. That deal ultimately grew to $100M for us.

Host:
Your whole investment framework is built on S-curves. Dive into your deep understanding of S-curves.

Alex:
Our framework has three parts: S-curve, competitive advantage, and undervalued profit potential.

When you’re on the right S-curve spot, you get exponential unit growth. With a strong business model (common in tech), profits grow exponentially, not linearly. The world usually doesn’t think exponentially, few believe you can predict 2, 3, 4 years ahead. But if you track S-curves, understand their patterns, model them, you can really predict the greatness.

Crucial: if you’re at the right S-curve position, you can buy the world’s best businesses at ultra-low PE ratios.We bought Nvidia at 4x PE in 2023.Bought Tesla (based on auto S-curve) at 5x PE in 2019.Held Apple at 4x PE.Bought Amazon, equivalent to free exposure to AWS.

S-curves matter because every technology follows this—a long dormant phase before eruption. Smartphones existed a decade before iPhone, internet twenty years before Netscape. Only after adoption barriers are cleared does demand tornado to life.

Host:
What factors determine S-curve “height” (ultimate scale)?

Alex:
When Amazon launched AWS, hidden in retail, its TAM was largest in enterprise IT history—facing a $600B IT systems market. We thought it’d bring 50% deflation, so TAM shrinks. But later saw building yourself was same cost as AWS, so TAM was even bigger. So you must assess time to takeoff, but also S-curve “height”—it determines when to sell, how long to hold.

S-curve failures exist too—like EVs. Thought they’d take 40–50% of car market, but hit big wall at 10–15%. You must adjust dynamically. Normally, when penetration hits 30–40%, exponential growth stops. We made a mistake on Apple: sold in 2012 when US smartphone penetration hit 50%. Apple stayed leader, compound growth at 20% via App Store, but 50–70% annual growth was done.

Host:
How do you judge buying timing? Especially during the flat part of the S-curve that can last ten years.

Alex:
Andy Grove said at the strategic inflection, don’t trust data—rely on intuition and anecdotal evidence. I like a book called "Dow Theory" (actually another book, but meaning as intended), talking about right and left brain. Best investors have a creative side, connect dots visually.

When investing in mobile games S-curve, early phones were small and weak. But I saw a 12-year-old playing great games on a big screen phone—at that moment I knew, the time had come. For enterprise, we went to Gartner IT conference, 30,000 US CIOs. We saw VMware virtualization, Splunk database booths jam-packed, showing enterprise demand. Saw AWS booth crowded from 9 to 11am—you can see demand accelerating before explosion.

Key: It’s ok to be late. Missing the first 1–3 years or 100% upside is ok—if the top is trillions, growth can last a long time. Peter Lynch said to me: “Erase the chart, only the future matters.”

S-curve slope (adoption speed) matters, too. We had a scholar study 100 years of S-curves. Radio’s S-curve was among fastest—7 years for 100% penetration. Dishwasher was much slower, needs backend. B2B usually slow, must connect to systems; consumer adoption is much faster. AI’s magic is that, whether consumer or enterprise, you just open browser—hence the vertical L-curve.

Host:
When several players compete in the same S-curve, how do you spot the ultimate winner?

Alex:
We research every company in the field, look for sustained competitive advantage. Many dislike tech stocks because future seems unpredictable, assets short-lived. But we found some digital-world advantages are even stronger than offline:Network effects: LinkedIn, Facebook.Industry standards: Oracle, Bloomberg. Huge DBA ecosystems and tuning software, forming lock-in.Rapid scale: With S-curve, Anthropic may reach $30B sales soon, Amazon got Walmart’s 40-year scale in 5 years.Key IP: Qualcomm (unavoidable for phones), ASML (lithography machines).Brand: Google, Amazon, Tesla almost never advertise.

AWS won in 2013, with 7-year lead, became platform/eco, scale 10x others, no one can match R&D.

For foundational models: Most of 50 companies disappeared, only 2–3 leaders remain. Why are they sustainable?Anthropic: Key IP in coding; strong enterprise brand (CIOs think first of Claude); reached escape velocity (10x sales growth); top code abilities feed models recursively, accelerating innovation.OpenAI: Strong consumer business, enterprise improving, coding tools accelerating.

In Internet era, leaders got bigger/faster and won. Exceptions (AOL to broadband, Netscape), but now consensus is, startups build apps atop these foundational models.

Host:
What does this mean for traditional software companies? You seem to have few large enterprise software stocks in your portfolio.

Alex:
Five years ago, maybe 40–50% invested in software. In April 2023, we thought software companies would win via AI API/data. But soon we found their AI products weren't good, couldn't drive growth or charge more. We sold almost all application software holdings, early this year even net short.

Traditional software faces:Drop in priority: CIOs prefer Anthropic tokens for faster ROI.Budget squeeze: AI spending crowds software budgets.Loss of pricing power: Annual price increases don't work any more.Employment impact: Companies freeze hiring, cut software seats.

Optimists say “no one will build their own ERP,” old tech is sticky (mobile games didn't kill consoles, tablets didn’t kill PC). But hard to imagine in 3–5 years, AI-native startups won’t challenge every software giant. Software’s “Rule of 40” (growth + profit margin = 40) is being challenged by AI’s “new rule of 40”: AI revenue share + market share. Traditional software’s AI revenue is only 1–2%—huge gap.

However, AI may reinforce existing software platforms. For example, Slack is the first thing you want to connect Claude—if it becomes key AI knowledge base, it’s locked in. Future AI agents may operate inside current enterprise software like people.

Host:
What about chips and hardware layer? You mentioned why this field is interesting now.

Alex:
For forty years, data centers barely changed—basically Intel x86. Workloads grew 25–40%/year, Moore’s Law about same, so hardware barely grew, whole supply chain commoditized.

But AI’s workloads are growing 10x/year, pushing every aspect of hardware to physical limits. This creates massive unit growth, which we call hardware de-commoditization.Servers: Old servers cost $5k—discarded if bad; AI servers worth $200–300k, liquid cooling, very hot, failure crashes system, now “critical infrastructure” like airplane parts—hard to replace once in.Memory (HBM): Used to be pure commodity, now HBM is 10-layer stacks, I/O boosted 10x, Samsung spent years.PCBs: Old servers need 10 layers, AI servers need 40—few suppliers can do.Networking: Used to upgrade 100G to 400G to 800G over 7-year cycles, now upgrades every year. Celestica, once a commodity OEM, retained IBM supercomputing legacy, now sole supplier of Google TPU servers, has 50–60% cloud Ethernet switch market share.Fiber: Corning holds huge fiber share, a Microsoft data center’s fiber length circles earth 4.5 times. Their fiber is thinner, more flexible, higher margin.Power supplies: Each Nvidia chip/rack uses 50–125% more power, boosting ASP for Delta and others.

The entire supply chain is tight, DRAM, NAND, PCB all 30% short.

Host:
You mentioned “AI revenue share” and “market share” as metrics—which matters more: absolute value or growth rate?

Alex:
Growth rate is very important. Going from 10% to 30% market share, your growth and profit margin both accelerate.

Host:
Given your framework (S-curve, competitive advantage, undervalued profit) has been proven over 25–30 years, why do most public investors still miss it?

Alex:
My mom asks me, why share the secret? Because it’s truly hard. Deep tech investing experience (we’ve done 20 years at Whale Rock), a team experienced in multiple cycles. Hardware and chips ignored for long, newbies fear them. Stock jumps, people fear heights so don’t buy. Nvidia surges for a year, flat for 6 months, they say “bubble will pop.” Many semi analysts missed it, never saw full picture of foundational model layers. You need big-picture vision, and to have studied dozens of S-curves across fields.

Host:
As bullish as you are on AI—what worries you most?

Alex:Negative public and government sentiment: Maine just banned data centers, only 20% optimistic about AI. Regulatory risk exists, but I think the genie is out of the bottle.Model progress slowdown: If models stall, open source catches up, becomes price war. Bad for model companies’ stocks, but actually good for chip firms (they just sell tokens no matter who wins).Top players lagging: If Meta or Oracle exit AI, reserved compute becomes an issue. But market is big enough, someone will fill the gap.

Host:
Why do you invest little in applications? Historically, apps' total market cap far exceeded infrastructure.

Alex:
Apps always come late. Three to four years after iPhone launched, the app ecosystem really took off. Risk is high now: boundaries between foundational models and apps unclear, hard for startup apps to build moats. Enterprise software giants (Salesforce) AI income share is still too low. Ecosystem isn’t clear yet, while chip/foundation layers are clear. There will be great app companies in future (e.g. Bret Taylor's Sierra), but this takes time, usually not in first 3–4 years.

Host:
You have a “research award wall” in your office for best research project of the year. In the AI era, what research lets you win this award?

Alex:
Our AI system is quite advanced, but hasn’t replaced analysts yet. We still do Philip Fisher’s "Scuttlebutt"—meet as many companies/managers as possible, talk to competitors, clients.

AI can help quickly learn new fields (e.g. ABF substrates), help draft earnings notes—high quality, but the top of notes must have "wisdom": what does it mean? How does it affect our investment themes? AI is an excellent “reporter," but can't predict future.

Our analysts' work on AppLovin: long-term tracking, meetings, chats, relationships—AI can’t do this yet.

Host:
What role does exchanging ideas with other investors play in your investing life?

Alex:
Philip Fisher also says: meet 10–15 similarly smart people nationwide, share ideas, build friendship. I call it a "tripod": I like it, my analyst likes it, and a respected fellow investor likes it—that greatly boosts my confidence.

Host:
How do you design different products for LPs? Any advice?

Alex:
For the first 15 years we only had a long/short fund. After 10 years, some asked for long-only, so we launched long-only in 2020 (now bigger than long/short). Around 2015, we considered private investing, gave LPs choice (15% or 25%), but only formalized in 2020. In 2021, launched hybrid fund (up to 80% private). Recently launched “Whale Rock Mega Cap Tech Fund”—invests in our chosen 12–13 stocks among top 30 global tech by market cap.

There’s structural under-allocation in big tech. Endowments and large LPs overweight private/international/small-cap, see large-caps as alpha-less. But to move Google/Nvidia requires 100 diversified fund managers changing views at once—while we can act ahead of 95% of them. Small-cap needs only one person, large-cap requires consensus shift.

Host:
If someone wants to understand Whale Rock, they should start with your research machine.

Alex:
Yes, we call it the “Whale Rock Learning Machine.” A team of ten highly experienced members. Like Buffett, we read books/blogs, but more importantly, talk to people in tech industry. Annually, we do 2,500–3,000 face meetings with management. We've compounding knowledge for 20 years, team is very stable (Andrew, Michael have been with me 19 & 18 years). The research engine supports both public/private investing; we don’t scan every opportunity, but once we see something fitting our system, we can move fast.

Host:
Final question: What’s the kindest thing anyone has done for you?

Alex:
Definitely my father. Cornell EE grad, moved to Wall Street, stellar Goldman Sachs career—corp finance in ‘80s, chairing private equity in ‘90s. Incredibly smart but humble, a true gentleman.

When I founded Whale Rock, he was the first investor. Then said, after 41 years at Goldman, why not join you, be the “gray hair”, do your compliance and chairman role while you keep doing your thing, I’ll help fundraising. We worked together six years until he passed in 2011. I was so lucky to work with him. He never raised his voice, mentored so many. After he died, I got countless letters on how he impacted, guided them. If I can be half the man he was, I’ll have totally won.

Host:
That’s moving. Alex, thank you very much for your time.

Alex:
Thank you.

Risk Disclaimer and Exemption ClauseThe market involves risk, investment should be cautious. This article does not constitute individual investment advice and does not consider any user's special investment goals, financial situation, or needs. Users should assess whether any opinions, views or conclusions in this article fit their specific situation. Investing accordingly is at your own responsibility.