Maybe some inspiration from:
Patrick O'Shaughnessy is the CEO of Positive Sum.
Guest: Alex Sacerdote, Founder and Portfolio Manager of Whale Rock Capital Management
# Why the AI Boom Is Just Getting Started - Key Points by Chapter
Source podcast: Invest Like the Best, Episode 477 - Investing on the S-Curve
TIMESTAMPS
0:00 Intro
9:55 AI's L-Curve
19:31 Whale Rock's S-Curve Playbook
26:14 Spotting Inflection Points
32:02 Finding AI Winners
40:04 AI vs Software
48:13 The Hardware Renaissance
58:04 Why Investors Miss AI
1:05:18 Whale Rock's Research Machine
LLM / AI:
Note on probabilities:
The probabilities below are subjective estimates of whether each claim is directionally right over the next 2-4 years. They are not investment advice and they are not formal statistical confidence intervals.
## Short overview
Alex Sacerdote's central claim is not simply that AI will be big. His more specific claim is that AI is a new compute paradigm, with a new technology stack, new winners, and new bottlenecks.
Whale Rock's framework is to look for the intersection of three things: a powerful adoption curve, a durable competitive advantage, and underappreciated future earnings power.
Arguably, the strongest claims in the episode are:
- AI coding is the first major practical enterprise unlock.
- AI infrastructure and hardware are entering a new renaissance.
- Foundation models may become an oligopoly rather than a commodity market.
- Classic SaaS and application software face pressure from AI.
- The most durable investment opportunities appear where adoption, scarcity, and moats overlap.
## 0:00 - Intro, Anthropic, the AI stack, and coding as the first unlock
Sacerdote frames AI as a new compute paradigm. When ChatGPT launched in November 2022, Whale Rock began a broad research effort across the whole AI stack: power, chips, cloud, foundation models, and applications.
At first, the clearest investment area was chips and infrastructure, because every possible AI winner would need more compute. Over time, Sacerdote says the model layer began to look less like a pure commodity market and more like a small group of leading companies: OpenAI, Anthropic, and Google/Gemini.
The strongest specific case in this opening section is Anthropic. Sacerdote argues that Anthropic focused on enterprise use cases, built a strong position in coding, and developed a broader ecosystem around Claude, APIs, SDKs, orchestration, and related tools.
Probability estimates:
- AI coding as the first major enterprise unlock: 75-80%.
- Foundation model oligopoly: 55-65%.
- Anthropic maintaining a clear coding lead: 45-60%.
Short quotes to search in the transcript:
- "power at the bottom, chips at the bottom"
- "we want to be in the chips and the infrastructure first"
- "three-horse race"
- "the big kicker was code"
## 9:55 - AI's L-Curve
Sacerdote argues that normal AI usage today is still mostly AI 1.0: a better search engine, assistant, or productivity tool. The real enterprise transformation - agentic workflows, company-specific skills, bots, and AI agents working across systems - is still barely penetrated.
For this reason, he says AI is not behaving like a normal S-curve. He calls it an L-curve, meaning a very sharp move upward from a tiny base. He also stresses the compute bottleneck: if enterprise AI is still early and compute is already scarce, demand may remain intense.
Probability estimates:
- Real agentic enterprise AI adoption is still early: 70-80%.
- Compute scarcity remains material for several years: 65-80%.
Short quotes to search in the transcript:
- "AI 1.0"
- "search engine on steroids"
- "10 bips of the knowledge workers"
- "less than 1% penetrated"
- "not enough compute in the world"
## 19:31 - Whale Rock's S-Curve Playbook
Whale Rock's core investment framework has three parts:
- S-curve
- Competitive advantage
- Underappreciated earnings power
The argument is that when a company reaches the steep part of a technology adoption curve, and also has a strong business model, earnings can grow nonlinearly. Sacerdote says investors often underprice this because markets are focused on the next quarter or the next year, not on what the earnings base might look like 2-4 years out.
He also emphasizes that the key question is not only whether a technology is growing. The harder question is how high the S-curve can go. In other words: how large is the final market, what penetration is realistic, and when does the curve stop being exponential?
Probability estimates:
- S-curves are a useful framework for technology platform shifts: 80-90%.
- Investors can reliably forecast 2-4 years out in specific technology curves: 45-60%.
Short quotes to search in the transcript:
- "S-curve"
- "competitive advantage"
- "underappreciated earnings power"
- "earnings don't grow linearly"
- "the world doesn't think exponentially"
- "how tall, how big is this S-curve"
## 26:14 - Spotting Inflection Points
Sacerdote says inflection points are often not visible in clean financial data at first. They show up through field evidence, customer conversations, conference behavior, supplier checks, and anecdotal signals.
He gives examples from mobile gaming, Splunk, VMware, and AWS. In these cases, the key signal was not just revenue growth, but visible demand pressure: crowded rooms, urgent customer interest, and obvious changes in user behavior.
He also argues that investors do not need to catch the very first year of a major S-curve. If the final market is large enough, being late by the first 100% can still leave a very long runway.
Probability estimates:
- Field research beats pure financial data at early inflection points: 70-85%.
- AI adoption can be faster than cloud or SaaS because access is simpler: 60-75%.
Short quotes to search in the transcript:
- "you can't trust the data"
- "intuition, anecdotal evidence"
- "standing room only"
- "it's okay to be late"
## 32:02 - Finding AI Winners
Sacerdote stresses that a great S-curve is not enough. A company also needs a moat. He lists several types of competitive advantage: network effects, scale, industry standard status, platform status, critical IP, brand, and distribution.
In AI, he thinks frontier model companies may develop moats through scale, compute access, enterprise trust, coding performance, feedback loops, and the ecosystem around the API. He still admits AI is faster moving and more complex than prior S-curves, so the risk is higher.
Probability estimates:
- Frontier AI companies will have durable moats: 60-75%.
- Leader-takes-most dynamics will be strong in AI: 50-65%.
- Open source fully commoditizes frontier models: 25-45%.
Short quotes to search in the transcript:
- "very powerful competitive advantage"
- "network effect"
- "critical intellectual property"
- "the leader goes bigger, faster, and wins"
## 40:04 - AI vs Software
This is one of the most negative sections for classic software. Sacerdote says Whale Rock previously had a large software allocation, but sold most of it and was even net short software at one point.
The reasons are clear: incumbent software companies have not yet created AI products that materially move revenue, CIO budgets are shifting toward AI tokens and model usage, pricing power may weaken, and seat-based models could be hurt if companies freeze hiring or reduce headcount.
At the same time, Sacerdote does not say all software is doomed. Systems of record such as CRM, HR, Slack, Workday, and other deeply integrated platforms may become more important if AI agents operate through them.
Probability estimates:
- Traditional SaaS faces real AI pressure: 60-75%.
- Core ERP or CRM incumbents are rapidly replaced: 30-45%.
- Systems of record remain sticky and may benefit from agents: 65-80%.
Short quotes to search in the transcript:
- "AI products were not very good"
- "sold almost all of our software"
- "priority list of any CIO has fallen"
- "agents"
- "existing incumbent software tools"
## 48:13 - The Hardware Renaissance
This is probably the strongest and most concrete thesis in the episode. Sacerdote argues that AI is decommoditizing hardware. For decades, much of the data center hardware stack was relatively commoditized: servers, memory, networking, printed circuit boards, and contract manufacturing.
AI changes this because workloads push every layer of hardware toward physical limits. Sacerdote mentions high-bandwidth memory, liquid cooling, AI servers, Ethernet switching, printed circuit boards, optical fiber, copper and fiber interconnects, and power supplies.
The investment implication is that some suppliers may get higher unit growth, higher ASPs, better margins, and longer visibility than they had in the old commodity hardware cycle.
Probability estimates:
- AI hardware renaissance continues: 75-90%.
- Selected hardware suppliers remain decommoditized: 70-85%.
- Margins stay elevated for four years across the whole chain: 45-60%.
Short quotes to search in the transcript:
- "decommoditization of the hardware industry"
- "workloads are growing 10x every year"
- "high bandwidth memory"
- "shortages of everything"
## 58:04 - Why Investors Miss AI
Sacerdote says many investors miss AI because it is hard to think across the whole stack. A chip analyst may not understand model-layer demand. A software analyst may not understand infrastructure scarcity. A generalist may be scared by charts that have already moved up.
He also emphasizes that the bear cases are not stupid. Real risks include regulation, public negativity toward AI, model progress slowing, open-source models catching up, and one or more major compute buyers reducing spending.
Probability estimates:
- Investors will continue missing some cross-stack AI winners: 60-75%.
- AI bull-case risks are material: 50-65%.
- Full AI model race to zero: 25-45%.
Short quotes to search in the transcript:
- "rate of change is very important"
- "it's harder than it seems"
- "holistic view"
- "if AI sort of slows down"
## 1:05:18 - Whale Rock's Research Machine
The final section is about process. Sacerdote describes Whale Rock as a learning machine built around experienced analysts, thousands of meetings, management relationships, supplier checks, customer calls, competitors, and other investors.
AI helps the research process. It can summarize, write notes, review quarters, and help analysts get up to speed on complex areas. But Sacerdote argues that AI does not replace judgment. The human analyst still has to answer what changed, why it matters, and how it affects the investment thesis.
Probability estimates:
- AI strongly augments investment research: 80-90%.
- AI fully replaces top analysts or portfolio managers soon: 20-35%.
- Human judgment plus AI plus field research remains superior: 75-85%.
Short quotes to search in the transcript:
- "not yet"
- "supplanting the job of the analysts"
- "scuttlebutt approach"
- "AI can be a great reporter"
- "can't quite pick into the future"
# How the probabilities were estimated
The probabilities are calibrated subjective estimates. They are based on three inputs:
- Base prior from earlier technology adoption cycles.
- Current empirical evidence from public sources.
- Risk discount for counterarguments and uncertainty.
## 1) AI hardware renaissance continues - estimated probability: 75-90%
Simple calculation:
- Base prior for a real infrastructure cycle: 60%.
- Add 15-20 percentage points because Nvidia's Q1 FY2027 results showed Data Center revenue of 75.2 billion USD, up 92% year over year, and Data Center networking revenue up 199% year over year.
- Add 5 percentage points from transcript evidence on shortages, supplier visibility, and decommoditization.
- Subtract 5-10 percentage points for cyclicality, export controls, power constraints, and future overcapacity risk.
Central estimate: about 80%.
Source:
## 2) Coding is the first major AI enterprise unlock - estimated probability: 75-80%
Simple calculation:
- Base prior for developer tools being early AI adopters: 55%.
- Add 10 percentage points because Stack Overflow 2025 reports that 84% of respondents use or plan to use AI tools, and 51% of professional developers use them daily.
- Add 5-10 percentage points because a 2026 meta-analysis of 23 studies found a moderate positive productivity effect for generative AI coding tools.
- Subtract about 10 percentage points because METR's 2025 randomized trial found experienced open-source developers were 19% slower with early-2025 AI tools in mature codebases.
Central estimate: about 75%.
Sources:
https://survey.stackoverflow.co/2025/ai
https://arxiv.org/abs/2605.04779
https://arxiv.org/abs/2507.09089
## 3) Enterprise agentic AI is still early but rising fast - estimated probability: 70-80%
Simple calculation:
- Base prior: 55%.
- Add 10-15 percentage points because McKinsey 2025 reports that 23% of organizations are scaling agentic AI somewhere and 39% are experimenting.
- Add 5 percentage points because the Stanford AI Index 2026 reports continued frontier progress rather than a clear plateau.
- Subtract 5-10 percentage points because enterprise deployment barriers remain high: governance, security, verification, system integration, and trust.
Central estimate: about 75%.
Sources:
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
https://hai.stanford.edu/ai-index/2026-ai-index-report
## 4) Foundation models become an oligopoly - estimated probability: 55-65%
Simple calculation:
- Base prior from cloud, search, and social concentration: 55%.
- Add 10 percentage points for capital intensity, compute scale, enterprise trust, brand, and tooling ecosystems.
- Add 5 percentage points because the Stanford AI Index 2026 says industry produced over 90% of notable frontier models in 2025, which supports the idea that frontier AI is institutionally concentrated.
- Subtract 10-15 percentage points for open-source models, open-weight models, sovereign AI, regulation, and possible model commoditization.
Central estimate: about 60%.
Source:
https://hai.stanford.edu/ai-index/2026-ai-index-report
## 5) Classic SaaS is under pressure from AI - estimated probability: 60-75%
Simple calculation:
- Base prior: 50%.
- Add 10 percentage points because Sacerdote describes weak incumbent AI monetization, budget pressure, and Whale Rock's own decision to sell most software exposure.
- Add 5-10 percentage points because AI coding tools reduce the cost of building internal or AI-native software.
- Subtract 10 percentage points because enterprise systems of record are sticky, heavily integrated, and may become more useful if AI agents operate through them.
Central estimate: about 65%.
Primary source:
https://colossus.com/episode/investing-on-the-s-curve/
# Final takeaway
Sacerdote's core argument is that AI is not only a product wave. It is a new compute stack. The most attractive opportunities are likely to appear where three things overlap: rapid adoption, scarce infrastructure, and durable competitive advantage.
In simple terms:
- Chips and infrastructure are the clearest current beneficiaries.
- Coding is the first major AI use case with visible enterprise value.
- Foundation models may become a concentrated oligopoly, but this is less certain.
- Classic SaaS faces pressure, but deeply embedded systems of record may survive or even gain importance.
- AI will help research, but judgment, fieldwork, and relationships still matter.
Main source:
https://colossus.com/episode/investing-on-the-s-curve/
Additional source:
https://www.capitalallocators.com/podcast/riding-s-curves-at-whale-rock/
Additional source:
https://medium.com/graham-and-doddsville/alex-sacerdote-of-whale-rock-capital-ee46fcbfd8eb
Additional source:
https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-First-Quarter-Fiscal-2027/default.aspx
Additional source:
https://survey.stackoverflow.co/2025/ai
Additional source:
https://arxiv.org/abs/2605.04779
Additional source:
https://arxiv.org/abs/2507.09089
Additional source:
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Additional source:
https://hai.stanford.edu/ai-index/2026-ai-index-report
Invest Like the Best with Patrick O'Shaughnessy
Podcast:
Timestamps:
(00:00:00) Welcome to Invest Like The Best
(00:00:00) Welcome to Invest Like The Best
(00:02:29) Alex Sacerdote
(00:03:08) Anthropic: Highest Conviction Position
(00:13:23) Investing in Private Markets at Scale
(00:19:08) S-Curves: The Full Framework
(00:25:08) When to Buy Tech Companies
(00:30:20) Identifying the Leader from the Pack
(00:34:04) Anthropic & OpenAI's Competitive Moats
(00:37:31) AI's Threat to Enterprise Software
(00:43:18) Network Effects in the Agent Era
(00:44:22) The Hardware Renaissance: Chips & Infrastructure
(00:53:56) Why So Few Investors Get This Right
(00:55:36) Key Risks to the AI Bull Case
(00:57:47) The Application Layer
(00:59:40) How AI Is Changing Research at WhaleRock
(01:02:53) The Role of Investor Networks & Idea Sharing
(01:03:40) Building a Multi-Product Firm
(01:07:58) WhaleRock as a Learning Machine
(01:09:15) The Kindest Thing
Žádné komentáře:
Okomentovat