Alex Sacerdote - How to Invest Through Technology Cycles - [Invest Like the Best, EP.477]

Summary

Alex Sacerdote says Whale Rock approaches technology by looking for S-curves, durable competitive advantages, and underappreciated earnings power, and he thinks AI is the clearest example of that framework today. His highest-conviction private position is Anthropic, which he views as one of a small set of foundational model leaders with real differentiation, especially in enterprise and coding. He argues the biggest commercial unlock is agentic code generation, and that AI demand is already reshaping the entire infrastructure stack, from HBM and networking to liquid cooling, PCBs, and fiber. Sacerdote is also cautious on incumbent software, saying many AI features are not yet monetizing well and that budgets are shifting toward foundation model usage. Across the conversation, he emphasizes that adoption is still early, compute supply is tight, and the speed of penetration matters more than static market-share snapshots.

Notes

• Whale Rock frames tech investing around identifying the right place in an adoption S-curve, then finding the strongest moat and the most underappreciated earnings power.
• He says new compute paradigms create a new stack, with chips, clouds, foundational models, and applications layered on top of each other.
• Anthropic is his highest-conviction private position.
• He describes the foundation-model market as narrowing from many entrants to a few leaders, with most startups falling away.
• He argues the leading-edge models matter disproportionately because a small benchmark step-up can unlock major use cases.
• Code is the key commercialization breakthrough for AI in his view.
• He thinks code alone is a huge market because token spend scales into meaningful annual software budgets across millions of coders.
• Anthropic was initially purchased at a 180 valuation, with upside implied toward a 9x valuation.
• Claude Code has become almost fully agentic, changing what users expect from AI coding tools.
• He contrasts cloud economics with AI model differentiation, saying clouds are mostly commodity infrastructure while models still vary materially.
• Anthropic is building a broader product ecosystem around the API, not just a model endpoint.
• He believes AI usage is still very early relative to the total knowledge-worker population.
• He argues the enterprise AI market is still less than 1% penetrated.
• Scarcity in compute is still central, and he says Anthropic only has about half the compute it needs.
• Sacerdote says his firm sold most software holdings because AI features were not yet good enough to monetize.
• He thinks software vendors face budget pressure because CIOs are prioritizing foundation-model spend where ROI is faster.
• He says many software businesses are only 1% to 2% penetrated on AI features.
• He sees the bear case for incumbents as AI turning them into databases while the interface moves elsewhere.
• AI may still fit inside incumbents if agents run on top of systems of record and preserve workflow control.
• He says AI is the most complex and fastest-changing S-curve he has studied.
• Anthropic’s enterprise brand and coding share help it reach escape velocity, in his view.
• He argues model gains can recursively improve the models themselves through code generation.
• His S-curve process starts by mapping the technology wave, then studying every participant to find the strongest moat.
• He lists network effects, standards, platforms, IP, brand, scale, and CAC/LTV as the main moat types in tech.
• He uses AWS as an example of a first-mover lead turning into an ecosystem and scale advantage.
• He says technology winners can look expensive by market standards, but still be cheap against their eventual earnings power.
• He uses EVs and smartphones to show that S-curves can accelerate, stall, or flatten at different penetration levels.
• He says it is acceptable to be late if the total S-curve is large enough.
• B2B adoption is slower than consumer adoption because enterprise tools must plug into existing systems.
• He says enterprise security and cultural friction can slow AI adoption.
• He thinks browser access is helping AI produce a sharper adoption curve.
• AI is reshaping data centers by pushing hardware to physical limits and decommoditizing once-stable parts of the stack.
• He says the data center was relatively static for decades before AI changed the hardware cycle.
• HBM is now a stacked, high-complexity bottleneck that requires multi-generation coordination with Nvidia.
• Celestica is his case study for a former commodity contract manufacturer finding advantage in AI servers and TPU supply.
• AI servers are much more expensive and hotter, making suppliers harder to replace.
• Ethernet upgrade cycles have compressed from seven years to annual cycles in AI infrastructure.
• Celestica’s cloud Ethernet switch share is very large, which he sees as proof of durable competitive advantage.
• AI server board design is more complex, with 40-layer PCBs replacing 10-layer designs.
• He says PCB volumes and margins are rising sharply as design cycles get locked in years ahead.
• Corning is benefiting from huge fiber requirements inside AI data centers.
• He expects scale-up networking to migrate from copper to fiber, expanding Corning’s opportunity.
• Power delivery is also a bottleneck, with Nvidia racks drawing much more power and lifting ASPs for suppliers like Delta and Advanced Energy.
• He says the firm is already short DRAM, NAND, and PCB supply by roughly 30%.
• He emphasizes rate of change over absolute share because accelerating penetration can drive both growth and margins.
• He thinks public-market investors are still structurally underweight the biggest tech winners.
• Whale Rock describes itself as a learning machine that relies on large volumes of face-to-face management meetings.
• He uses AI for speed and summarization, but not as a replacement for analyst judgment.
• He says AI can write better notes, but humans must supply the thesis and interpretation.