← Home
Invest Like the Best · May 13, 2026 · 1h 16m

Krishna Rao - Anthropic's CFO on Compute, Scaling to $30B ARR, and the Returns to Frontier Intelligence - [Invest Like the Best, EP.471]

Krishna Rao, CFO of Anthropic, joins Invest Like the Best to discuss how he navigates compute procurement and allocation using a 'cone of uncertainty' framework, leveraging three chip platforms fungibly across Trainium, TPUs, and GPUs. He explains the daily meetings that allocate compute among model development, internal research, and customer demand, and why Anthropic believes the returns to building frontier intelligence are still increasing. The conversation also covers the blurry line between platform and application and why Anthropic builds its own products like Claude Code. Throughout, Rao shares insights from two years at one of the fastest-growing businesses in history.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Highlights

Cone of uncertainty in compute procurement
Krishna Rao describes using a 'cone of uncertainty' framework to plan compute purchases, acknowledging a wide range of possible future needs rather than single-point forecasts.
Fungible use of three chip platforms
Anthropic runs its workloads fungibly across Amazon Trainium, Google TPUs, and NVIDIA GPUs, switching between them based on cost, availability, and performance.
Daily compute allocation meetings
Anthropic holds daily meetings to decide how to split its compute capacity among model development, internal research, and serving customer demand.
Returns to frontier intelligence keep rising
Rao argues that the returns to building more capable models are increasing, not diminishing—especially for enterprise use cases.
Building products on your own platform
Anthropic deliberately builds its own applications like Claude Code on top of its models, blurring the traditional line between platform and application.
Was this useful?