Скачать или смотреть The New Economics of AI. Wei Zhou (SemiAnalysis) on Tokenomics, Caching, & the Future of RL Systems

📍 Recorded live at AI INFRA SUMMIT 4, Convene San Francisco

Tokenomics is reshaping how AI is built, deployed, and priced. In this fireside chat, Wei Zhou from SemiAnalysis joins Val Bercovici to break down the real economic forces behind AI infrastructure, why marginal cost is no longer near zero, and how memory bound inference is redefining the business models of AI native companies.

From the rise of prompt caching to the shift toward agent swarms and reinforcement learning, Wei explains the technical and economic transitions that will define 2025 and beyond. He also shares the vision behind InferenceMax, SemiAnalysis’ benchmarking suite, and what it will take to build a transparent industry standard for AI performance.

Highlights from the session:
Why software is entering a high marginal cost era driven by token generation
The layer cake of tokenomics across GPU providers, model builders, agents, and enterprise users
How prompt caching and KV cache hit rates influence total cost and user experience
Why agentic systems depend on memory optimization and long lived context
The economics of reinforcement learning and why RL is increasingly memory bound
Inside InferenceMax and what it takes to benchmark real world AI workloads

📣 Super early bird available — sign up for the next AI INFRA SUMMIT → https://luma.com/aiinfra5