OpenAI’s Compute Chief: We Can’t Build Fast Enough | Sachin Katti

The AI industry is currently undergoing the largest infrastructure build-out in history, driven by an insatiable demand for compute that consistently outstrips supply. As Head of Industrial Compute at OpenAI, Sachin Katti highlights the physical reality of this expansion, where data centers function as massive factories converting power into intelligence. Key challenges include scaling liquid cooling systems to manage intense thermal loads and securing reliable energy sources, such as nuclear and grid-integrated power, to support massive GPU clusters. OpenAI’s strategy employs a portfolio approach, blending partnerships with hyperscalers and custom silicon development like the "Jalapeno" chip to maximize tokens per watt. Furthermore, AI is increasingly assisting in its own infrastructure design, including chip architecture and network routing protocols like MRC, which ensure reliability across massive GPU clusters by masking hardware failures in real-time.