
OpenAI Podcast Episode 8 - OpenAI x Broadcom and the future of compute
186 snips
Oct 13, 2025 Sam Altman, CEO of OpenAI, and Greg Brockman, Co-founder and President of OpenAI, delve into their exciting partnership with Broadcom. They explore the significance of custom chips designed for AI workloads, emphasizing the critical role compute plays in reaching artificial general intelligence. Altman highlights plans for massive compute scale, while Brockman discusses the optimization of infrastructure. The duo also reflects on historical advancements in chip design, underscoring the collaboration needed for the future of AI development.
AI Snips
Chapters
Transcript
Episode notes
Vertical Integration Drives Efficiency
- OpenAI and Broadcom built a custom chip and full system together to optimize AI workloads end-to-end.
- Vertical integration across transistor to token delivers large efficiency and performance gains.
Optimize The Whole Stack
- Optimize across the full stack from chips to racks to software to cut cost per token and increase throughput.
- Expect demand to grow faster than efficiency gains, so build significant incremental capacity.
AI Helped Design The Chip
- Greg described using AI to design chip components and finding optimizations that saved months of engineering time.
- He noted many model-suggested optimizations matched expert ideas but were discovered far faster.


