Multi-chip strategies diversify AI compute risk
The guest argued that a multi-chip and multi-CSP strategy is essential for frontier AI labs to mitigate supply risks and shift capital expenditures to operational expenses.
The argument
By partnering with multiple cloud providers like Microsoft, Oracle, and Google, and utilizing chips from NVIDIA, AMD, Cerebras, and Broadcom, OpenAI aims to maintain maximum flexibility. This approach allows them to scale compute dynamically while shifting heavy upfront CapEx into pay-as-you-go OpEx.
The thesis, stress-tested
✓ What validates it
- ✓Successful deployment of OpenAI's custom Broadcom chip in active clusters
- ✓First training runs completed on NVIDIA's Vera Rubin architecture
▸ Risks discussed
- ▸Transitioning to built-to-suit data centers (e.g., with SoftBank) will require higher upfront CapEx
- ▸Dependence on third-party cloud providers for financing and infrastructure scaling
Hear it yourself
AFFILIATE LINK · ZORTIX MAY EARN A COMMISSION · NEVER A RECOMMENDATION TO TRADE