Anthropic and OpenAI Diversify AI Compute Beyond NVIDIA

@nathanbenaich· August 1, 2026 View original
Anthropic and OpenAI Diversify AI Compute Beyond NVIDIA

Key takeaways

  • Major AI labs are actively diversifying their compute infrastructure.
  • AMD is gaining significant traction as an alternative to NVIDIA for AI workloads.
  • This trend indicates a push for supply chain resilience and competitive pricing.
  • The AI hardware market is becoming more competitive with multiple key players.

Who benefits

TechnologyCloud ComputingAI DevelopmentSemiconductor ManufacturingData Centers

Summary

The latest State of AI compute index reveals that Anthropic has significantly increased its use of AMD MI450s, making non-NVIDIA silicon account for 7 out of its 8 gigawatts of contracted compute. OpenAI also shows substantial diversification, with 16.75 out of 26.75 gigawatts of its compute coming from non-NVIDIA providers like AMD, Broadcom, and Cerebras.

Recent updates to the State of AI compute index highlight a notable trend among leading AI companies: a strategic move towards diversifying their hardware infrastructure beyond NVIDIA. Anthropic, for instance, has reportedly integrated up to 2 gigawatts of AMD MI450 processors, resulting in 7 out of its total 8 gigawatts of contracted compute now being sourced from non-NVIDIA suppliers. Similarly, OpenAI is also expanding its compute partnerships. The company's current infrastructure indicates that 16.75 gigawatts of its 26.75 gigawatts of compute capacity are now powered by a mix of non-NVIDIA vendors, including AMD, Broadcom, and Cerebras. This shift suggests a broader industry effort to reduce reliance on a single chip manufacturer and potentially optimize for cost, performance, or supply chain resilience.

Why it matters

This diversification in AI compute infrastructure signals a maturing market and potential shifts in competitive dynamics among hardware providers. Professionals in AI engineering and strategy should be aware of these trends to make informed decisions about future hardware investments and partnerships.

How to implement this in your domain

  1. 1Evaluate alternative AI hardware providers like AMD, Broadcom, and Cerebras for specific workload needs.
  2. 2Assess the cost-effectiveness and performance benefits of diversifying compute infrastructure.
  3. 3Develop a multi-vendor strategy for AI hardware procurement to mitigate supply chain risks.
  4. 4Stay informed on emerging chip architectures and their potential impact on AI development.

Original post by @nathanbenaich

"i refreshed the @stateofai compute index (cutoff 1 aug 26): - anthropic added up to 2gw of amd mi450s, which makes non-nvidia silicon now 7 of its 8gw of contracted compute - openai: 16.75 of 26.75gw is non-nvidia across amd, broadcom and cerebras read more:"

View on X

Originally posted by @nathanbenaich on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses