OpenAI says it won't use Google AI chips for now

OpenAI announced it has no current plans to use Google’s tensor processing units (TPUs) for its operations.

This follows reports of the AI lab testing Google’s chips to address rising demand.

A spokesperson confirmed the company is in the early stages of testing but does not intend to deploy them widely at this time.

Google did not provide a comment.

Currently, OpenAI uses Nvidia’s GPUs and AMD’s AI chips to support its systems. It is also developing its own chip, with plans to finalize its design later this year.

While OpenAI has partnered with Google Cloud to boost computing capacity, most of its power is expected to come from GPU servers run by CoreWeave.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ Strategic chip diversification is the new normal in AI

OpenAI’s approach to chips reflects a broader trend of AI companies managing supply chain risk through diversification while developing proprietary solutions.

The company currently relies on Nvidia GPUs (which control approximately 80% of the AI accelerator market) while also utilizing AMD chips to power its growing computational demands 1.

Simultaneously, OpenAI is finalizing its first custom AI chip design expected to reach “tape-out” this year, with production utilizing TSMC’s advanced 3-nanometer technology planned for 2026 2.

This multi-pronged strategy, using market-leading chips while developing in-house alternatives, mirrors approaches taken by other tech giants like Google and Meta who maintain relationships with commercial chip providers while investing in custom silicon.

This balanced approach helps explain why despite signing up for Google Cloud services earlier this month, OpenAI isn’t rushing to adopt Google’s TPUs at scale. Maintaining flexibility remains the priority.

2️⃣ Custom silicon development reflects shifting AI economics

OpenAI’s pursuit of custom chips highlights how the economics of AI infrastructure are driving vertical integration across the industry.