Microsoft CEO says company will keep buying Nvidia, AMD chips

Microsoft has begun deploying its first homegrown AI chips, named Maia 200, in its data centers, with plans to expand deployment soon.

Despite developing Maia 200, Microsoft CEO Satya Nadella said the company will continue purchasing chips from Nvidia and AMD due to ongoing supply challenges.

The chip is designed for AI inference tasks and claims to outperform recent chips from Amazon and Google.

Nadella emphasized that Microsoft’s approach involves both innovation and collaboration with existing vendors.

The Maia 200 will be used by Microsoft’s Superintelligence team to develop its own AI models, potentially reducing reliance on third-party providers like OpenAI and Anthropic.

The chip will also support OpenAI’s models on Microsoft’s Azure cloud platform.

Mustafa Suleyman, leading the Superintelligence team, said his team will be the first to use Maia 200 for frontier AI development, highlighting ongoing hardware access issues across the industry.

🔗 Source: TechCrunch

🧠 Food for thought

Implications, context, and why it matters.

Key details on Maia 200’s customer access and pricing remain unconfirmed

Initial internal use has been stressed for Microsoft’s Superintelligence team, the AI specialists building Microsoft’s own frontier models. Support has also been described for OpenAI models on Azure. Timing and location for broader Azure customer access to Maia 200 have been left unclear.
To judge Maia 200’s impact against Nvidia and AMD, more information still needs to be provided. The Azure virtual machine (VM) instance types that will run Maia 200 have not been named, the regions where those instances will launch have not been listed, and pricing has not been compared with Nvidia- or AMD-based options.

A new market opens for tools that simplify AI deployment across multiple chip vendors

Software tool providers may see more friction as Microsoft keeps buying AI chips from Nvidia and AMD while rolling out its own Maia 200. Azure customers could end up juggling workloads across different hardware, which raises portability headaches.
Companies that build vendor-agnostic optimization software can step into that gap. These tools help developers run AI models efficiently across chip platforms without writing or maintaining separate code for each vendor.