CNBC verified on Thursday that Microsoft is in negotiations to provide Anthropic with its proprietary artificial intelligence processors.
Microsoft, which trails cloud competitors Amazon and Google in providing customers with special-purpose AI chips, might benefit from a partnership. Microsoft has not yet made its second-generation Maia AI chip accessible via its Azure cloud after announcing it in January. OpenAI’s GPT-5.2 model would run on the Maia 200 processor, according to the company.
This possible deal comes after Microsoft’s announcement that it will invest $5 billion in the AI business in which Anthropic agreed to spend $30 billion on Microsoft’s Azure cloud services under that initial agreement.
According to a person acquainted with the arrangement who wished to remain anonymous in order to discuss internal issues, Anthropic has not yet reached a contract with Microsoft about the use of the Maia. The talks that took place earlier on Thursday were covered by The Information.
The discussions centre on a core asset. Anthropic is exploring leasing server capacity that is powered by Microsoft’s second-generation Maia 200 custom artificial intelligence chip.
Regarding target workloads, the Maia 200 chip has a specific purpose. It is designed strictly for AI inference. This means it runs existing models like Anthropic’s Claude rather than training new ones.
There is also a notable efficiency boost. Microsoft claims the Maia 200 chip delivers over 30% improved tokens per dollar compared to previous hardware in its data centres.
As for the current status, negotiations are still in the early stages. No official contract or chip supply agreement has been finalized or signed.
Microsoft’s stock saw no movement, and the company said in November that it would invest $5 billion in Anthropic, while Anthropic promised to spend $30 billion on Azure. Additionally, Anthropic depends on Google and Amazon cloud services.
The CEO and co-founder of Anthropic, Dario Amodei, had stated during an event earlier this month that the company has experienced “difficulties with compute.”
This year, the popularity of its Claude assistant and Claude Code tool for AI-assisted programming has increased, making Anthropic’s demand for processing power even more urgent.
SpaceX had revealed on Wednesday that Anthropic will pay $1.25 billion a month for processing power through May 2029.
For the purpose of training and executing generative AI models, Anthropic has traditionally relied significantly on Nvidia graphics processing units. Anthropic said in April that it would use special Trainium chips from Amazon Web Services in a 10-year deal valued at over $100 billion. Google’s Tensor Processing Unit processors will be used by Anthropic, the company said in October.
Anthropic chose not to respond. A request for comment was not immediately answered by Microsoft.
CEO Satya Nadella stated during the company’s April earnings call that Microsoft’s Maia 200 “offers over 30% improved tokens per dollar, compared to the latest silicon in our fleet.”
The chips are currently operating in Microsoft data centres located in Iowa and Arizona, he said.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
