Microsoft is discussing ways to supply Anthropic with its in-house artificial intelligence chip.

On May 21 local time, CNBC reported the two companies are negotiating Anthropic's use of Microsoft's customised AI chip, Maia.

If a deal is reached, it would be a meaningful case for Microsoft. Microsoft has been seen as lagging Amazon and Google in supplying special-purpose AI semiconductors to external customers. The company unveiled the second-generation Maia AI chip in January but has not yet offered it externally via Azure cloud.

A contract has not been finalised. A person familiar with the matter said Anthropic is discussing adopting Maia with Microsoft but has not reached a final agreement. Microsoft shares were little changed on the day.

The talks come as Anthropic's computing demand expands rapidly. Dario Amodei (대리오 아모데이), Anthropic's co-founder and CEO, said at an event earlier this month that the company is "struggling with compute." Demand for computing resources has increased further this year as use of Claude and the AI coding tool Claude Code has risen.

Anthropic has already outlined plans for large-scale infrastructure spending. SpaceX said in a May 20 filing that Anthropic plans to pay $1.25 billion per month through May 2029 in exchange for computing power. Anthropic has also relied heavily on Nvidia graphics processing units for training and inference and is widening its chip sourcing.

Anthropic said in April it would use Amazon Web Services' in-house Trainium chip under a contract worth more than $100 billion over 10 years. In October last year, it announced plans to also use Google's tensor processing units. If talks with Microsoft also result in a deal, Anthropic would further strengthen a structure in which it uses multiple Big Tech in-house chips alongside Nvidia's.

The two companies are already tied through a cloud contract. Microsoft said in November last year it would invest $5 billion in Anthropic, and Anthropic committed to spend $30 billion on Azure. Anthropic is also using cloud services from Amazon and Google.

Microsoft has also consistently stressed the competitiveness of its in-house chips. Satya Nadella (사티아 나델라), Microsoft's CEO, said in an earnings call in April that the Maia 200 chip has improved tokens processed per dollar by more than 30 percent compared with the latest semiconductors currently in operation. He said the chip is running at data centres in Arizona and Iowa.

In this flow, the talks are expected to be a test of whether Microsoft can link its in-house AI semiconductors to external customer demand. For Anthropic, how to secure computing resources stably to handle rapidly growing service demand remains a key task.