Anthropic is in early discussions to use Microsoft's custom Maia 200 artificial intelligence chips, though no deal has been finalised according to CNBC, citing a person familiar with the matter who asked not to be named.
The talks come as CEO Dario Amodei acknowledged earlier this month that the company has had "difficulties with compute" - a candid admission that reflects surging demand for Claude and Claude Code through the first half of 2026.
Why Maia matters to Microsoft
For Microsoft, landing Anthropic as a customer would be a meaningful step forward for the Maia platform, announced in January but still unavailable to Azure clients, while rivals Amazon and Google have moved faster to place custom silicon with external AI developers.
On the company's April earnings call, Satya Nadella said the Maia 200 delivers more than 30% better tokens per dollar than other silicon in Microsoft's fleet, with deployments already running across data centres in Arizona and Iowa.
The potential Maia arrangement would add another layer to a compute strategy that has expanded rapidly across multiple providers.
In April, Anthropic signed a 10-year AWS Trainium deal valued at more than US$100 billion (A$140.3 billion), building on an earlier commitment to run workloads across Google's tensor processing units.
SpaceX disclosed on Wednesday that Anthropic will pay $1.25 billion per month for computing capacity through May 2029 - a figure that frames the scale of infrastructure spending now required to compete at the frontier.
The two companies deepened their financial relationship in November 2025, when Microsoft committed a $5 billion investment in Anthropic alongside a pledge by the startup to direct $30 billion in spending toward Azure.
Beyond additional capacity, a Maia rental arrangement would also allow Anthropic to shape future chip iterations around its model architecture - an influence typically reserved for the largest cloud spenders.
