Anthropic in talks to use Microsoft's custom AI chip after $5 billion investment deal. Here is what we know

Microsoft is in active talks to supply Anthropic with its custom-built Maia artificial intelligence chip, CNBC has confirmed. No agreement has been finalised, according to a person familiar with the matter who requested anonymity to discuss internal negotiations. The Information first reported the discussions on Thursday. What Is the Microsoft Maia 200 Chip and Why It Matters Microsoft unveiled its second-generation Maia AI processor in January, positioning it as a competitive alternative to the custom silicon offered by cloud rivals Amazon and Google. The chip has not yet been made available to Azure customers in a commercial capacity. However, Microsoft chief executive Satya Nadella confirmed in April that it is already operational in the company's data centres in Arizona and Iowa. Nadella said on the company's April earnings call that the Maia 200 "offers over 30% improved tokens per dollar, compared to the latest silicon in our fleet." Microsoft has also confirmed that the processor will be used to run OpenAI's GPT-5.2 model. A deal with Anthropic would represent a meaningful commercial milestone for the chip, giving Microsoft a high-profile client and validating the Maia platform at scale. Why Anthropic Needs More Computing Power Anthropic's interest in additional chip supply reflects a broader and increasingly urgent challenge facing the company. Dario Amodei, Anthropic's co-founder and chief executive, acknowledged at an industry event earlier this month that the company has had "difficulties with compute." The strain has been building. Anthropic's Claude AI assistant and its Claude Code tool for AI-assisted programming have both grown considerably in popularity this year, placing greater demand on the company's existing computing infrastructure. That growth has made securing sufficient capacity a pressing operational priority. The scale of that need was thrown into relief on Wednesday, when SpaceX disclosed in a filing that Anthropic will pay $1.25 billion per month through May 2029 for computing power. Anthropic's Expanding Web of Cloud and Chip Arrangements Anthropic has historically relied on graphics processing units from NVIDIA to train and run its generative AI models, but the company has been steadily broadening its hardware partnerships. In April, Anthropic announced it would adopt Amazon Web Services' custom Trainium chips under a 10-year arrangement valued at more than $100 billion. The company also announced plans in October to use Google's tensor processing unit chips. Alongside those agreements, Anthropic continues to draw on cloud services from both Amazon and Google. Also Read | Anthropic Raising $30 Billion More as AI Labs Absorb Majority of VC Funding The potential Microsoft deal would add a fourth major chip relationship to that portfolio, underscoring how aggressively the company is moving to shore up its computational foundations. The $5 Billion Microsoft Investment and the $30 Billion Azure Commitment The talks over Maia chip access sit within an already substantial financial relationship. In November, Microsoft announced it would invest $5 billion in Anthropic. As part of that arrangement, Anthropic committed to spending $30 billion on Azure cloud services over the life of the agreement. Also Read | Anthropic to present Mythos AI cyber risk findings to global financial watchdogs A chip supply deal would therefore build on an existing partnership rather than establishing an entirely new one, giving Microsoft a deeper operational role in powering Anthropic's AI systems beyond simply hosting them on its cloud infrastructure. Microsoft Playing Catch-Up in Custom AI Silicon For Microsoft, the stakes of a deal extend beyond the revenue it would generate. The company currently trails both Amazon and Google in the market for custom AI silicon supplied to enterprise clients. A high-profile arrangement with one of the world's most closely watched AI companies would help address that gap and signal broader commercial confidence in the Maia platform.