Anthropic is considering a diversification of its AI computational requirements — pushing it to look beyond players like Nvidia. The AI company has reportedly started discussions with software giant Microsoft to ink a chip deal. If finalized, Anthropic is likely to rent Microsoft’s in-house Maia 200 chip which is manufactured by the Taiwan Semiconductor Manufacturing Co.
The talks between Anthropic and Microsoft are still in the early phase, Reuters reported on Thursday citing sources familiar with the matter. A timeline by which more clarity on the deal may emerge remains unclear for now.
What has powered Anthropic so far?
Anthropic’s AI capabilities have been deriving computational power from a shiny lineup of chipsets from players including Nvidia and Amazon.
Anthropic relies heavily on the Nvidia H100 and H200 GPUs for running live model inference across major cloud platforms like AWS and Google Cloud. The chips are also utilized for the core daily training of its Claude LLM models.
Microsoft and Nvidia recently co-invested up to $15 billion into Anthropic. The deal entailed the allocation of large Nvidia server capacity for Claude operations.
Meanwhile, from Amazon, Anthropic rents the AWS Trainium chips that Anthropic uses for high-performance model trainings. Both the companies are also partners in running Project Rainer, which is a a massive, custom-built AI supercomputer and data center cluster developed by AWS in collaboration with Anthropic.
If Anthropic’s reported deal with Microsoft pulls through, the AI company could put it to use for the advantage for its chatbots, the Reuters report said.
Microsoft’s Maia 200 chip, launched in January 2026, is built on TSMC’s 3nm process. This allows it to power LLMs more cost-effectively and reduce the concentration of deals with heavyhitters like Nvidia.
The Maia 200 is loaded with Static Random-Access Memory.(SRAM), which has proven to give speed advantages for chatbots for other AI systems that face large numbers of users.
Insiders react
Microsoft could finally see profits from its chip-related endeavours as AI companies are still struggling with the shortage of AI compute, insiders from the tech industry have pointed out that
Shay Boloor, the Chief Market Strategist at market intelligence firm Futurum Equities said on X that Anthropic is definitely in need for compute to power the advancements it plans for Claude.
“This shows how severe the AI compute shortage still is with frontier labs needing every viable source of capacity they can get,” Boloor said. “OpenAI and Anthropic are locking up capacity years in advance because whoever controls compute controls how fast AI products, agents and inference revenue can scale.”
Insiders have reiterated that the demand for AI chips within the U.S. itself is far higher higher than the supply which has resulted in frontier labs stopping to care about “which chip wins” and starting to tap every viable stack they can get.
Siddhartha Saxena, the CTO of ThineAI joined the conversation on X and said, “Compute diversification is becoming mandatory at the frontier. Also, Maia is being positioned for inference efficiency, not replacing Nvidia training clusters.”
Whether or not the deal will see the light of the day remains a matter to wait and watch for now.