Microsoft (MS), long seen as a latecomer in the artificial intelligence (AI) chip race, is pushing a plan to supply its in-house AI chips to Anthropic, the developer of "Claude." If the transaction goes through, it would mark the first time MS provides its own semiconductors to an external AI model developer.
The Information, a U.S. information technology (IT) outlet, reported on the 21st (local time), citing multiple sources, that Anthropic is discussing a plan to rent servers based on MS-designed AI chips called "Maia." As usage of Claude surges and computing demand jumps, the company is moving to secure additional computing resources.
If the talks conclude, Anthropic will use MS's latest AI chip, "Maia 200," for inference on Claude models. Inference is the process by which an AI model generates answers to user questions, and the expense burden grows as service usage increases. In-house chips from cloud providers such as MS, Amazon, and Google are seen as more cost-efficient than Nvidia graphics processing units (GPUs) for some inference workloads.
If Anthropic adopts Maia 200, it would become the first among major AI model developers to use all three cloud providers' in-house AI chips—Amazon, MS, and Google. Anthropic has already outlined a strategy combining Amazon Web Services (AWS) Trainium, Google's TPU, and Nvidia GPUs.
MS unveiled Maia 200 in January this year as an AI accelerator for inference. The chip is built on TSMC's 3-nanometer process and focuses on reducing the expense of generating answers for large AI models. If the talks are finalized, MS would gain an opening to expand its in-house chip business—previously centered on internal services—to external customers.
However, the negotiations are still in the early stages, and it is uncertain whether a contract will be signed. OpenAI, Anthropic's competitor, is also pursuing development of its own AI chips, and MS, a major investor in OpenAI, is said to have already allocated part of its data center space for those chips.