According to CNBC, Microsoft is in talks with Anthropic regarding plans to supply its self-developed Maia AI chips. If finalized, this collaboration would mark a significant breakthrough for Microsoft in the AI chip sector, helping narrow the gap with Amazon and Google in custom AI chip ecosystems. In January of this year, Microsoft launched its second-generation Maia AI chip, the Maia 200, though it has not yet been officially made accessible to external customers via Azure cloud services. Previously, Microsoft stated that the Maia 200 had already been deployed to run OpenAI's GPT-5.2 model. During April's earnings call, Microsoft CEO Satya Nadella revealed that compared to the company’s existing chips, the Maia 200 delivers over a 30% improvement in "tokens per dollar" productivity, with related deployments now operational at data centers in Arizona and Iowa. However, sources familiar with the matter indicated that no formal agreement between Anthropic and Microsoft has yet been signed. Earlier reports by The Information first disclosed details about these negotiations. Over recent years, demand for computing power within Anthropic has surged dramatically. With rapid adoption of the Claude assistant and Claude Code programming tool earlier this year, reliance on training and inference resources has intensified. In early May, Anthropic co-founder and CEO Dario Amodei publicly acknowledged the company currently faces "compute constraints." Currently, Anthropic primarily relies on NVIDIA GPUs for training and running generative AI models while expanding partnerships with cloud providers. In April this year, Anthropic announced a decade-long commitment to adopt AWS' Trainium custom AI chips worth more than $100 billion. Last October, the company also declared intentions to utilize Google's TPU chips. Additionally, SpaceX recently disclosed that Anthropic will pay $1.25 billion monthly before May 2029 to secure computational resources. This further underscores the immense need for high-performance computing among leading AI companies competing in large-scale language model development.