Today, we announced our largest investment to date in our own silicon," said Satya Nadella, Chairman and CEO of Microsoft, on social media on October 12 local time, revealing Microsoft's latest plans for its custom-designed chips.
Microsoft announced that the Azure Maia 200 is now in production and has released the new Azure Maia 300 chip. The company also revealed its new deployment plan: starting in 2027, the Azure Maia series will be scaled to multi-gigawatt deployments, making Maia the primary inference system for Microsoft’s products and services.
Affected by this news, Microsoft's stock price rose 1.14% to $541.15, bringing its market capitalization back to $4 trillion.
According to Microsoft’s progress update, the Maia 200 is already in production, powering Microsoft workloads including Microsoft Copilot and supporting the MAI team in developing next-generation models. GitHub Copilot and Microsoft Foundry will also be operational by the end of the year. Across these workloads, Maia 200 runs OpenAI models, Microsoft’s proprietary MAI models, and leading open-source model families.
Microsoft states that the Maia 300 is designed for the new types of AI workloads emerging today, including long-context reasoning, latency-sensitive conversational experiences, and continuous multi-step agent tasks. Microsoft expects the Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems.
Microsoft has accelerated its development of custom chips. In November 2023, Microsoft launched the Maia 100, a chip built on TSMC’s 5-nanometer process, designed to power cloud-based training and inference for AI workloads such as OpenAI models, Bing, GitHub Copilot, and ChatGPT.
Two years later, in January this year, Microsoft announced the launch of its second-generation AI chip, the Maia 200. At the time, Microsoft stated that the Maia 200 was the most efficient inference system the company had ever deployed, offering 30% better performance per dollar compared to its then-most-recent hardware.
Less than nine months after releasing Maia 200, Microsoft has now launched Maia 300.
In a November interview last year, Satya Nadella mentioned the rationale behind Microsoft’s in-house chips. He stated that Microsoft plans to create a closed loop between its MAI models and chips, granting the company the right to design its own microarchitectures according to its needs and continuously update its models.
Regarding Microsoft’s announcement of its plan to deploy custom chips, outsiders believe it aims to reduce costs and decrease reliance on suppliers. Alongside the release of Maia 300, Microsoft also stated that its future will not be defined by a single model, a single chip, or a single innovation. Model and chip selection are inherent components of Microsoft’s vision for transformative advancement. Microsoft’s focus is on developing responsible, affordable, and scalable platforms.
In addition to Microsoft, companies such as Google, Amazon, Meta, and OpenAI are accelerating their efforts in developing proprietary chips. Amazon’s Q2 earnings report revealed that its annualized revenue from AI and chip businesses both exceed $25 billion. Regarding its chip business, Amazon disclosed that an increasing number of AI startups, including unicorns like NEURA Robotics and Odyssey, are adopting Trainium.