"Today we announce the largest bet to date on our own-developed chips." On October 12th local time, Satya Nadella, Chairman and CEO of Microsoft ( Satya Nadella ), revealed the latest plans for Microsoft's self-developed chips on social media.
Microsoft stated that the Azure Maia 200 has now been put into production, and they have also released the latest chip, the Azure Maia 300. At the same time, Microsoft announced a new deployment plan: starting from 2027, they will expand the Azure Maia series to a scale of several gigawatts, making the Maia the main reasoning system for Microsoft's products and services.
Affected by this news, Microsoft's stock price rose 1.14% to $541.15 on that day, and its market value returned to $4 trillion.
According to the progress announced by Microsoft, Maia 200 is currently in use, serving Microsoft's production workloads, including Microsoft Copilot, and providing support for the MAI team in developing the next generation of models. GitHub Copilot and Microsoft Foundry are also expected to be put into use by the end of the year. Within these workloads, Maia 200 runs OpenAI models, Microsoft's own MAI models, as well as a leading series of open-source models.
Microsoft stated that Maia 300 is specifically designed for the new types of AI workloads that have emerged today, including long-context reasoning, latency-sensitive dialogue experiences, and continuous multi-step agent tasks. Microsoft expects that, compared to leading systems based on GPU, Maia 300 can produce 40% more token (tokens) per megawatt.
Microsoft has accelerated the development of its own chips. In November 2023, Microsoft released Maia 100, which uses TSMC's 5-nanometer manufacturing process and is designed to support cloud-based training and inference for artificial intelligence workloads such as OpenAI, Bing, GitHub Copilot, and ChatGPT.
Two years later, in January of this year, Microsoft announced the launch of its second-generation artificial intelligence chip Maia 200. At that time, Microsoft stated that Maia 200 was the most efficient inference system the company had deployed to date, with its performance per dollar improving by 30% compared to Microsoft's latest generation of hardware at that time.
Less than 9 months since Microsoft released Maia, Microsoft has once again released Maia 300.
In an interview last November, Satya Nadella mentioned the logic behind developing in-house chips. He stated that Microsoft plans to establish a closed loop between its own MAI models and chips, granting the company the right to develop its own chips independently. This means designing microarchitectures according to its own needs and then continuously updating its models.
Regarding Microsoft's announced plan to deploy self-developed chips, the outside world believes it is aimed at reducing costs and decreasing dependence on suppliers. On the occasion of the release of Maia 300, Microsoft also stated that in the future, it will not be defined by a single model, a single chip, or a single innovation. The choice of models and chips is an inherent part of Microsoft's vision for forward-looking transformation. Microsoft's focus is on developing responsible, affordable, and scalable platforms.
In addition to Microsoft, companies such as Google, Amazon, Meta, and OpenAI are also accelerating their efforts in developing their own chips. Amazon's second-quarter financial report shows that the annual revenue from its artificial intelligence and chip business exceeded $25 billion. In terms of the chip business, Amazon revealed that an increasing number of AI startups are adopting Trainium, including unicorn companies such as NEURA Robotics and Odyssey.
