Microsoft Deploys Custom Maia 200 Chip To Reshape Cloud AI Economics

The Maia 200 deployment demonstrates that custom silicon has matured from experimental capability to production infrastructure at hyperscale.

Feb. 1, 2026 at 8:47am

Got story updates? Submit your updates here. ›

Microsoft has begun deploying its second-generation artificial intelligence processor, the Maia 200, in select data centers as part of a broader effort by major cloud providers to reduce their dependence on Nvidia hardware. The Maia 200 is an application-specific integrated circuit built exclusively for AI inference workloads and delivers approximately 10 petaflops of performance at four-bit precision.

Why it matters

The new chip matters for enterprise technology leaders because it reflects a fundamental shift in cloud computing strategy. As AI workloads grow and inference costs become a dominant line item for cloud customers, the hyperscalers building their own chips can offer price advantages that ripple through the entire ecosystem. For organizations planning AI deployments, understanding this hardware evolution helps inform decisions about platform selection and long-term cost structures.

The details

The Maia 200 contains 144 billion transistors manufactured on Taiwan Semiconductor Manufacturing Company's 3-nanometer process and is equipped with 216 gigabytes of high-bandwidth memory capable of transferring data at 7 terabytes per second. Microsoft claims the chip delivers three times the compute performance of Amazon's Trainium processors on certain benchmarks and exceeds Google's latest tensor processing unit on others, while achieving 30 percent better performance per dollar compared to competing accelerators.

  • Microsoft has begun deploying the Maia 200 in its data center in Des Moines, with plans to expand to facilities in Phoenix.
  • The company reports the chip powers workloads, including Microsoft 365 Copilot and OpenAI's GPT-5.2 models.

The players

Microsoft

A multinational technology company that has developed the Maia 200, its second-generation artificial intelligence processor, to reduce its dependence on Nvidia hardware.

Nvidia

A leading manufacturer of graphics processing units (GPUs) that has historically dominated the data center compute market within the hyperscaler tier, but is now facing increased competition from custom silicon developed by cloud providers.

Google

A cloud provider that pioneered the development of custom silicon with its tensor processing units nearly a decade ago and now operates seventh-generation chips that handle over 75 percent of its Gemini model computations.

Amazon Web Services (AWS)

A cloud provider that launched its own custom chip, Trainium, in 2022 and has expanded to third-generation chips deployed at scale for customers including Anthropic.

Taiwan Semiconductor Manufacturing Company (TSMC)

The world's largest contract chipmaker, which manufactured the 144 billion transistors in Microsoft's Maia 200 chip using its 3-nanometer process.

Got photos? Submit your photos here. ›

What they’re saying

“The Maia 200 deployment demonstrates that custom silicon has matured from experimental capability to production infrastructure at hyperscale.”

— Janakiram MSV (Forbes)

What’s next

Microsoft has indicated broader customer availability of the Maia 200 chip will come in the future, but has not specified timelines. Enterprise buyers cannot currently select Maia-based instances for their own applications.

The takeaway

The Maia 200 deployment reflects a fundamental shift in cloud computing strategy, as major cloud providers like Microsoft, Google, and Amazon develop their own custom silicon to reduce dependence on third-party hardware and offer price advantages to customers. This trend changes the competitive dynamics of cloud AI pricing and requires enterprise technology leaders to understand the hardware evolution to inform their platform selection and long-term cost structures.