Microsoft Built Its Own AI Chip and Plans to Run Its Office Suite on It

Key points

• Microsoft announced the Maia 300, a custom silicon chip built to run large AI models at scale inside Azure data centres, in October 2026.
• Microsoft engineered the chip alongside the servers and racks that house it, not as a standalone component.
• Microsoft AI CEO Mustafa Suleyman described the approach as "chip-model co-design for frontier superintelligence."
• Microsoft 365, the office suite used by hundreds of millions of workers worldwide, will be among the first applications the chip powers.
• No independent benchmark figures exist yet; every performance claim so far comes from Microsoft's own executives.

Microsoft has built a chip. Not bought one, not leased capacity on someone else's silicon: built its own, tuned specifically for the AI workloads running inside Azure.

The chip is called the Maia 300. Announced in October 2026, it sits at the centre of what Microsoft calls a "production-scale, full-stack AI advantage." That phrase sounds like marketing, but the technical reality is concrete. Instead of designing a chip and then figuring out where to put it, Microsoft engineered the Maia 300 together with the server chassis and the rack system around it, the way you'd design a car engine and the body simultaneously so neither limits the other.

Every AI answer you get from a Microsoft product costs the company money in computing time. A chip shaped to match Microsoft's own AI models runs those models more efficiently, which means faster responses and lower costs that can be passed on through pricing or reinvested in more capable products.

What will Maia 300 actually power?

Microsoft 365 is one of the first destinations. Jim Kleewein, a Technical Fellow at Microsoft, specifically named Maia as the engine behind AI features in Microsoft 365. Practically, that means the Copilot assistant inside your Word document or your Teams meeting summary could soon run on Microsoft's own silicon rather than chips it buys from others. We've covered Microsoft 365 seven times in the past 90 days; the Maia announcement is the first that ties the chip directly to the product layer most users actually touch.

Mustafa Suleyman, CEO of Microsoft AI, framed the ambition larger, describing Maia 300 as part of "chip-model co-design for frontier superintelligence." Co-design means the chip and the AI model are shaped around each other from the start, a bit like a glove shaped around a hand. That approach tends to produce better efficiency than fitting a general-purpose chip to a specialised model. We first covered Suleyman's thinking on AI architecture on 20 September, when he warned that AI setting its own goals could end up competing with humans for resources.

Should this worry rivals like Nvidia and AMD?

It should get their attention. Microsoft deploying its own silicon at scale means it needs to buy fewer of the expensive GPUs (the specialised processors that handle AI's heavy computation) that Nvidia and AMD sell. This is the clearest signal yet that Microsoft intends to control its AI infrastructure from the chip upward.

The competitive framing is sharp. Google has its TPU line. Amazon has Trainium. Now Microsoft has Maia 300, and it isn't a prototype: the announcement came with server and rack renders showing a production system.

The honest caveat is one Microsoft didn't address publicly. There are no independent benchmark figures yet, and every performance claim comes from the company's own executives. Those numbers need external validation before anyone can assess how Maia 300 actually stacks up. Watch for those tests.

What matters most here is the vertical integration play, not the chip itself. Microsoft is betting that owning the full stack from silicon to software gives it a cost and speed advantage that's hard to match by assembling commodity parts. Whether Maia 300 delivers on that bet is a question the benchmarks will answer, not the launch event.