Microsoft Unveils Maia 300 Chip, Betting on Its Own Silicon Over Nvidia

Microsoft (MSFT) just told the market it plans to lean far harder on its own AI chips. On Monday it unveiled Maia 300, its third-generation AI accelerator, and said it will mainly use its own silicon to handle incoming AI requests.

That's a real shift for a company whose AI business was built largely on Nvidia GPUs. As CNBC reported Monday, Microsoft is ramping up homegrown chips to bolster its AI stack, which spans the Azure cloud and the Copilot products.

Microsoft's own announcement is blunt about the scale. The company said it will grow the Maia family to multi-gigawatt deployments starting in 2027, making Maia the primary inference system for its products and services.

Timing is worth a close look. Maia 200 is already in production, while Maia 300 production starts in 2027. Alistair Speirs, a Microsoft general manager, wouldn't give a firm deployment timeline and said the company wants to "reserve a little bit of optionality" in a fast-changing industry.

The hardware numbers are the headline. Microsoft plans to pack 96 Maia 300 chips into each data center rack, double the Maia 200 count. It says a rack can run up to seven times more AI tokens, with 40% more tokens per megawatt than GPUs.

Reuters added useful context on power. Its report said Maia 300 racks will draw about 167 kilowatts, against roughly 142 kilowatts for Nvidia's latest GB300 racks, which hold 72 chips. More chips per rack, a bit more power, and Microsoft is betting the output per megawatt wins out.

Why would Microsoft bother? Cost. Finance chief Amy Hood said at a Goldman Sachs conference last month that using its own chips has helped drive "margin improvement over the course of a decade."

CEO Satya Nadella added in July that Maia 200 delivers 30% better performance per dollar than the latest hardware in Microsoft's fleet.

"We need new architectures that will give us essentially better performance, better use of the power that we have." — Alistair Speirs, General Manager, Microsoft

Wall Street is already sizing it up. In a client note last Wednesday, Evercore ISI analysts estimated Maia will make up about 10% of Microsoft's AI chip spending by the first quarter of 2027. They recommend buying the stock.

For MSFT holders, the thesis is simple. AI infrastructure is the biggest expense line in the story, and every workload that moves from rented-by-the-unit GPUs to in-house chips is a chance to protect margins. That's a slow-burn tailwind, not a one-quarter event.

Nvidia holders shouldn't panic. A 10% share of chip spending is a dent, not a collapse, and Microsoft still partners with Nvidia and AMD on model training. Jensen Huang appeared at a Microsoft event just last week, where a Surface laptop built around an Nvidia processor was announced.

Now the risks. Unlike Google's TPUs or Amazon's Trainium chips, Maia can't be rented by the hour to other companies. "Nobody has tested them outside of Microsoft yet," said Patrick Moorhead, CEO of Moor Insights and Strategy.

Microsoft's own AI models also haven't made a big splash against OpenAI, Anthropic and Google, so the chip story carries a lot of weight.

So what do you do with this? There's no clean price level here, because the payoff plays out over years. Watch Microsoft's next earnings call for any change in gross margin or capital spending commentary tied to Maia.

Also watch for two signals: whether Microsoft ever opens Maia to outside customers, and whether Evercore's 10% estimate creeps higher. Nvidia owners should track how much of Microsoft's inference work actually migrates.

Bottom line: Microsoft is turning its chip bill into a margin lever, and the next few earnings calls will show whether it's working.