Microsoft’s big bet on custom artificial intelligence hardware just got a lot more complicated. The tech giant’s next-generation AI chip, known internally as Braga (or the Maia AI chip), won’t be ready for mass production until 2026, pushing back the original timeline by at least six months. This isn’t just a minor scheduling hiccup. Microsoft had hoped to start rolling out these chips in their data centers this year, which would have helped them cut costs and reduce their heavy dependence on Nvidia’s expensive graphics processors. Now they’re stuck paying those premium prices for longer while their competitors race ahead.
The problems started piling up fast. According to sources close to the project, the development team kept running into unexpected design challenges. Every time they tried to add new features, including some specifically requested by Microsoft’s AI partner OpenAI, the chip would become unstable during testing simulations.
But the technical issues weren’t the only headache. The project has been bleeding talent, with some teams losing up to 20% of their staff. The intense pressure and constantly shifting requirements apparently became too much for many engineers, leading to a brain drain that further slowed progress.
“It’s been a perfect storm of technical and organizational challenges,” explained one industry analyst familiar with the situation.
Microsoft’s AI Chip Ambitions Face Major Setback
This delay couldn’t have come at a worse time for Microsoft. The company had been banking on the Braga chip to help them break free from their expensive relationship with Nvidia, which currently dominates the AI hardware market. Instead, they’ll be writing even bigger checks to their competitor while waiting for their own solution.
Making matters worse, when the Braga chip finally does arrive, it’s expected to perform significantly worse than Nvidia’s latest Blackwell processor, which hit the market in late 2024. That performance gap could make it harder for Microsoft to justify the switch, even to itself.
While Microsoft struggles with its chip development, its cloud computing rivals are pulling ahead. Google has already launched its seventh-generation Tensor Processing Unit, and Amazon is preparing to release its Trainium3 AI chip later this year. Both companies seem to have figured out the custom chip game better than Microsoft.
This competitive pressure makes the delay even more painful. Cloud providers are in an arms race to offer the most powerful and cost-effective AI services, and having your own custom chips is becoming essential to winning that battle.
The Braga setback raises serious questions about Microsoft’s entire AI chip roadmap. The firm had been working on a number of processors, including the Braga-R and Clea chips, which were due to be launched in 2026 and 2027, respectively. If they fail to meet the deadline for the first chip, then what does that say for the remainder of the lineup?
The issue was compounded by the fact that Microsoft also ostensibly canceled a separate AI training chip project in early 2024. That action, combined with the current delay, suggests the firm might have to rethink its overall plan for bespoke hardware.
Can Microsoft Overcome Its AI Hardware Hurdles?
Microsoft’s failure proves how hard it is to make competitive AI chips. The business case is obviously cheaper, faster, and less reliant on suppliers, but bringing that to life has proven to be extremely hard.
The AI ecosystem changes extremely fast with new developments occurring constantly. When you have designed and developed a chip, the technology landscape may have shifted completely by the time. Microsoft is learning the hard way.
Yet Microsoft has not abandoned its goal of creating a custom chip. It keeps pouring money into designing its own processors, aware that proprietary hardware will be a significant driver of future success in AI and cloud computing.
The question now is whether Microsoft can learn from these mistakes and catch up before its competitors leave it too far behind in the custom chip game.
