Microsoft's in-house AI chip delayed again with crappy performance, can't compete with NVIDIA

Microsoft has been working on its in-house AI chip for a while now, but more and more hurdles are getting in its way with new delays of up to 6 months or more in new reports.

In a new article from The Information, we're hearing that Microsoft's ambitious in-house AI chip efforts have been delayed by up to 6 months, with initial performance evaluations not looking great at all. Microsoft's in-house AI chip is codenamed "Braga" and is now aiming for mass production in 2026, and that the performance of the AI chip is lower than NVIDIA's current Blackwell AI chips, let alone their next-gen Rubin AI chips powered by HBM4 memory.

Microsoft has had mammoth R&D costs and chip design revisions that have forced the company to push out its production timelines by 6 months, with its main reasons to make an in-house AI chip is to reduce its dependency on NVIDIA AI GPU hardware. NVIDIA AI GPUs are in gigantic demand, where it's selling as many as it (well, TSMC) can make.

Microsoft isn't the only one so reliant on NVIDIA hardware, as big US tech giants like Amazon and Google are reducing their reliance on NVIDIA AI GPUs by making their own in-house AI chips in their data centers (as well as using NVIDIA AI GPUs in their data centers).

NVIDIA CEO Jensen Huang recently acknowledged that the fact that the competition (Amazon, Google, Microsoft, and other companies) are making custom AI chips, but he also asked why bother creating custom AI chips if NVIDIA makes the best of the best (and he's right).