Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia
Microsoft is betting its own silicon. MAIA chips moving from training to inference means Nvidia just lost a major customer—and enterprises get a third option.
Why it matters
Microsoft is shifting inference workloads from Nvidia GPUs to its homegrown MAIA chips, reducing compute costs and vendor lock-in while establishing itself as a credible silicon alternative. For enterprises, this signals a real diversification path; for Nvidia, it marks the start of a margin squeeze on inference, the larger compute volume.
The key facts
4 to knowMicrosoft deploying MAIA chips for AI inference at scale, reducing Nvidia GPU dependence
Margin expansion and AI stack control are stated drivers
Inference (not training) is the focus—the higher-volume, longer-tail workload
Story key exists: microsoft-maia-300-inference-gigawatt-scale-2027 — MAIA deployment is active and tracked
Go to the source
CNBC Technologycnbc.com
Publisher excerpt: By increasingly counting on its own chips, Microsoft stands to expand margins and bolster its AI stack.