From Startup Fortune · · 1m 10s
Microsoft Runs Maia Chips in Production to Cut Nvidia Costs
Microsoft's Maia 200 chip cuts Nvidia's AI costs by 30% in Azure production.
In brief
Microsoft's Maia 200 chip cuts Nvidia's AI costs by 30% in Azure production. Microsoft's custom Maia 200 chip reduces Azure inference costs by 30% compared to Nvidia hardware. Originally reported by Startup Fortune.
Microsoft's Maia 200 in Production
Microsoft's Maia 200 chip runs AI workloads in production including Copilot and OpenAI models.
“Microsoft now says Maia 200 is serving production workloads, including Microsoft Copilot, and is running OpenAI models, Microsoft's own MAI models and open-source model families inside Azure”
Maia 200 Technical Specs
Built on TSMC's 3-nanometer process with 140 billion transistors and 216gb hbm3e memory.
“Built on TSMC's 3-nanometer process, Maia 200 packs more than 140 billion transistors, delivers over 10 petaflops at 4-bit precision, and carries 216gb of hbm3e memory with 7 terabytes per second of bandwidth”
Cost Advantage Over Nvidia
Maia 200 offers 30% better performance per dollar and 40% better performance per watt for MAI models.
“Microsoft says that adds up to more than 30% better total cost of ownership than the latest-generation hardware in its fleet, and later told investors Maia 200 offers 30% better performance per dollar while delivering 40% better performance per watt when running MAI models”
Nvidia's Margin Pressure
Nvidia's 73-75% gross margin drives hyperscalers to build custom AI chips.
“Nvidia's margins are the reason everyone is racing to build their own silicon”
Nvidia's Customer Dependence
Four direct customers account for over 61% of Nvidia's revenue with no identified names.
“as of Nvidia's fiscal 2026 third quarter, four direct customers each accounted for more than 10% of Nvidia's revenue, together making up roughly 61% of it, up from about 36% a year earlier”
Maia 300 Deployment Timeline
Microsoft plans multi-gigawatt Maia deployments starting in 2027 with Maia 300 as primary AI system.
“Microsoft announced Maia 300 on October 12, 2026, and said it will scale Azure Maia to multi-gigawatt deployments starting in 2027, making Maia 300 the primary AI inference system for Microsoft products and services”
Competitive Positioning
Maia 200 beats Amazon's Trainium and Google's TPU on specific AI workloads.
“Maia 200 beats Amazon's third-generation Trainium on FP4 throughput by roughly three times and edges out Google's seventh-generation TPU on FP8 workloads”
Market Shift in AI Infrastructure
Custom AI accelerator shipments projected to grow 44.6% in 2026 versus 16.1% for merchant GPUs.
“Custom AI accelerator shipments are projected to grow 44.6% in 2026, against 16.1% for merchant GPUs”
Here's the gist
Microsoft's custom Maia 200 chip reduces Azure inference costs by 30% compared to Nvidia hardware.
