Maia 200 specs
Microsoft · Shipping now · first available Jan 2026. Dense compute, memory, bandwidth and power from the vendor's documents; each precision stays separate.
All accelerators · Compare with Trainium3
Specifications
| Dense BF16 · FP32 accumulate | Not published |
|---|---|
| Dense FP8 · FP32 accumulate | 5,072 TFLOPS |
| Dense FP4 · FP32 accumulate | 10,145 TFLOPS |
| Memory | 216 GB · HBM |
| Memory bandwidth | 7 TB/s |
| Maximum chip power | 750 W |
| Interconnect | Ethernet scale-up |
| Process node | 3nm |
| First availability | Jan 2026 |
| Availability | Shipping |
Microsoft’s inference chip, live in Azure Jan 2026; serves GPT-5.x-class models at claimed 30% better perf/$. Its 1,268 TFLOPS BF16 peak runs on the vector unit (the tensor unit handles FP8/FP6/FP4), so no tensor BF16 rate is listed.
Dense tensor rates with FP32 accumulation, never sparse or FP16-accumulate headline figures. Memory is in decimal GB. The est marker flags a value its vendor-document review did not confirm.
Rent it
No tracked rental quote. The site tracks providers' published list prices; this part may still be rentable under terms it does not record.
Value ratios
| $ per dense BF16 PFLOP-hour | No dense BF16 rate |
|---|---|
| Dense BF16 TFLOPS per watt | No dense BF16 rate |
| Bandwidth bytes per 1,000 BF16 FLOP | No dense BF16 rate |
| $ per dense FP8 PFLOP-hour | No on-demand quote |
| Dense FP8 TFLOPS per watt | 6.76 |
| Bandwidth bytes per 1,000 FP8 FLOP | 1.38 |
| $ per GB of memory per hour | No on-demand quote |
Prices divide the cheapest on-demand list price; theoretical ratios do not measure application speed.
Sources
Specifications reviewed . Peak-specification table: 5,072 TFLOPS FP8, 10,145 TFLOPS FP4, 750 W. BF16 (1,268 TFLOPS) is a vector-unit rate, so tf16 stays empty.
First availability: Microsoft: Maia 200 ↗ (2026-01-26): “Maia 200 is deployed in our US Central datacenter region”