01CGPU cloud · comparison· quotes observed 2026-10-11 to 2026-10-14 · specs reviewed 2026-10-15
H200 vs B200 rental price and specs
H200 and B200 side by side: the same provider-class price for each, price per dense PFLOP-hour at one precision at a time, and their spec sheets. Every price carries its venues and observation dates.
Rental price by provider class, US$ per GPU-hour
Each row compares one class aggregate with the same aggregate for the other chip — never one class or billing model with another. The on-demand rows use provider list prices only; the marketplace row is variable-availability asks.
| Class aggregate | H200 | B200 | B200 vs H200 |
|---|---|---|---|
| Typical — median of specialist clouds | $5.35median of 6 · $3.99–$6.31 · observed to | $8.09median of 6 · $6.00–$8.60 · observed to | +51% |
| Big-cloud list price | $7.91AWS (p5en) · lowest of 3 · observed | $14.00Oracle OCI · lowest of 2 · observed | +77% |
| Marketplace floor | $3.68Vast.ai · the only quote · observed | $6.25Vast.ai · the only quote · observed | +70% |
Price per dense PFLOP-hour, one precision at a time
Each chip's cheapest specialist on-demand list price divided by its dense (non-sparse) tensor rate at the same precision. A precision a spec sheet does not list is left blank, never converted from another.
| Precision | H200 | B200 | B200 ÷ H200 |
|---|---|---|---|
| Dense BF16 | $4.03Hyperstack $3.99/GPU·hr · | $2.67Hyperstack $6.00/GPU·hr · | 0.66× |
| Dense FP8 | $2.02Hyperstack $3.99/GPU·hr · | $1.33Hyperstack $6.00/GPU·hr · | 0.66× |
| Dense FP4 | —no dense rate at this precision | $0.67Hyperstack $6.00/GPU·hr · | — |
Spec differences
From each chip's spec-sheet row on /hardware, with its vendor-document review date; est marks a figure that review did not confirm.
| Spec | H200H200 SXM · reviewed 2026-10-15 | B200B200 (HGX) · reviewed 2026-10-15 | B200 ÷ H200 |
|---|---|---|---|
| Dense BF16 | 989 TFLOPS | 2,250 TFLOPS | 2.28× |
| Dense FP8 | 1,979 TFLOPS | 4,500 TFLOPS | 2.27× |
| Dense FP4 | —not listed | 9,000 TFLOPS | — |
| Memory | 141 GB | 180 GB | 1.28× |
| Bandwidth | 4.8 TB/s | 7.7 TB/s | 1.6× |
| TDP | 700 W | 1,000 W | 1.43× |
Attribution for these prices and specifications: “AI Compute Tracker (aicomputetracker.com), CC BY 4.0; third-party terms apply to data from Vast.ai, Amazon Web Services, Microsoft Azure, Oracle Cloud Infrastructure, GPU cloud pricing pages, RunPod, Chip vendor documentation”.