Blog ▸ B200 GPU Pricing 2026: Wholesale Rates, Per-Hour Cost, and How It Compares to H200
GPU Infrastructure
B200 wholesale rates run $3.75 to $6.25 per GPU-hour. Here is the current market range, how B200 compares to H200 on cost per token, and what to expect from a quote.
B200 GPU Pricing 2026: Wholesale Rates, Per-Hour Cost, and How It Compares to H200
GPUaaS.com Team
GPU Infrastructure
July 21, 2026
The world's most wanted GPU, NVIDIA B200 bare metal DC in US West - live on packet.ai →→(Access it from Bare metal CTA on top after login)
B200 wholesale rates run $3.75 to $6.25 per GPU-hour depending on provider and contract term. The market median sits at $6.01 to $6.25/hr as of July 2026, up roughly 12% from $5.50/hr a year earlier.
That's the number most searches for "B200 price" are actually looking for. Here's what sits behind it.
Key takeaways
B200 on-demand rates range $2.69 to $18.53/hr across the market, with a median of $6.01 to $6.25/hr as of July 2026
The market median has risen roughly 12% year-over-year, from $5.50/hr in July 2025
Wholesale rates through GPUaaS.com typically land in the lower half of the market range, closer to $4.00 to $5.00/hr
192GB of HBM3e memory, roughly 35% more than H200's 141GB, reduces sharding and quantization overhead for the largest models
Neoclouds typically discount 15 to 30% for 4 to 12 week commitments
The spread across the market is wide. Lowest verified on-demand rate: $2.69/hr on spot pricing. Highest named provider: GCP at $18.53/hr. Lambda Labs currently undercuts the pack at $5.29/hr, with Northflank close behind at $5.87/hr. That's a 7x range for the same chip, depending entirely on who's quoting and what commitment structure comes attached.
The median has moved. A year ago, the market median sat around $5.50/hr. It's now $6.01 to $6.25/hr, roughly a 12% rise. That's not runaway inflation, it reflects genuine demand pressure as more workloads move onto Blackwell-class hardware while supply hasn't fully caught up.
◆ B200 VS H200 COST PER TOKEN
Where the memory premium pays off
B200 carries 192GB of HBM3e memory and 8 TB/s of bandwidth, roughly 35% more memory than H200's 141GB. That headroom matters directly for workloads running 70B+ parameter models with long context windows, where H200's memory ceiling forces sharding or aggressive quantization that B200 doesn't need.
For a workload that fits comfortably inside H200's memory already, B200's higher rate doesn't pay for itself. For a workload pushing against H200's ceiling, B200's extra memory can reduce enough overhead to come out ahead on cost per token, even at the higher hourly rate.
◆ CONTRACT LENGTH AND DISCOUNTS
Commitment moves the rate more than the provider does
Contract length moves the rate meaningfully. Neoclouds typically discount 15 to 30% for four to twelve week commitments. Hyperscalers offer one-year reserved tiers at a discount too, but the floor rate itself starts much higher on hyperscaler platforms regardless of term.
$6.01-$6.25
the median B200 on-demand rate per GPU-hour across tracked providers as of July 2026, up roughly 12% from $5.50/hr a year earlier
AIMultiple GPU Index, getdeploying.com, July 2026
GPUaaS.com sources B200 capacity from vetted providers, typically landing in the lower half of the current market range rather than the hyperscaler-adjacent ceiling. On a $6.25/hr market median, that puts a wholesale quote closer to $4.00 to $5.00 for most standard configurations.
Submit a spec, GPU count, region, contract length. Get a quote back within 24 hours.
Get a B200 wholesale quote in 24 hours.
Typically below the market median. No buyer fees. For single GPUs, packet.ai handles self-serve access with 24/7 human support.
B200 on-demand rates range from $2.69 to $18.53 per GPU-hour across the market, with a median of $6.01 to $6.25 per hour as of July 2026. Wholesale rates through vetted providers typically land in the lower half of this range, in the $4.00 to $5.00 range for most standard configurations.
Depends on the workload. If a model fits comfortably in H200's 141GB of memory, B200's higher rate doesn't pay for itself. If a workload is pushing against H200's memory ceiling and forcing sharding or aggressive quantization to fit, B200's 192GB can reduce that overhead enough to come out ahead on cost per token despite the higher hourly rate.
The market median rose roughly 12%, from $5.50/hr in July 2025 to $6.01 to $6.25/hr in July 2026. This reflects growing demand as more workloads move onto Blackwell-class hardware, while supply has not fully caught up to that demand.
Yes. Neoclouds typically discount 15 to 30% for 4 to 12 week commitments. Hyperscalers offer 1-year reserved tiers at a discount as well, though their floor rate starts considerably higher than specialized providers regardless of contract term.
Submit a workload spec, GPU count, region, and contract length, and GPUaaS returns a competitive quote from a vetted provider within 24 hours, typically landing in the lower half of the current market range.
Last reviewed: 22 July 2026. Pricing data from AIMultiple GPU Index, getdeploying.com B200 Cloud Pricing, Tech Insider Blackwell GPU Pricing Report, gpus.io NVIDIA B200 Price Comparison, and Thunder Compute NVIDIA B200 Pricing analysis. Get a B200 cluster quote on GPUaaS.com.
