NVIDIA H200 Price Comparison (Reviewed November 2026)

The NVIDIA H200 features 141GB of HBM3e memory for large-model inference and memory-heavy training. But this extra capacity carries a premium.

This guide covers the NVIDIA H200 pricing landscape, including on-demand rental rates and an approximate price to purchase. It also breaks down the full H200 spec sheet, SXM vs NVL variants, and comparisons to other relevant GPUs.

NVIDIA H200 Cloud Pricing

H200 - Price Trend

The mean on-demand H200 price across 11 providers rose from $5.95 per GPU-hour on Aug 5 to $6.06 on Nov 3, 2026 (+1.9%). Google Cloud is excluded because its H200 is reservation-only.

Recent H200 Price Changes

H100 and H200 - Cloud Rates Comparison

Across 9 providers that feature both cards, the H200 costs an average of 13% more per GPU-hour than the H100. The gap runs from 0% at Oracle to 33% at Runpod.

H200 Cost Benchmark

Note

At $1.09, an hour on Thunder Compute's A100 buys about 17 minutes of an H200 at Vast.ai's $3.85/hr marketplace median.

H200 Hardware Price: Buy vs Rent

The main difficulty when weighing whether to buy or rent an H200 is that NVIDIA doesn't publish list prices for data-center GPUs. These calculations resort to OEM reseller quotes and secondary market listings. Needless to say, sourcing this kind of hardware is not easy even for companies that have the funds.

This breakeven calculation is based on Hyperbolic's fixed rate of $3.99/hr. It includes electricity at $0.13/kWh: 10.2 kW for the 8-GPU system and 0.9 kW for a single NVL card and its host. It excludes rack space, staff and resale value.

A single H200 NVL bought from a reseller is the cheapest way to own an H200. The SXM H200 only comes in 8-GPU HGX systems, so the comparison there is a full system against renting eight GPUs.

H200 Specifications

The H200 comes has two variantes: SXM and NVL. The main difference between them is power and interconnect.

• The H200 SXM runs at up to 700 W and only comes in 8-GPU HGX systems, where NVSwitch links every GPU to every other at full NVLink speed.
• The H200 NVL is capped at 600 W, which costs it about 16% of the SXM's peak throughput, but it's a standard PCIe card that fits an ordinary air-cooled server. NVLink bridges connect two or four cards.

H200 vs H100

The H200 uses the same GH100 die as the H100. But it got a memory upgrade; the H100 SXM's 80 GB of HBM3 was replaced with 141 GB of HBM3e. This gives the H200 76% more capacity and 43% more bandwidth.

When to Choose the H200

The H200 earns its cost in three scenarios:

• A model exceeds 80GB VRAM and would otherwise require multi-GPU tensor parallelism on H100.
• When serving with long context windows (128K+ tokens) where KV cache consumes 30–50GB per request on a 70B FP8 model.
• For large-batch production inference where the 37–42% throughput gain reduces GPU count and total cost per token.

When to Choose the H100

For development, fine-tuning, and any inference workload that fits inside 80GB, the H100 is the cost-effective choice. The compute is comparable, and Thunder Compute's H100 80GB is available at $3.20/hr.

Spin up an H100 instance now.

H200 vs B200

The B200 is NVIDIA's Blackwell successor to the H200. It is a full generational rebuild: a new die, native FP4 support, and more memory than the H200.

What LLMs Fit on H200 GPUs?

With 141GB of VRAM per card, an H200 runs mid-sized models on a single GPU, while the largest models need multi-GPU nodes or multi-node clusters. The estimates below are published weight sizes only; add 10–30% for KV cache, activations, and framework buffers at a moderate context window to gauge the real deployment footprint.

Read next

Explore the full Cloud GPU market landscape in AI GPU rental market trends analysis.

Last Thoughts on NVIDIA H200 Pricing

The H200 is the right GPU when your workload genuinely needs more than 80GB of VRAM. For everything else, H100 availability may be the more practical constraint.

Looking at recent price moves, H200 rates are likely to keep rising through the rest of 2026.

FAQ