H200 Cloud GPU Pricing

About the H200

The NVIDIA H200 is a memory-upgraded variant of the H100, sharing the same Hopper compute architecture but with 141 GB of HBM3e at 4.8 TB/s. This 76% increase in memory capacity over the H100 directly translates to higher throughput on memory-bound inference. For teams already on Hopper but bottlenecked by 80 GB, the H200 is a drop-in upgrade with no software changes.

Key Specifications

Cloud Pricing Context

H200 on-demand pricing starts around $2.14/hr and runs up to $3.59/hr. Pricing is settling between H100 and B200 rates. The H200 offers strong value for inference where extra memory eliminates tensor parallelism across multiple H100s, effectively halving infrastructure costs.

Best For

• Inference serving of 70B+ parameter models on a single GPU
• Memory-bound workloads where HBM3e bandwidth is the limiting factor
• Teams upgrading from H100 who need more VRAM without requalifying software
• KV-cache-heavy serving (long-context LLMs) where 141 GB keeps throughput high

Available From

Related GPUs

Compare Providers