Rent NVIDIA H100 SXM in the Cloud

Three pricing tiers shape the cost of cloud NVIDIA H100 SXM:

- On-demand — kicks off at $1.57 per hour at Latitude.sh. Flexible, no commitment, best for interactive work.
- Spot / preemptible — from $1.49 per hour at Novita AI (~5% off). Subject to interruption; best for checkpointable pre-training.
- Reserved — ~up to 40% discount for multi-month commitments. Best for steady production serving.

Across the 7 providers we monitor, rates drift weekly as capacity loosens or tightens, so always cross-check live pricing before committing to a long run.

The cheapest NVIDIA H100 SXM right now is on Latitude.sh at $1.57/hr; Vultr and Vast.ai also list it — pick the region and billing that fits your workload.

The NVIDIA H100 SXM datasheet shows a Hopper-class card with 80 GB of HBM3 memory, 3,350 GB/s memory bandwidth, 990 TFLOPS of FP16 compute, 67 TFLOPS of FP32 compute, and a 700W thermal envelope. It reached market in 2023.

Those specs make it capable of handling the full modern AI stack: pre-model training moderate-sized models, fine-tuning most LLMs, running production real-time serving at useful batch sizes, and accelerating diffusion / image generation workloads.

Ready to deploy? Latitude.sh has NVIDIA H100 SXM from $1.57/hr. Vultr and Vast.ai also carry it.

Compute specs on NVIDIA H100 SXM: 990 FP16 TFLOPS, 67 FP32 TFLOPS, with 3,350 GB/s of memory bandwidth on a 80 GB Hopper package.

In practice, that's enough horsepower to handle large-batch FP16 pre-training with near-peak utilisation and to sustain production inference on large language models under typical batch-serving patterns. Performance-per-dollar is competitive at $1.57 per hour cloud pricing, especially for teams that can push batch sizes to fully saturate tensor cores.

Ready to deploy? Latitude.sh has NVIDIA H100 SXM from $1.57/hr. Vultr and Vast.ai also carry it.

Cloud providers carrying NVIDIA H100 SXM in July 2026: DigitalOcean, Latitude.sh, Massed Compute, Novita AI, RunPod, Vast.ai, and others (7 total). Latitude.sh has the most affordable on-demand rate ($1.57 per hour); Novita AI has the most affordable spot rate ($1.49 per hour).

When evaluating, go beyond the per-hour rate: check typical queue times, whether instances are bare-metal or virtualised, what interconnect is available for multi-GPU model training, and what support channels you get. These factors often matter more than a 10-15% rate difference.

Rent NVIDIA H100 SXM today from Latitude.sh, Vultr, or Vast.ai — compare live pricing and spin up in minutes.

NVIDIA H100 SXM use cases cluster around Large-scale AI training, distributed workloads, LLM pre-training. Those fit its spec envelope: enough VRAM (80 GB) for real AI work, enough compute in its Hopper tensor cores for fast training, and enough bandwidth for efficient real-time serving on transformer models.

Teams typically pick NVIDIA H100 SXM when they need more headroom than consumer-class cards but don't yet need frontier-class memory. At $1.57 per hour cloud pricing, it's a workhorse option for research labs, startups, and production real-time serving services.

Rent NVIDIA H100 SXM today from Latitude.sh, Vultr, or Vast.ai — compare live pricing and spin up in minutes.