H100 GPU Cloud Availability and Pricing

Availability #

• Lambda Labs - At least 1x (actual max unclear) H100 GPU instant access
  • Max H100s avail: 60,000 with 3 year contract (min 1 GPU)
  • Pre-approval requirements: Unknown, didn’t do the pre-approval.
  • Pricing: $2.40 $1.99 per/GPU per/Hour on demand ($1.89/hr with largest reservation)
    • Update: The Lambda CEO just commented that they’ve dropped their prices to $1.99/hr.
• FluidStack - 1x (and seemingly up to 10x) H100s available with instant access
  • Max H100s avail: 1,024 GPUs with approval, 10 without talking to sales (min 1 GPUs)
  • Pre-approval requirements: Unknown, didn’t do the pre-approval
  • Pricing: price lowered to $1.99/hour since publishing ($2.40 per/GPU per/Hour with reservation)
• Runpod - limited access after pre-approval.
  • Max H100s avail: 64 GPUs (listed as 8x, so likely min 8 GPUs, but the reservation form offers 1x so the min may be smaller, and the max may be higher)
  • Pre-approval requirements: Fill out reservation form on website.
    • Information required:
      • What is your use case?
      • What specs do you require (PCIE/SXM/NVL, network speed, infiniband or not, storage volume, CPU, RAM)?
      • How many H100 do you require? (1-50+)
      • How long do you plan on using H100s?
  • Pricing: no instant pricing ($2.30 per/GPU per/Hour listed with reservation)
• CoreWeave - limited access after pre-approval.
  • Max H100s avail: Unknown, didn’t do the pre-approval. (min 1 GPU)
  • Pre-approval requirements: Unknown
  • Pricing: $4.25/hour ($2.23 per/GPU per/Hour with reservation)
• Azure - limited access after pre-approval.
  • Max H100s avail: Unknown, didn’t do the pre-approval.
  • Pre-approval requirements: Unknown.
  • Pricing: Unknown.
• Oracle Cloud - limited access after pre-approval.
  • Max H100s avail: Unknown, didn’t do the pre-approval.
  • Pre-approval requirements: Unknown.
  • Pricing: Unknown.
• AWS - none.
• Crusoe Cloud - none.
• Google Cloud - none.
• Vast.ai - none.

Pricing #

1. Lambda Labs - $2.40 per H100 per hour ($1.89/hour with largest reservation)
   • Max H100s available with Lambda Labs = 60,000 GPUs
   • Hoops needed to get access to H100s on Lambda Labs:
     • Any H100s = no hoops
     • Max H100s - contact sales department
2. CoreWeave - $4.25 per H100 per hour ($2.23/hour with largest reservation)
   • Max H100s available with CoreWeave = unknown
   • Hoops needed to get access to H100s on CoreWeave:
     • Any H100s = contact sales department
3. Runpod - no instant pricing available, requires reservations (starting at $2.30/hour with largest reservation)
   • Max H100s available with Runpod = 64 GPUs
   • Hoops needed to get access to H100s on Runpod:
     • Any H100s = fill out reservation form on website
4. FluidStack - no instant pricing available (starting at $2.40/hour with reservations)
   • Max H100s available with FluidStack = 1,024 GPUs
   • Hoops needed to get access to H100s on FluidStack:
     • Any H100s = contact sales department

The best provider if you need 100+ H100s and want minimal costs is likely: Lambda Labs or FluidStack.