Prices updated 2026-06-23 — sourced from live provider data

GCP H100 Pricing Options

All available H100 instance types on GCP, sorted by lowest hourly rate.

Region Instance Type Price/hr GPUs vCPUs RAM
us-central1 a3-highgpu-8g On-Demand $32.7730/hr Cheapest 8× GPU 208 vCPU 1872 GB
us-central1 a3-megagpu-8g On-Demand $98.3200/hr 8× GPU 208 vCPU 1872 GB

About GCP H100 Cloud Pricing

Google Cloud Platform offers GPU compute through its A2, A3, and G2 machine series, with deep integration into Vertex AI, TPU pods, and Google's proprietary ML infrastructure.

Why Choose GCP for H100

  • TPU v5p available alongside GPU options for specific workloads
  • Vertex AI for managed ML pipelines
  • Sustained use discounts (automatic, no commitment needed)
  • Preemptible VMs for up to 60–91% savings on batch jobs

Best suited for: Teams using TensorFlow/JAX who want access to both GPU and TPU hardware, plus managed ML services via Vertex AI.

H100 Technical Overview

ArchitectureHopper
Memory80 GB HBM3
Best ForLarge language model training, foundation model pre-training, and high-throughput inference

The H100 SXM5 variant with NVLink delivers up to 3.35 TB/s aggregate GPU-to-GPU bandwidth, making it the standard for large-scale distributed training. The PCIe variant trades interconnect bandwidth for broader server compatibility.

Pricing Analysis

GCP's cheapest H100 option starts at $32.7730/hr, which is 798% above the market average of $3.6501/hr. GCP currently offers on-demand pricing only for H100. This means no commitment required, but you won't access the deeper discounts available from reserved or spot pricing.

Note: GCP's sustained use discounts provide automatic savings without commitments, but their GPU availability can be more constrained than AWS in certain regions.

Market context: H100 is the most in-demand GPU in the cloud market as of 2026. Prices have stabilized after the 2024 shortage, but availability still varies significantly by provider and region.

Top Alternatives to GCP for H100

Side-by-side cost comparison vs the cheapest GCP H100 rate ($32.7730/hr).

Provider Price/hr Type vs GCP
Vast.ai $1.4900/hr Spot -95% cheaper vs GCP
Corvex $1.5900/hr Reserved -95% cheaper vs GCP
DataCrunch $1.6500/hr Spot -95% cheaper vs GCP
CoreWeave $1.7900/hr Reserved -95% cheaper vs GCP
Shadeform $1.7900/hr On-Demand -95% cheaper vs GCP

View all H100 providers →

Calculate Your Exact H100 Cost

Enter your hours, job type, and GPU count. Our calculator compares GCP against all 32 providers instantly.

● $19 — no subscription
Track GCP H100 prices — daily alerts
Get emailed when GCP drops below your budget. $19 for 30 days. No account needed.
Set Price Alert → Calculator →

Frequently Asked Questions

How much does GCP H100 cost per hour in 2026? +
GCP H100 starts at $32.7730/hr (on-demand). The market average for H100 across all providers is $3.6501/hr. GCP offers 2 H100 configurations ranging from $32.7730 to $98.3200/hr. Use the GPU cost calculator to estimate total cost based on your hours and workload type.
What are the cheapest alternatives to GCP for H100? +
There are 27 providers offering H100 cheaper than GCP: Vast.ai at $1.4900/hr, Corvex at $1.5900/hr, DataCrunch at $1.6500/hr, and 24 more. 2 other providers offer H100 at higher rates. See the full comparison table for all H100 providers.
Does GCP offer reserved or spot H100 pricing? +
GCP offers 1 pricing type for H100: on-demand. With on-demand only, you pay per hour with no commitment. While you won't get reserved or spot discounts, this provides maximum flexibility for unpredictable workloads.
What workloads is GCP H100 best suited for? +
The H100 (Hopper architecture, 80 GB HBM3) is best suited for Large language model training, foundation model pre-training, and high-throughput inference. GCP is a hyperscaler provider — Teams using TensorFlow/JAX who want access to both GPU and TPU hardware, plus managed ML services via Vertex AI.
How much does it cost to run H100 on GCP for a month? +
At GCP's cheapest H100 rate of $32.7730/hr: running 24/7 costs ~$786.55/day or ~$23924.29/month per GPU. For training jobs, costs scale linearly — a 100-hour training run would cost ~$3277.30. Multi-GPU configurations multiply accordingly. Use the cost calculator for precise estimates with your specific parameters.
Compare across all providers

Not sure which GPU configuration fits your budget?

Get a personalized recommendation based on your monthly budget, workload, and real pricing data from all 58 providers.

Budget Optimizer — $19 →
$19 One-Time

Is a reserved contract worth it here?

Get the break-even math for GCP H100 — calculated from real 90-day pricing data, not an LLM.

Get Analysis — $19