HARDWARE TRACKER

NVIDIA H200 Availability & Lead Times (2026)

Summary • 12 Data Sources

When will NVIDIA H200 be available?

The NVIDIA H200 is currently shipping to hyperscalers and elite CSPs. For general enterprise orders via OEMs (Dell, HPE, Supermicro), lead times remain at 24-32 weeks. The H200's primary bottleneck is HBM3e supply, which limits the total available capacity in the spot market.

Key Data Points

Memory: 141GB HBM3e at 4.8 TB/s
OEM Lead Time: 24-32 weeks
Cloud Availability: Available Now (Select Regions)
Inference Speed: 1.9x faster than H100 (Llama 2 70B)
Estimated Price: $35,000 - $45,000 per GPU

Check H200 Lease Rates →

Supply Chain Status

⚠

OEM Servers

24-32 Week Lead Time
(Dell, HPE, Supermicro)

✓

Cloud Instances

Available Now (Select Regions)
(AWS p5e, Azure ND MI300X)

Spot Market

Extremely Scarce
Markup: 40-50% over MSRP

Why the H200 Shortage?

It's All About HBM3e

The H200's defining feature is its memory: 141GB of HBM3e at 4.8 TB/s. This is 1.8x the memory capacity and 1.4x the bandwidth of the H100.

HBM3e production yield from SK Hynix and Micron is the primary bottleneck. Every major chip (H200, B200, MI300X) is fighting for the same limited supply of HBM stacks.

Performance Uplift (vs H100)

Llama 2 70B Inference1.9x Faster
GPT-3 175B Inference1.6x Faster
HPC Workloads1.1x Faster

Frequently Asked Questions

Can I upgrade H100 baseboards to H200?

Yes! The H200 uses the same SXM5 form factor as the H100. It is a drop-in compatible upgrade for H100 HGX systems, requiring no changes to cooling or power infrastructure.

Is H200 worth waiting for vs H100?

If you are running memory-bound inference workloads (LLMs), absolutely. The extra memory allows running larger models on fewer GPUs. For compute-bound training, H100 is still very capable and more readily available.