H200 servers, bought and hosted in India.

The H200 keeps the H100’s compute and adds 76% more memory per GPU. For large-model inference that memory is often the difference between one server and two.

Illustration of a 8-gpu hgx server, 1.1 tb of gpu memory with its lid removed.
GPU memory
141 GB HBM3e per GPU
GPU power
Up to 700 W per GPU
Server power
Up to ~10.2 kW per server
Rack space · cooling
5U–8U, by chassis · Air

Shipping Lead time, October 2026: 2–4 weeks · All GPU lead times

Good fit for

Who buys H200.

GPU server hardware in detail.

The hardware

What an H200 server looks like.

8-GPU HGX server, 1.1 TB of GPU memory. 141 GB HBM3e per GPU, up to 700 W per GPU, 5u–8u, by chassis of rack space.

  • Server power: Up to ~10.2 kW per server
  • Cooling: Air

Hosting

Where an H200 server runs.

One H200 server draws up to 10.2 kW, more than most offices can supply. We place it in a rack provisioned for its full rated draw, with each power supply on a separate feed.

How managed hosting works

Yotta NM1 Navi Mumbai

  • Uptime Institute Tier IV certified
  • Dual utility feeds from separate substations
About Yotta NM1

Yotta D1 Greater Noida

  • Dual 220 kV feeds with an on-site substation
  • 48 hours of on-site backup power
About Yotta D1

H200 questions

Is H200 better than H100 for inference?

For memory-bound work, usually yes. 141 GB per GPU lets larger models and longer contexts fit on fewer GPUs. For small models that already fit on an H100, the gain is smaller.

How much power does an 8-GPU H200 server need?

NVIDIA rates the DGX H200 at up to 10.2 kW. Plan the rack and the feeds around that peak, not the average.

Can you host an H200 server I already bought?

Yes. Send the model and configuration and we’ll confirm the rack and power placement before you ship it.

Other GPU servers

Owning one

Talk to OwnGPU

Give your GPUs a managed home.

One line is enough. We’ll call or email you back.

I want to

Or email [email protected] · Privacy