datacenter gpu

NVIDIA H100 SXM

The GPU that started the modern AI datacenter buildout

$28,000

NVIDIA does not publish a list price; this reflects typical broker/OEM pricing in 2026.

Specs

  • GPUs in this unit1
  • Total GPU memory80 GB
  • Memory bandwidth (per GPU)3,350 GB/s
  • Total power draw700 W
  • Compute (dense BF16/FP16)989 TFLOPS

HBM3 memory, SXM5 socket, sold only as part of an integrated server (see DGX systems below).

What this actually means for you

This is a purpose-built datacenter chip — it has no display output and needs a specialized SXM server chassis with liquid or high-static-pressure cooling. You don't buy one to put in a desktop; you buy it as part of an 8-GPU server, like the ones below.

Solo capacity: at the standard FP16/BF16 rule of thumb (2 GB of GPU memory per billion parameters, plus 20% overhead for context), this unit's 80 GB can hold an open-source model up to roughly 33 billion parameters before you'd need more hardware.

Power & running cost

Continuous draw
0.70 kW
Monthly electricity
$95
Everyday comparison
~58% of one home

Run 24/7, this unit uses 16.8 kWh/day — about the same continuous power draw as 0.6x an average U.S. home (~29 kWh/day, EIA). Monthly cost assumes the 2026 U.S. residential average of 18.8¢/kWh.

Request this build

No payment, no account — just tell us your name and we'll follow up with a real quote.