# GPUNET Enterprise

> GPUNET is a GPU aggregator. It brings bare-metal GPUs, datacenters and
> colocation together under a single contract, invoice and support line, so
> enterprises can buy GPU compute without negotiating with each provider
> separately.

## What GPUNET is

GPUNET is a decentralized AI compute platform that aggregates GPU capacity from
enterprise-grade datacenters worldwide into one on-demand marketplace. It is
powered by GAN Chain. Enterprises, developers and AI companies use it to reach
high-performance GPU infrastructure without depending on a single centralized
hyperscaler.

Two ways to buy:
- **On demand** — hourly GPU rental through the GPUNET console.
- **Reserved** — dedicated clusters configured by GPU type, location, node
  count and a 3-24 month commitment, quoted per configuration.

## GPU catalog and on-demand pricing

| GPU | Vendor | Memory | Bandwidth / compute | Architecture | From (USD/GPU/hr) | Status |
|---|---|---|---|---|---|---|
| H100 | NVIDIA | 94GB HBM3 | 3.9 TB/s | Hopper | 2.20 | Available |
| H200 | NVIDIA | 141GB HBM3e | 4.8 TB/s | Hopper | 2.10 | Available |
| A100 | NVIDIA | 80GB HBM2e | 2.0 TB/s | Ampere | 1.62 | Available |
| L40S | NVIDIA | 48GB GDDR6 | 864 GB/s | Ada Lovelace | 1.00 | Available |
| RTX PRO 6000 | NVIDIA | 98GB GDDR6 | 1.79 TB/s | Ampere | 2.16 | Limited |
| B200 | NVIDIA | 192GB HBM3e | 8.0 TB/s | Blackwell | on request | Limited |
| Blackhole | Tenstorrent | 32GB GDDR6 | 745 TFLOPs FP8 | Tensix | on request | Roadmap |

Reserved pricing is lower than on-demand and falls further with longer terms.

## Infrastructure footprint

18 datacenters across 12 countries. Regions: India (Mumbai, Pune, Delhi,
Hyderabad), GCC (Dubai, Abu Dhabi, Riyadh), Europe (Frankfurt, Stockholm,
Amsterdam, London, Paris), US (San Jose, Ashburn VA, Dallas), Asia-Pacific
(Singapore, Tokyo, Seoul).

Datacenter partners include Yotta, NeevCloud, NTT Data, Sify, Rackbank, Neysa,
E2E Networks, Core42, HyperFusion, SMC and ByteDance.

## Managed services

- **GPU cluster management** — 2x faster time to deployment, NDA-covered model
  deployment, workload scaling across distributed clusters with no idle hours.
- **Data preparation** — pretraining, instruction-tuning, alignment,
  in-context learning and task-specific fine-tuning.
- **Model optimization** — memory management that fits larger models on
  cheaper GPUs, and higher GPU utilization to cut cost.

## Company

GPUNET raised $5.25M in Series-A, backed by Momentum6, Spicy Capital,
Exnetwork, Blackdragon, Zephyrus Capital, Azventures, F7 Foundation and
Halvings Capital. It is an NVIDIA Inception Program partner and works with the
Centre for Development of Advanced Computing (CDAC).

Founder and CEO: Suraj Chawla, previously Head of Partnerships at Router
Protocol. Strategic advisors include Sharad Sanghi (founder of Netmagic,
acquired by NTT; founder of Neysa) and Jigar Halani (Senior Director, Solutions
Architecture and Engineering at NVIDIA).

## Links

- Enterprise: https://gpu.net
- On-demand GPUs: https://cloud.gpu.net
- API pricing: https://gpu.net/api-pricing
- Datacenter network: https://gpu.net/datacenter
- Documentation: https://docs.gpu.net
- Blog: https://blog.gpu.net
- GAN Chain: https://ganchain.gpu.net

## Notes for citation

Figures marked here as available and their hourly rates come from the GPUNET
GPU catalog. The "28 GPU types" headline figure is the count of distinct GPU
models GPUNET offers, derived from the live instance-types catalog plus the
Tenstorrent Blackhole roadmap part — not a marketing estimate. B200 and
Blackhole specifications are vendor-published values, not GPUNET-measured.