Fight for the best price on your 128+ H100 cluster

Place a bid and get your H100 cluster: it’s a special offer for those training large models. Reserve a 128+ cards cluster with NVIDIA® H100 Tensor GPUs and 3.2 Tbit/s InfiniBand for at least 2 weeks.

How it works

Follow these steps — it’s really simple. For more details, see our FAQ.

Place your bid

Place your bid on a cluster from 128 to 512 GPUs till May 31. We will contact you as soon as you place your bid to confirm it is real.

Track your position

Your bid price will be visible in the Bid Tracker below. You can track your place against other participants and update your bid if needed.

Win and take it all

It’s a special offer for a part of our capacity, that’s why the winner gets the cluster first. Other bidders will get their resources in descending order.

Host configuration

  • 8 × NVIDIA® H100

  • 8 × 20 Intel® Sapphire Rapids vCPUs

  • 8 × 160 GB RAM

  • InfiniBand 3.2 Tbit/s host

Minimum GPU capacity

The minimum GPU capacity requested is 16 hosts or 128 GPUs.

Architect support

Dedicated architect support is provided free of charge.

Bid starting price

The starting price for one GPU per hour is 2,5 USD.

Make your bid

Bid Tracker

Buyer
USD/GPU/h
#GPU
Days
i***o@k***t.co.il
2.52
240
30
i***n@m***v.io
2.50
128
15

The bidders list is updated every day at 12 a.m. CET. Bidding close date is May 31st, results are published on June 3rd.

What makes Nebius AI great for multi-host training

Start fast

Set up your training infrastructure with ready-to-deploy Terraform and Helm solutions for training, inference and storage based on Kubernetes, Slurm, Kubeflow and other popular tools.

Restart in minutes

All layers of our full-stack cloud are built with resilience in mind — infrastructure, IaaS and managed services. You only need minutes to recover VMs or GPU hosts.

Get real support

We do listen to the feedback from our customers — have a look at the Recraft experience and how direct communication with our dedicated solution architects helped them to train a 20B foundational model.

Terms and conditions

Your bid is not legally binding for either side. It merely serves as an expression of interest and does not create any contractual obligations.

The bid cannot be less than 2.5 USD per GPU/hour. Winning bid allows you to get access to high-volume GPU resources without any wait lists.

The minimum period of consumption in the bid is 14 days. There is no limits for the maximum period.

The winner should make a 100% prepayment for the period they have indicated in the bid.

The minimum amount of GPUs is 128, the maximum amount is 512 GPUs. Your number should be a multiple of eight, as there are 8 GPUs in the host.

The bid with the biggest price wins, and we guarantee that we provide GPU cluster to the winner as soon as the payment is made.

FAQ

Can I get a trial to test the cluster?

We can make a live demo of the platform and grant access to the demo environment before you sign the contract.