Dedicated GPU nodes

Bare-metal GPU rentals for AI teams.

Tell us what you need and when. Tera gives you a straight answer on configuration, start date, and economics.

B300  ·  B200  ·  H200  ·  A100

Node configurations

B300, B200, H200, and A100.

Choose a GPU to start. Availability, timing, and terms are confirmed against your request.

Blackwell Ultra

B300

8× NVIDIA B300 SXM

288 GB / GPU
2.30 TB HBM3E / node

For large-model training, inference, and memory-intensive workloads.

Blackwell

B200

8× NVIDIA B200 SXM

180 GB / GPU
1.44 TB HBM3E / node

For large training runs and high-throughput production inference.

Hopper

H200

8× NVIDIA H200 SXM

141 GB / GPU
1.1 TB HBM3E / node

For memory-heavy inference, fine-tuning, and training workloads.

Ampere

A100

8× NVIDIA A100 80 GB SXM

80 GB / GPU
640 GB HBM2E / node

For established training, fine-tuning, inference, and CUDA stacks.

Configurations shown are not live inventory. Final hardware, region, price, and start date are confirmed per offer.

Capacity brief

Tell us what you need.

Share the workload, GPU, node count, and needed-by date. We will reply with what looks viable and what still needs confirmation.

Configuration. Start date. Economics. In writing.

Your requirements

Capacity brief

Start here
Technical details (optional)

We use your details only to evaluate and respond to this request.

Tera Compute is a product and brand of Hashbranch. Hashbranch is the contracting entity and seller. Submitting requirements does not reserve capacity. Availability, configuration, price, support, and start dates are confirmed only in a written offer.