About TensorDock

TensorDock aggregates GPU capacity from independent data centers and resells it through one console and one API. It sits between a marketplace like Vast.ai and a first-party cloud: the hardware belongs to vetted operators in real facilities, but you are not buying from a single vertically integrated provider.

The useful consequence is granular sizing. Rather than picking from a fixed instance menu, you choose the GPU, then dial vCPU, RAM, and storage independently. If your workload is GPU-bound and needs almost no CPU, you stop paying for CPU you will not touch.

Pricing

Hourly and considerably cheaper than the majors. RTX 4090s sit around $0.37/hr, an A6000 48GB around $0.70/hr, and H100s around $2.25/hr. There is no monthly commitment, and instances can be stopped to stop GPU billing while keeping the disk, which still costs a small amount.

Because supply comes from multiple operators, availability of any given card varies by location and by day. Check before you plan around a specific configuration.

Service Details

Instances are full virtual machines, not containers, which is a meaningful difference from RunPod. You get your own kernel, so custom drivers, kernel modules, and Docker-inside-the-VM all work normally. Ubuntu is the default; other Linux images and Windows are available.

The API covers listing available hardware, deploying, and destroying, so you can script a fleet that spins up for a batch job and disappears afterwards. That is where the hourly pricing actually pays off.

Consider this if you want marketplace prices but would rather not run production work on a stranger’s home server.

Available in: United States, Canada, Europe, Asia

Learn More

For more information about TensorDock, visit their website at https://www.tensordock.com/

Plan	Price/mo	CPU	RAM	Storage	Notes
RTX 4090	—	8 vCPU	32GB	100GB NVMe	Around $0.37/hr
RTX A6000 48GB	—	12 vCPU	64GB	200GB NVMe	Around $0.70/hr
H100 SXM5	—	24 vCPU	180GB	500GB NVMe	Around $2.25/hr