Coming soon · Private beta

The world's compute,
at your fingertips.

Grid is a crowd-sourced GPU cloud for distributed ML training. Train at hyperscaler speed for a fraction of the cost — or share your idle hardware and earn.

We're launching soon — join the list for early access.

10,000+
GPUs in the grid
~70%
Lower cost vs. hyperscalers
190+
Regions worldwide

Everything you need to train AI models at scale

From a single GPU to a multi-node run, Grid gives you production-grade infrastructure without the production-grade complexity.

Distributed by default

Launch multi-node training with a world size you set. Grid handles orchestration, sharding, and fault recovery across the grid.

Bring your own code

A drop-in Python SDK and notebook workflow. Point Grid at your training script — no infra rewrites, no Kubernetes to babysit.

Reproducible runs

Pinned runtime images, captured seeds, and full run lineage. Re-run any experiment and get the same result, every time.

Pay for what you use

Per-second metered compute at a fraction of hyperscaler pricing. No reserved capacity, no idle bills, no lock-in.

Live observability

Stream metrics, logs, and hardware telemetry from every node in real time. Compare runs side by side as they train.

Secure & isolated

Workloads run sandboxed on provider hardware with end-to-end encryption and signed agent attestation.

From script to scale in three steps

01

Connect your code

Install the Grid SDK, point it at your training entrypoint, and pick a runtime image.

02

Pick your scale

Set a world size. Grid finds matching GPUs across the grid and provisions them in seconds.

03

Train & track

Watch live metrics stream in. Checkpoints and lineage are captured automatically.

Got idle GPUs?
Put them to work.

Your idle GPUs are leaving money on the table. Plug them into the grid and turn downtime into income — fully automated, fully under your control, with payouts for every second your hardware trains.

Join the waitlist
  • Turn idle hardware into passive income
  • You decide when your GPUs join the grid
  • Sandboxed, attested, secure workloads
  • Automatic payouts for every second of compute

Be first on the grid

We're opening access soon. Join the waitlist and we'll let you know the moment it's ready.

Join the waitlist