Thunder Compute
Verified Oct 2, 2026 ActiveOn-demand cloud GPUs for AI and ML work, rented by the minute and managed from a CLI, VS Code or an API.
Listed on EffectHub ·
Pricing
- Pricing model
- Paid
Pay as you go. The pricing page lists instances from $0.35 per hour for an RTX A6000 up to $3.20 per hour for an H100, billed by the minute, plus $0.219 per GB per month for disk beyond the first 100 GB and $0.05 per GB per month for snapshots. Sandboxes start at $1.95 per GPU-hour billed per second. These are hourly rates, so no monthly figure is recorded.
Use tool ↗About Thunder Compute
Thunder Compute rents NVIDIA GPUs, from RTX A6000 to H100, as persistent instances you reach over SSH, a VS Code extension, a CLI or the browser, billed by the minute. A second product, Sandboxes, gives short-lived CPU or GPU environments billed per second for evaluations, reinforcement-learning rollouts and agent runs. Instances support snapshots and templates, a REST API and an MCP server for automation, and the vendor says instances are never preempted. It is run by a San Francisco startup that went through Y Combinator in 2024.
- Platforms
- Web · API · VS Code extension
- API
- Yes
- Last verified
- Oct 2, 2026
Verified sources
- Pricing page: instance GPU rates per hour billed by the minute, storage and snapshot rates, sandbox rates per GPU-hourthundercompute.com ↗
- Home page: persistent instances via SSH, VS Code, CLI and browser; sandboxes; REST API and MCP serverthundercompute.com ↗
- Y Combinator profile: San Francisco, Summer 2024 batch, founded 2022ycombinator.com ↗
Is Thunder Compute legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Last checked
- Oct 2, 2026The site answered
- On the market since
- 2022The vendor's own material, read by us
- Operated by
- Thunder Compute, United StatesOur research, card verified Oct 2, 2026
Pros & cons of Thunder Compute
Thunder Compute is a straightforward pick for researchers and small teams who want an A100 or H100 for a few hours without a contract. Per-minute billing and a VS Code workflow keep it simple. Large training clusters and regulated workloads call for a bigger provider.
Pros
- Per-minute instance billing and per-second sandbox billing
- Published hourly prices for A6000, L40, A100 and H100
- No data egress charge listed
- VS Code extension, CLI, REST API and MCP server for automation
Cons
- Disk beyond the included 100 GB and snapshots are billed separately
- Smaller GPU fleet than hyperscalers, so capacity for big clusters needs reserved pricing
- No managed inference or training platform on top of the machines
Best for
- Researchers and students who need a GPU for experiments by the hour
- Developers running parallel evaluation or RL rollouts in short-lived sandboxes
Not for
- Teams that need multi-region, compliance-certified infrastructure at large scale
- Anyone wanting managed model serving rather than raw GPU machines
Watch out: Disk is billed hourly while an instance runs and snapshots are charged per GB per month, so forgotten snapshots and large disks add cost even when you use little GPU time.