Snapshot
Continue to About ↓- What it does
- Thunder Compute rents on-demand NVIDIA GPUs as persistent Ubuntu instances billed per minute, and as short-lived isolated GPU sandboxes billed per second for agent workloads.
More detail
- Instances are SSH-able machines for training, notebooks and remote editor development
- sandboxes are aimed at RL rollouts, model evals and GPU CI/CD that scale to zero when idle.
- Best for
- ML engineers, RL researchers and small AI teams who want an SSH-able GPU server or many isolated, parallel GPU environments billed by the minute or second without hyperscaler overhead.
More detail
- Platform teams that would rather spin up GPU CI jobs on demand than maintain GPU runners.
- Pricing
- Pay-as-you-go, no monthly subscription.
More detail
- Instances are billed per minute only while running, with separate per-hour rates: RTX A6000 $0.35, L40 $0.79, A100 80GB $1.09, H100 SXM $3.20
- extra vCPU $0.04/vCPU-hr
- first 100 GB disk per GPU included, further disk $0.219/GB/mo
- snapshots $0.05/GB/mo.
- Sandb
- Main limitation
- Teams that need multi-GPU sandboxes (one GPU per sandbox today), guaranteed capacity at a moment's notice (availability is live and can sell out), or sandbox access without a gated request to sales.
More detail
- Also a poor fit for buyers who need published region, residency or compliance details or independent third-party reviews before committing.
- Record
- Living and monitored
- Reviewed
- Evidence
- Public web research
- Machine access
- Structured profile ↗
About
- Thunder Compute rents on-demand NVIDIA GPUs as persistent Ubuntu instances billed per minute, and as short-lived isolated GPU sandboxes billed per second for agent workloads.
- Instances are SSH-able machines for training, notebooks and remote editor development
- sandboxes are aimed at RL rollouts, model evals and GPU CI/CD that scale to zero when idle.
What it helps with
- GPU cost and idle time: teams overpay for GPU machines that sit unused between experiments, and hyperscaler list prices
- Agentic workloads need many short, isolated GPU environments rather than one long-lived server; managing that fleet by h
- GPU CI pipelines often require dedicated runners that cost money between builds.
- Agents can write hostile code; workloads need isolation that containers may not give, and evals sometimes need networkin
- Experiments get interrupted or expensive to pause without a way to preserve a disk between runs.
- Cross-cloud data movement fees make it costly to pull model data and artifacts back out.
Capabilities
A practical view of what is available and where limitations apply.
| Offering | Available | Detail |
|---|---|---|
| GPU types offered | Yes | RTX A6000 48 GB, L40 48 GB, A100 80GB and H100 SXM 80 GB on instances. |
| Instance scaling | Limited | 1x/2x/4x/8x GPU instance configurations; hot-swappable specs on running instances. |
| Multi-GPU sandboxes | No | Sandboxes currently support one GPU each; larger configurations require contacting Thunder Compute. |
| Sandbox isolation | Yes | Ubuntu microVMs, one kernel per sandbox, so agent code cannot reach other workloads; systemd and Docker Compose work. |
| Network control for sandboxes | Yes | Outbound internet can be blocked at create time for closed-network evals; private and cloud-metadata networks stay blocked. |
| Per-second sandbox billing with scale to zero | Yes | Resources metered separately and billed per second until the sandbox stops or its TTL expires. |
| Preemption | No | Instances and sandboxes are stated as never preempted; they run until stopped or TTL expiry, so cost accrues if agents do not terminate them. |
| Persistent storage | Limited | Instances support snapshots billed at $0.05/GB/mo; terminating a sandbox destroys its filesystem, so data must be downloaded first. |
| Data egress fees | No | Listed at $0.00/GB. |
| Agent and editor tooling | Yes | CLI, VS Code/Cursor/Devin Desktop extensions, Python SDK, REST API and MCP server for Claude Code, Cursor and Codex; OpenClaw plugin is in public beta. |
Bot trust
How confidently AI systems can verify this profile from public evidence.
—of 100
ChatGPT—/100
Claude—/100
This profile has not completed a Bot Score assessment yet.
Evidence
Where this record came from, when it was reviewed, and how confidently it can be used.
- Verification
- Scribble researched these facts from the linked public sources.
- The owner has not manually confirmed every field.
- Last reviewed
- Primary source
- thundercompute.com
Technical profile details
Record ID
4f9c0b84-11c4-4733-8711-a00d9a8d1efbCanonical URLhttps://profiles.scribble.network/listing/thunder-compute-4f9c0b84