# Thunder Compute

Official pricing: https://www.thundercompute.com/pricing  
Category: gpu-cloud · Isolation: container

## Pricing regimes (raw)

- **GPU instances (on-demand, per GPU incl. base vCPU/RAM)** (resource): $0/vCPU-h, $0/GiB-h
- **GPU sandbox (Firecracker, per second, TTL)** (resource): $0.051/vCPU-h, $0.016/GiB-h
- **Reserved clusters (8-512+ GPUs, fixed term)** (resource) [sales]

## Features

Yes: snapshots, fork/clone, persistent disk, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, Docker inside, SSH, public IPv4, HTTPS preview URLs, egress allowlist, open internet, self-hosting, BYOC, SOC 2, GPU, Python, live resize, live fork (no pause), MCP server, snapshots on demand, inbound access rules, firewall inside (nftables)

No: memory snapshots, pause/resume, volumes, idle auto-stop, full VM (own kernel), nested virtualization, browser, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, static egress IP, open source, HIPAA, SSO, arm64, Windows, macOS, Node.js, HTTP method/path egress rules, wake on request, memory fork, EU data residency, automatic snapshots, agent harness API, hosted agent API (their own agent), gVisor or VM (no shared kernel), secret proxy, secret proxy for any API, extra volumes, shared volumes

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/thunder-compute.json → feature_evidence

## Caveats

- All prices are complete base GPU-instance bundles at the minimum CPU count, not stand-alone GPU surcharges. Minute billing; no egress charge.
- Additional CPU $0.04/core-h, usually 8 GB RAM included per core. Running disk first 100 GB included then $0.03/100GB-h; stopped storage uses snapshots $0.05/GB-month.
- Multi-GPU non-linear: 4×/8× A100 and L40 cost more per GPU than 1×/2×. Do not multiply single-GPU price blindly.
- GPU mode priced per GPU from the 1x bundles (vcpu_h/ram_gib_h 0 because base CPU/RAM is bundled). Extra vCPU ($0.04/vCPU-h) beyond the bundle minimum is not modelled; A100/L40 4x and 8x bundles cost more per GPU than 1x (understated). Running disk beyond 100 GB: $0.03 per 100 GB-hour = $0.219/GB-month at 730 h.
- Null values mean unknown, not free/unlimited. No model-token cost, tax or unverified ancillary charge included.
- Native browser/device/task/credit prices with unknown CPU/RAM are deliberately non-priceable. Check regime before interpreting any partial resource estimate.
- Instances bill 'for every minute they exist, from creation until you delete them'. There is no stop state, so an idle instance pays the full GPU bundle rate. The only cheap parked state is snapshot + delete ($0.05/GB-month; API snapshot_gb $0.00006849/GB-h).
- Included disk is 100 GB PER GPU (8x instance includes 800 GB). Instance disk ranges 100-500 GB (1x) up to 100-4000 GB (8x). Disk can be expanded, not shrunk. The card's included_disk_gib 100 is right for 1 GPU only.
- Multi-GPU bundles include more vCPU/RAM per GPU: A100 4x 36-60 vCPU/228-420 GB ($5.96), 8x 96-120 vCPU/648-840 GB ($11.92); H100 4x 60 vCPU/480 GB ($12.80), 8x 120 vCPU/960 GB ($25.60); L40 4x 24-40 vCPU ($3.96), 8x 48-80 ($7.92). A6000 is 1x only. Public API (GET /v2/pricing, no auth) confirms every rate, plus additional_vcpus $0.04/h and disk_gb $0.0003/GB-h.
- Account restrictions: service unavailable in 18 countries (incl. Brazil, Mexico, China, Indonesia, Malaysia, Vietnam, Saudi Arabia) and B2B-only (VAT ID required) in 27 more (incl. India, South Korea, Switzerland, Turkey, UAE). One account per user. Preloaded credit never expires and is non-refundable. Promo credit is revocable. Cards are charged at rising 'threshold' checkpoints; there is no hard spend cap.

## How this provider charges


YC: Summer 2024; directory status **Active**. Research scope: public-product-researched. Source: https://www.ycombinator.com/companies/thunder-compute

GPUs for Agents

## Regime table

| Regime | When it applies | How billed | Numbers | Source |
|---|---|---|---|---|
| GPU instances | Public listed mode | resource; CPU alloc, memory alloc | Unknown or native-unit billing; see facts below | https://www.thundercompute.com/pricing |
| Usage | Plan limits apply | Monthly fee; credits only as stated | fee=0; included usage=$0; concurrent=None; max session h=None | https://www.thundercompute.com/pricing |
| Retained disk / GiB-month | Lifecycle/scope must be checked | Native add-on | unverified, not $0 | https://www.thundercompute.com/pricing |
| Snapshots / GiB-month | Lifecycle/scope must be checked | Native add-on | $0.05 | https://www.thundercompute.com/pricing |
| Egress / GiB | Lifecycle/scope must be checked | Native add-on | $0 | https://www.thundercompute.com/pricing |
| IPv4 / month | Lifecycle/scope must be checked | Native add-on | unverified, not $0 | https://www.thundercompute.com/pricing |

## Verified billing facts and gotchas

1. All prices are complete base GPU-instance bundles at the minimum CPU count, not stand-alone GPU surcharges. Minute billing; no egress charge.
2. Additional CPU $0.04/core-h, usually 8 GB RAM included per core. Running disk first 100 GB included then $0.03/100GB-h; stopped storage uses snapshots $0.05/GB-month.
3. Multi-GPU non-linear: 4×/8× A100 and L40 cost more per GPU than 1×/2×. Do not multiply single-GPU price blindly.
4. Card has gpu_only mode with null rates because host inclusion and count-specific bundles cannot safely be represented by one GPU additive table.

Native pricing (not coerced into incompatible engine units):
```json
{
  "extra_vcpu_hour_usd": 0.04,
  "extra_disk_gb_hour_usd": 0.0003,
  "included_disk_gb": 100
}
```

## Worked example

Requested: 4 vCPU/8 GiB, 50 simultaneous × 8 h/day × 22 days = **8,800 instance-hours**, 30% CPU; 50 GiB snapshots and 100 GiB egress. A month is 730 hours only when explicitly used in rate conversion.

The specified workload has no GPU, so there is no directly comparable CPU-only offer. If forced onto cheapest 1× RTX A6000 (6 vCPU/48 GB minimum), 8,800×$0.35=$3,080 plus $2.50 for 50 GB snapshot-month, egress $0. This overprovisions GPU/CPU/RAM and is not a CPU-only quote.

## Sources and coverage

- https://www.ycombinator.com/companies/thunder-compute
- https://www.thundercompute.com/pricing
- https://www.thundercompute.com/docs/technical-specs
- https://www.thundercompute.com/docs/llms.txt
- https://www.thundercompute.com/
- https://www.thundercompute.com/docs/vscode/quickstart
- https://www.thundercompute.com/llms.txt

Page captures and documentation indexes/bundles are in `raw/`; provider JSON lists the evidence paths. HN query results were captured separately as `<id>--hn.txt`; they are discovery leads, not current tariff authority. Downloaded docs do not imply every feature was verified; unsupported assertions remain null.