Voltage Park
US neocloud selling on-demand and reserved bare-metal NVIDIA HGX H100 clusters from $1.99/hr; acquired TensorDock (2025) and merged with Lightning AI (Jan 2026).
A cheap, simple A100/H100 cloud with real funding and a clear virtualisation story, but it is young, publishes no legal entity or security certifications, and all reliability figures are self-reported.
site reachable · checked 11 h ago · daily check
by the bestcloudmining editorial desk · curated review · prose updated 2026-10-02 · facts as of 2026-10-02 · how we verify
Thunder Compute is a budget GPU cloud backed by Y Combinator, operated by Thunder GPU, Inc. (Delaware) [2]. Its pricing is published and competitive (A100 at $1.09, H100 SXM at $3.20 per GPU-hour) [1] with per-minute billing and no egress fees [1]. The ToS (updated September 2026) names the entity and governing law - a transparency upgrade from our earlier data [4].
On-demand GPU instances (A100 80GB, H100 SXM) with per-minute billing and no egress charges [1]. Sandboxes for per-second billing (H100 at $2.95, A100 at $1.95) [1]. Reserved clusters of 8 to 512-plus GPUs at a discount [1]. No minimum purchase, no crypto payment - card only via Stripe [1]. There is no hosting or DePIN element - this is a traditional cloud with published per-GPU pricing [1]. The product catalog is small (A100, H100 SXM, L40S and a few others), which means less choice than larger clouds but also less complexity in comparing [1]. The Sandboxes product is positioned for AI agents - hence the "GPUs for Agents" tagline on its Y Combinator page [7].
The terms (updated September 2026) name Thunder GPU, Inc. (Delaware, 251 Little Falls Drive, Wilmington) with Delaware governing law and AAA arbitration [4]. All purchases are non-refundable [4]. KYC: no identity-verification policy is published [4]. Payments are card-only via Stripe [4]. The "no-egress-fees" claim is specific to the pricing model - network transfer is included [1].
The status page shows 100 percent uptime from July through October 2026 [4], and the pricing page is server-readable with all rates [1]. The entity is named and registered in Delaware [4]. The 13-million-dollar Series A (Matrix-led, August 2026) is corroborated by Yahoo Finance and SiliconANGLE [6]. GetLatka independently estimates under 1 million dollars in annual revenue [5]. Trustpilot is blocked from datacenter IPs during checks [7]. Trustpilot is blocked from datacenter IPs during checks [7]. What stays unverifiable: the partitioning technology's real-world performance claims, and there is no independent audit [1].
Founded as a Y Combinator company [7], backed by a 13-million-dollar Series A [5]. Reddit presence is entirely self-promotional [6]; no independent operational threads surfaced [6]. Trustpilot is blocked from datacenter IPs during checks [7]. No fraud, loss or insolvency events surfaced [6].
Published and live: A100 80GB at $1.09 per GPU-hour, H100 SXM at $3.20 [1]. Sandboxes (per-second): H100 at $2.95, A100 at $1.95 [1]. Reserved: A100 at $1.49 per GPU-hour for 8-512-plus GPU clusters [1]. Payment: card only via Stripe [4]. No KYC policy published [4]. Non-refundable per the terms [3]. No minimum purchase published [1].
Thunder Compute is the simplest budget GPU cloud on our list - published pricing, named Delaware entity, a live status page, and no DePIN complexity [1][4]. The "no-egress-fees" and "per-minute" billing are honest conveniences [1]. Its constraints are the same as any small cloud: no SLA remedy [4], non-refundable purchases [3], and a Y-Combinator-scale company with a 13-million-dollar round [5]. The partitioning technology that underpins the pricing is a genuine engineering trade-off - you are splitting one physical GPU across two tenants, so peak performance will be lower than a dedicated instance [1]. For cost-sensitive workloads that tolerate this, the pricing advantage is real [1]. For budget AI experiments and short bursts, it is a clean, verifiable option [1].
| Documented price | $0.350 per GPU/hr |
|---|---|
| Notes | Observed 2026-09-24 at https://www.thundercompute.com/pricing per GPU/hr (1x): RTX A6000 $0.35, L40 $0.79, A100 80GB $1.09, H100 PCIe $3.20. Multi-GPU tiers: 2x H100 $6.40, 4x $12.80, 8x $25.60. Add-ons: additional storage $0.03/100 GB/hr (first 100 GB included), snapshots $0.05/GB/mo, extra vCPU $0.04/vCPU/hr, no data-egress fees. Billed per minute. |
Run the long math before paying: cloud mining calculator →
Built only from what is documented on this page as of 2026-10-02 - anything not documented stays a gap on purpose.
Everything happens on thundercompute.com - we count the click, nothing else changes for you.
Price from $0.350 per GPU/hr - start with the minimum, not with the plan the checkout page pushes.
maintenance . A term you did not read is money you cannot get back.
No independent proof is documented. Withdraw the first payout early and compare it with the pool worker or your own wallet.
Then run the math: mining calculator · contract vs. HODL · all GPU compute providers.
Short answer: Based on our checks as of 2026-10-02: Thunder Compute scores 6.7/10 in our weighted review. We logged no red flags. User reports and operator background are documented with dated sources in the meta-review below.
| Status | active |
|---|---|
| Operator | Thunder Compute · HQ United States (medium) |
| Founded | 2023 |
| Proof | none published |
| KYC | not documented |
No red flags logged as of 2026-10-02.
A real, operating provider can still lose you money: with a contract you own no hardware, revenue follows coin price and difficulty, and most published contracts are priced so the seller comes out ahead. "Legit" means the operator exists and pays what its terms promise - not that the deal is profitable. 13 dated sources back this page; they are listed under sources.
Same category, scored on the same criteria - so the comparison stays honest. Compare only in the same unit, and read the minimum, KYC and proof columns before switching.
US neocloud selling on-demand and reserved bare-metal NVIDIA HGX H100 clusters from $1.99/hr; acquired TensorDock (2025) and merged with Lightning AI (Jan 2026).
Distributed consumer-GPU cloud with published per-second prices and measured host earn rates; no token, fiat only.
Established US GPU cloud: on-demand H100/B200 instances, 1-Click Clusters and Superclusters, billed per minute with no ingress/egress charges.
Open permissionless marketplace for containerised GPU/CPU compute; providers bid, tenants pay per block.
Ready to try Thunder Compute? Start small, verify one payout, and only scale up once the hashrate or compute actually matches what you paid for.
Based on our checks as of 2026-10-02: Thunder Compute scores 6.7/10 in our weighted review. We logged no red flags. User reports and operator background are documented with dated sources in the meta-review below.
No comments yet - be the first. Posting runs a short proof-of-work in your browser (anti-spam), no account needed.