First-party clouds United States · Live data

Thunder Compute.

First-party GPU cloud billed per minute. A virtualized GPU tier (RTX A6000, L40, L40S, A100 80GB, H100) undercuts the dedicated passthrough tier on every card, and instances attach straight to VS Code or Cursor via the tnr CLI. Storage is billed separately per GB-hour and data egress is free. Crypto mining is prohibited by their terms of service.

At a glance

Business model: First-party clouds
Tier: First-party cloud
Trust tier: Tier 2
Headquarters: United States
Pricing model: On demand
Min rental: 1 min
Mining: Prohibited

Instance menu

Every Thunder Compute configuration we track

Per-GPU price is the comparable number; the instance total is what you're billed for the whole machine. Storage, extra vCPU, and egress are priced separately by most providers.

GPU	Configuration	GPUs	vCPU	RAM	Disk	$/GPU/hr	$/hr total
Nvidia RTX A6000	a6000_x1	1	6	48 GB	100 GB	$0.35/hr	$0.35/hr
Nvidia L40	l40_x1	1	6	48 GB	100 GB	$0.79/hr	$0.79/hr
Nvidia L40	l40_x2	2	12	96 GB	100 GB	$0.79/hr	$1.58/hr
Nvidia L40	l40_x4	4	24	192 GB	100 GB	$0.99/hr	$3.96/hr
Nvidia L40	l40_x8	8	48	384 GB	100 GB	$0.99/hr	$7.92/hr
Nvidia A100 80GB PCIe	a100xl_x1	1	8	64 GB	100 GB	$1.09/hr	$1.09/hr
Nvidia A100 80GB PCIe	a100xl_native	1	8	64 GB	100 GB	$1.79/hr	$1.79/hr
Nvidia A100 80GB PCIe	a100xl_x2	2	16	128 GB	100 GB	$1.09/hr	$2.18/hr
Nvidia A100 80GB PCIe	a100xl_x4	4	36	288 GB	100 GB	$1.49/hr	$5.96/hr
Nvidia A100 80GB PCIe	a100xl_x8	8	96	768 GB	100 GB	$1.49/hr	$11.92/hr
Nvidia H100	h100_x1	1	4	32 GB	100 GB	$3.20/hr	$3.20/hr
Nvidia H100	h100_native	1	4	32 GB	100 GB	$3.20/hr	$3.20/hr
Nvidia H100	h100_x2	2	8	64 GB	100 GB	$3.20/hr	$6.40/hr
Nvidia H100	h100_x4	4	60	480 GB	100 GB	$3.20/hr	$12.80/hr
Nvidia H100	h100_x8	8	120	960 GB	100 GB	$3.20/hr	$25.60/hr

When to pick Thunder Compute

Best for

Per-minute billing — a 12-minute fine-tune costs 12 minutes, not a rounded-up hour.
Interactive development straight from VS Code or Cursor over Thunder's CLI.
Cheap single-GPU work — the virtualized tier undercuts the dedicated rate on every card.
Bursty experiments where you stop the instance the moment the job finishes.

Avoid for

Crypto mining — Thunder's terms prohibit it outright: instant ban, credits revoked, usage billed in full.
CUDA features the virtualized tier can't run — managed/unified memory, MPS, GPU-accelerated FFmpeg, and Nsight-class profilers. Take the dedicated tier if you need them.
Multi-node distributed training — instances top out at 8 GPUs with no InfiniBand fabric between hosts.
Workloads that need guaranteed H100 capacity — availability is fleet-wide and does read unavailable.

Price history

Daily median on Thunder Compute's top GPUs.

GPUs available on Thunder Compute

GPU	Tier	VRAM	$/hr	Range	Offers
Nvidia RTX A6000	workstation	48GB	$0.35/hr	$0.35/hr–$0.35/hr	1	Compare →
Nvidia L40	datacenter	48GB	$0.79/hr	$0.79/hr–$0.99/hr	3	Compare →
Nvidia A100 80GB PCIe	datacenter	80GB	$1.09/hr	$1.09/hr–$1.79/hr	5	Compare →
Nvidia H100	datacenter	80GB	$3.20/hr	$3.20/hr–$3.20/hr	0	Compare →

Compare with peers

Other providers in the same bucket — quick way to sanity-check pricing before committing.

RunPod

93% host

Cheapest: $0.12/hr · Nvidia RTX A2000 · 46 GPUs

Community Cloud lets independent hosts list GPUs. Consumer-friendly onboarding, weekly USD payouts via Stripe Connect.

Lambda Labs

No fresh listings in window

First-party AI cloud — Lambda owns and operates the GPUs. Transparent on-demand per-GPU pricing on B200, H100, GH200, A100. No marketplace commission, no hos...

DeepInfra

Tier 1

No fresh listings in window

First-party AI cloud — DeepInfra also rents B200 GPU instances on demand (1x/2x/4x/8x configs). Same provider as the hosted-inference API, just a different s...

Novita AI

Tier 2

Cheapest: $0.33/hr · Nvidia GeForce RTX 4090 · 4 GPUs

AI and agent cloud platform providing on-demand GPU rentals for training and inference workloads with high performance and cost efficiency.

Frequently asked

How is Thunder Compute billing measured?

Most providers bill per-second once the instance is running, with a small minimum (often 60 seconds). Some first-party clouds round up to the minute. Headline $/hr is the right comparison unit.

Which regions does Thunder Compute offer GPUs in?

P2P marketplaces aggregate hosts worldwide so region varies per offer. First-party clouds and hyperscalers expose explicit region pickers (US, EU, APAC). Filter on the provider's site after clicking through if region matters.

Does Thunder Compute offer an SLA?

Hyperscalers (AWS, GCP, Azure, Oracle) publish formal SLAs. First-party clouds (Lambda, CoreWeave) offer support contracts. P2P marketplaces and decentralized networks have no SLA — uptime depends on the individual host.

What's the refund / cancellation policy on Thunder Compute?

Per-second billing means you only pay for compute used — stop the instance and billing stops. Pre-paid credits and committed-use discounts have provider-specific terms; check the provider's billing docs before pre-paying.