Pay for the seconds
you actually use
Usage-based GPU rates for renters and provider earnings, metered per second with clear pricing.
$
Every accelerator, every price
Sample renter estimates. Marketplace prices move with supply and demand.
Per GPU-hour, billed per second · no minimums · egress is free · excludes storage & tax · updated Jun 2026
What it actually costs
Example jobs, estimated bills
Per-second pricing is abstract until you see a rough receipt. These examples show the intended comparison pattern, not a live quote.
Fine-tune a 7B model
8x H100 · 10 hours
Overnight training run
1x H100 · 12 hours
H100 80GB. Illustrative Kracht rates and hyperscaler reference pricing; verify current prices before production use.
Plans
Start free, scale when you're ready
Renter pricing stays metered per second. Choose on-demand for flexibility or reserved terms for steady workloads.
1-year reserved saves about 22% versus on-demand for sustained workloads.
Free
For experiments and learning. No card required.
- $5 starter credit
- On-demand GPUs
- Per-second billing
- Community support
Reserved
Reserved discounts for steady workloads; exact per-GPU rates stay in the table.
- 1-year reserved
- Priority capacity
- Reserved scheduling
- Email + chat support
Enterprise
Dedicated fleets and terms for teams at scale.
- Volume discounts
- SLA options
- SSO & audit logs
- Dedicated support engineer
List hardware
Have idle GPUs? See the provider path.
Provider estimates are separate from renter costs. List spare hardware, set availability, and estimate earnings when demand matches.
Example payout
LISTENINGsample demandIllustrative H100 scenario at high utilization. Actual earnings vary by demand, pricing, and availability.
Provider readiness
Listing starts with trust checks
Hardware listing should make the provider path obvious without promising instant production availability.
- Verify identity. KYC + payout account before any listing goes live.
- Benchmark hardware. Automated checks confirm the GPU performs as listed.
- Meet demand. Vetted renters matched to your floor price.
Reliability
What you pay for actually stays up
FAQ
Questions, answered
Compute is metered per second and drawn from your prepaid credit balance. You only pay while an instance runs.
No egress fees, no minimums, no platform fees. The rate you see is the rate you pay.
Up to ~38% versus on-demand for sustained workloads, with priority scheduling and capacity planning.
Cards, iDEAL and SEPA via Mollie, plus invoicing for Enterprise accounts. Payments are not switched on yet, so nothing is charged today.
On-demand has no commitment. Stop your instance and billing stops. Reserved terms run until the period ends.
Still curious? There's a lot more.
Browse all FAQs →Two sides, one marketplace
Spin up the GPUs you need - or earn from the ones you already own
Rent on-demand compute by the second, or list idle hardware and let it pay for itself. Same marketplace, both directions.
Join 12,000+ builders / $5 free credit / no card required
Stay in the loop
Product updates and early access to new GPU classes. No spam.