Skip to content
ArkNet logoArkNet
Usage-based pricing

Pay per execution,
metered by kernels.

ArkNet charges for the work your program actually performs — measured directly from kernel execution, not wall-clock time. Scale across fragmented GPUs with a mathematically guaranteed max burn.

Check live rates Billing docs
Spot
Per execution
One logical run = one bill
Metering
Kernel-based
Work units, not seconds
Mixed GPUs
Allowed
H100 → 5090 → A100 seamlessly
Max burn

Hard caps per execution, enforced before work begins.

Each execution declares a maximum burn. ArkNet escrows that maximum, runs the execution (potentially across multiple GPUs), then settles the final cost from kernel-metered receipts. Unused escrow is instantly released.

Max burn docs
Set a max burn
Your runner sets a hard cap per execution. If escrow can’t be placed, the execution never starts.
$ARKNET_MAX_BURN=25.00USD
Optional: ARKNET_MAX_BURN_USDC=25.00
Fail closed
No escrow → no execution → no surprise bill.
Predictable spend
Max burn is the absolute upper bound.
Kernel-metered settlement
Providers attest usage as kernel work on specific SKUs. A single execution may include multiple GPU segments; total cost is the sum of metered work.
01
Authorize (hold)
Escrow max burn for the execution.
02
Execute
Run kernels on one or more GPUs (segments).
03
Receipt → settle
Settle kernel-metered usage, release remainder.
Example Receipt
H100
1,240 WU
Attention + matmul
RTX 5090
610 WU
Post-processing
Billedsum(WU × rates)
Most popular

Spot market

For scalable inference and fault-tolerant jobs.

Variableper execution
  • Per execution billing (not time-based)
  • Kernel-metered usage (work-based)
  • Mixed GPU segments allowed
  • Escrow-first settlement with max burn
  • Marketplace rates from hardware owners
View marketplace
Recommended for prod

Reserved cluster

Guaranteed availability for production APIs.

Fixedmonthly rates
  • Priority access guarantees
  • SLAs and support targets
  • Private networking options
  • Monthly invoicing (predictable)
  • Capacity planning assistance
Deploy cluster
Custom solutions

Enterprise

For high-compliance and bespoke deployments.

Custom
  • Full isolation + dedicated capacity
  • Custom identity + networking
  • Compliance programs (SOC2, HIPAA)
  • 24/7 dedicated support options
  • Bespoke GPU configurations
Contact sales

A quick comparison

Cloud bills by time. ArkNet bills per execution using kernel-metered work. “Eqv/hr” helps you sanity-check market rates against legacy providers.

Comparison table
ArkNet spot settles per execution from cryptographic receipts.
ProviderH100 (80GB)A100 (80GB)RTX 4090 (24GB)Billing model
ArkNet SpotExecution-based
$1.85 eqv / hr$1.10 eqv / hr$0.25 eqv / hrPer execution (kernel-metered)
AWS (p5.48xlarge)
$4.10 / hr$3.67 / hr—Per hour
Google Cloud (A3)
$3.95 / hr$3.40 / hr—Per minute
Lambda Labs
$2.49 / hr$1.29 / hr—Per hour

* “Eqv/hr” is a convenience projection from kernel-metered work on recent workloads. You are strictly billed per execution.

Frequently Asked Questions

Straight answers on per-execution billing, kernel metering, and the max burn mechanics.

Ready to deploy?

Start on spot executions, set your max burn constraints, and scale effortlessly across the decentralized marketplace.

Browse providers Understand billing