Skip to content
ArkNet logoArkNet
The distributed runtime

The computer that
never sleeps.

ArkNet is a peer-to-peer network of GPU providers that forms a single runtime across thousands of fragmented hardware sources — with verifiable execution and market-driven pricing.

Explore the grid Read architecture
Routing
Constraints + bids
VRAM, region, reliability
Verification
Proof-backed
Output matches code + input
Latency
Edge-aware
p95 routing targets
Interactive Lifecycle

Select a stage below to explore the mechanics. A job is dispatched with explicit constraints, optimally routed, executed in isolation, and mathematically verified.

Constraint Engine
Define explicit VRAM, region, and maximum latency limits directly in your kernel manifest.
Mempool Admission
Cryptographic validation of job signatures and parameters before entering the routing queue.
Cost Estimation
Upfront dynamic escrow calculation based on real-time spot market pricing and execution limits.

Validators

Validators maintain the integrity of the ledger. They meticulously verify execution proofs submitted by providers and finalize global state transitions.

  • Verify execution proofs (STARKs)
  • Consensus on global state
  • Slash malicious providers
  • Audit routing + settlement
Validator docs

Providers

Providers lend their silicon to the network. Any hardware owner can join and passively earn yield from idle capacity — from a single RTX 4090 to an H100 cluster.

  • Execute WASM/PTX kernels
  • Cache model weights for warm starts
  • Bid into the marketplace
  • Earn rewards per-second
Become a provider

Global footprint

A single unified runtime spanning regions — continuously optimized for locality, lowest latency, and mathematically verified correctness.

2,400+
Active nodes
42
Countries
84 PB
Total VRAM
99.9%
Uptime (YTD)
Live network metrics. Integrations pull directly from /api/metrics/summary.

Ready to join the grid?

Whether you’re consuming raw compute or supplying it, ArkNet gives you a verifiable, frictionless path from code to silicon.

Become a provider Find compute