meshlinking nodes · 142,391 gpus online
142,391 gpus onlineRent compute
decentralized compute · mainnet

rent the world'sidle gpus.

Mesh turns idle GPUs into one global supercomputer — rent compute on demand, pay per job, no data center.

↵

from $0.31/hr · pay per job · 142,391 gpus online

cluster · live● online
h100 · train
84 nodes
l40s · infer
1,204 nodes
mi300x
312 nodes
4090 · batch
9,811 nodes
jobs matched today2.1M
H100 · 80GB · $1.10/hr ▲A100 · 80GB · $0.62/hr ▬L40S · 48GB · $0.48/hr ▲RTX 4090 · 24GB · $0.31/hr ● hotMI300X · 192GB · $0.88/hr ▬GH200 · 96GB · $0.95/hr ▲A10 · 24GB · $0.19/hr ▬L4 · 24GB · $0.22/hr ▲H100 · 80GB · $1.10/hr ▲A100 · 80GB · $0.62/hr ▬L40S · 48GB · $0.48/hr ▲RTX 4090 · 24GB · $0.31/hr ● hotMI300X · 192GB · $0.88/hr ▬GH200 · 96GB · $0.95/hr ▲A10 · 24GB · $0.19/hr ▬L4 · 24GB · $0.22/hr ▲
compute / instances

every gpu,
on demand.

Rent by the hour or the job — from a single 4090 to a rack of H100s. No minimums, no contracts, no data center.

live / job queue

watch the mesh
work.

Every job is matched to idle nodes, computed, and proven — on chain. This is the feed.

mesh · live job queue● streaming
H100 · training run · 8 nodesrunning
A100 · fine-tune · 4 nodespaid
L40S · inference · 512 batchrunning
MI300X · embedding · 96GBpaid
RTX 4090 · render · 2,400 framesrunning
gpus online142,391
jobs / day2.1M
avg job time11s
$ settled on chain$4.6M
utilization86%
gpus online
jobs run / day
avg job time
of supply staked
the network

142,391 gpus,
one mesh.

nodes online
142,391
countries
94
avg payout
$0.84 / hour
proofs checked
412M
how it works

from ask to
answer.

01

request

Describe the job — model, batch, budget. The scheduler reads it in seconds.

02

match

Your job is matched to idle nodes with the right silicon and a proof stake.

03

compute

Nodes run the work. Every step is logged and hashed for verification.

04

prove & pay

A proof is checked on chain, escrow releases, and $MESH settles instantly.

benchmarks

faster, and a
lot cheaper.

Independent runs, published weekly. Scores relative to a hyperscaler baseline of 100 — the higher, the better.

training throughput
92
inference latency
71
render · 4k
83
cost per job
66
manifesto

compute should
be a commons.

The world has more idle GPUs than active ones — in gaming rigs, render farms, and quiet crypto miners. Meanwhile labs and startups queue for weeks to rent silicon they'll only use for an hour.

Mesh wires those machines into one network, pays their owners for every verified second, and lets anyone rent the world's spare compute for the price of the electricity it burns. No data center. No gatekeeper. Just math, proof, and $MESH.

— Soren VossFounder, Mesh

roadmap

where the mesh
is going.

2026 · Q1now

mainnet · 142k gpus · proof-of-compute live

2026 · Q2next

$MESH staking + node rewards go live

2026 · Q3next

gpu rental for inference at edge locations

2026 · Q4later

mesh SDK · submit jobs from your own code

2027 · Q1later

federated training across 100k nodes

faq

short answers.

What is Mesh?+

Mesh is a decentralized GPU network. Anyone with spare silicon earns $MESH for running jobs; anyone who needs compute rents it by the job, proven on chain.

Who can run a node?+

Anyone. A modern gaming GPU, 16GB of VRAM, and a stable connection is enough to start earning. No KYC, no datacenter.

How is compute verified?+

Nodes submit proofs-of-compute that are checked by the network before escrow releases. If a node lies, its stake is slashed.

What does it cost?+

You pay per job or per hour, from $0.31/hr on a 4090. No minimums, no reserved capacity, no surprise bills.

Is this like cloud compute?+

It is the same silicon — you just rent it from a network of owners instead of a hyperscaler. Prices are set by supply, settled instantly in $MESH.

join the mesh.

Get early access, rent compute, or run a node. 142,391 gpus are already online.