01
Buy credits
You pay in fiat or stablecoins. A gateway sources AVT, permanently burns it, and issues Compute Credits — one credit per CU. You never have to touch a token.
Describe the task. Aventus Compute turns it into metered jobs, runs them across a live network of community-run nodes, and hands back a signed result.
aventus compute
liveRe-check this bridge message before we release it
bridge.attest
9 CU7 owner groups · unanimous · committee tier
7 of 7 owners agree
signedhash matches the source event · safe to release
Describe the work…
Nodes online
0
live and growing — every node bought adds capacity
CPU cores pooled
~0
≈654 TFLOPS peak · ~57 TB RAM
Unbroken uptime
0 days
921,062 blocks, zero halts
Heartbeats signed
0 M
eight per node, every day
Getting real businesses onchain meant building a network of real machines. Most people with work to do don’t need a chain — they need a result they can rely on, and the same network delivers it.
Enterprise adoption
Since 2017 the work has been getting real businesses onchain — across aviation, energy, loyalty and prediction markets. Each one needed infrastructure underneath it that actually held up.
The network it took
Serving them took more than a hosting service. It took thousands of community-run nodes, each one staked, identified and provably alive eight times a day.
Generalised compute
That network does not only run chains. It can run work — one metered unit at a time, for anyone, whether or not a blockchain appears anywhere in the picture.
A Compute Unit (CU) is one core-hour of metered work executed on a live node on the Aventus network — run by a member of its community — and returned with a signed result. Every workload the network performs is priced in CUs — so there is one number to budget against, whatever you are running.
| Workload | What you get | Indicative CU | Availability |
|---|---|---|---|
| Appchain hosting | Node capacity for a dedicated chain — committed capacity, not metered. | Subscription | Live |
| Bridge attestation | An independent re-check of a cross-chain message before it is released. The Aventus bridge is the first workload on the meter. | 8–12 / message | Live |
| Model execution | A deterministic model run — an illiquid-asset price, say — reproduced by a committee of separate machines and signed. | 10–40 / update | In alpha |
| Oracle and data queries | A fetch from a permitted source, signed, with provenance attached. | 1 / query | Next |
| Distributed retrieval | The same request made from N residential networks in M countries — uptime checks, geo-verification, censorship and availability monitoring. | 2–6 / sweep | Next |
| Test execution | A build, release or feature exercised from real residential machines and networks. Load, integration, canary and pre-release runs, signed per run. | 4–20 / run | Next |
| Custom Wasm jobs | Your code, sandboxed, run to a deadline, result signed. | Metered | Next |
CU figures are indicative. Every workload class is benchmarked on real machine hardware, and carries a published ceiling, before it is offered for sale.
Every node on the network runs a light client. That duty uses about 2.5% of what it can do. We never sell the whole machine — the hardware belongs to the people who run it — so the pool available to customers is capped at a quarter of each machine’s cores. That cap alone is a large, standing supply.
One machine on the network
Available cores
0
25% of ~21,800 pooled cores
Daily capacity
0 CU
core-hours available every day
Annual capacity
0 M CU
at today's 3,631 nodes
Marginal power
~0 kW
fleet-wide · no new data centre
Derived from the 31 August 2026 network report: roughly six cores per node across 3,631 nodes. Hardware varies, some owners run several nodes, and owners choose the tiers they take part in.
There are cheaper ways to rent a core. There is no other way to rent a live network of 3,631 community-run nodes, spread across real connections and real places — each one staked, key-registered and continuously proving that it is online.
Work that has to originate outside a data centre — availability checks, geo-differentiated reads, endpoint verification, censorship monitoring — cannot be bought from a cloud region. It can be bought here.
Many owners run more than one node, so committees are selected by disclosed beneficial owner, with a cap per owner group. A result signed by a committee drawn from separate owners carries an assurance no single-operator cloud can offer.
Nothing new gets built to run your work. The nodes are already deployed and running, bought and powered by the people who run them; paid work draws down idle headroom, for a few kilowatts more across the whole network.
Eight signed heartbeats a day, 2.27 million to date. Scheduling relies on liveness that is already recorded on-chain, not on a vendor's status page.
Five steps, the same for every workload class. You see an API and an invoice; underneath, credits are burned and work is assigned across machines that have never met each other.
01
You pay in fiat or stablecoins. A gateway sources AVT, permanently burns it, and issues Compute Credits — one credit per CU. You never have to touch a token.
02
A job names its workload class, its code or policy version, its inputs, a deadline, the assurance tier it needs and the credits it will spend. Credits are reserved before anything runs.
03
Eligible live machines are selected by verifiable random draw, filtered by benchmarked capability and capped by owner group, so no single owner can dominate a job.
04
Machines run the job in a deterministic sandbox. Where your tier requires it, several re-execute independently and must agree before the result stands. Disagreement widens the reported confidence or fails the job safely.
05
You get the output plus a signed receipt: code hash, inputs commitment, machine identities and owner groups, and the credit-burn reference.
Every receipt can be verified with an open verifier you run on your own machine. You never have to take our API’s word for what happened — the CU definition, the receipt format and the verifier are all published.
Writing software has never been cheaper. Being sure about it has never been harder. Aventus Compute runs your builds, releases and canaries from nodes run across the community — real connections, real hardware, real geography — and signs the record of every run.
A cloud test rig tells you the build works from a data centre. These machines tell you it works on consumer bandwidth, behind consumer NAT, from the places your users actually are — and hand back a signed record of which ones ran it.
One release can mean thousands of runs. Pay per run, with no reserved fleet to keep warm and nothing to provision between releases.
Beyond automated runs, machine owners can opt in as a paid pre-release panel: install the build, exercise it, report back. A distributed tester network is not something a cloud vendor can rent you.
We are taking a small number of design partners for the testing beta — pipelines we can benchmark end to end before this opens generally.
Put a pipeline forwardThat symmetry is the whole design. The network pays for its capacity in newly issued AVT on a fixed schedule. Customers pay for that capacity in fiat or stablecoins, and the credits they receive exist only because AVT was permanently destroyed to create them. Utilisation is the dial between the two.
AVT burned
CUs purchased × USD price per CU
÷ disclosed multi-day AVT reference price
Net supply change
emissions (fixed) − burns (usage-driven)
Compute Credits are non-transferable accounting units. They cannot be traded, staked, redeemed for cash or used to claim revenue. They meter capacity and nothing else — they are not a second token.
The 14,625 AVT daily pot and its annual halving stay exactly as published. No job bonus, no variable reward, no new bond. Compute assignments become part of normal node duty.
The 3,631 is a live count, and it grows as people buy nodes. A new licence does not increase emissions; it shares the same pot. What it adds is roughly 1.5 sellable cores — about 13,100 CU a year of new capacity the network can sell.
Horizontal axis: the share of available capacity actually sold. Illustration of the mechanism, not a forecast — held at today’s 3,631 nodes with a fixed price on both sides.
The more of the available pool the network sells, the more AVT is destroyed against a schedule of issuance that never moves. At low utilisation the effect is small. At high utilisation the burn is the largest sink the token has ever had.
Nothing here is a promise about price. It is a description of a mechanism: work sold, credits burned, both published.
One ledger. Issuance, emissions, burns and net supply are published together, in one place. Capacity sold for a customer’s own token is reported as its own line, so fleet utilisation is never confused with AVT burn.
Start metered and pay for what you use. Move to committed capacity when you need guaranteed throughput. Or keep the arrangement that already works for appchains and pay node owners in your own token.
| Path | Best for | How it is priced | AVT effect |
|---|---|---|---|
| Metered credits | Variable or bursty workloads, and anyone starting small. | Pay-as-you-go, per CU, in fiat or stablecoins. Volume bands, no minimum. | Direct burn |
| Committed capacity | Production workloads that need guaranteed throughput. | A monthly minimum for reserved CUs, plus an SLA. | Direct burn, predictable |
| Custom arrangements | Projects that would rather pay node owners in their own token. | Negotiated — the existing appchain arrangement, generalised to any workload. | Reported separately |
Customers get
Work, with a receipt
results from machines you do not have to trust blindly, and a record of exactly which ones ran the work
Node owners get
The same rewards, more duty
the published AVT schedule, unchanged, plus any customer token negotiated for their capacity
AVT holders get
A burn per unit sold
token demand tied to real work performed, not to sentiment
Workload classes open in order, and each one has to prove itself on real hardware first. We put our own bridge on the meter before we asked anyone else to buy the same capability.
Enterprises and projects running a dedicated chain on Aventus tech.
Cross-chain messages independently re-checked before release. We put our own bridge on the meter first.
Markets that depend on a model price with no spot anchor, and need it reproduced by independent machines.
Protocols and venues that need a result with provenance, not infrastructure to run.
Engineering teams who need every build exercised against real machines on real networks before it ships.
Anyone with work a data centre cannot credibly perform. No chain required.
Aventus Compute adds a customer to the network. It does not rewrite the deal anybody already has with it.
14,625 AVT a day, halving annually, weighted by stake and liveness, exactly as published.
No more than a quarter of a machine's cores are ever sold, and owners choose which tiers they take part in.
More jobs completed does not mean more AVT for a machine, and never a share of customer revenue.
Compute Credits are non-transferable, non-redeemable, carry no yield and no governance. A meter, not an asset.
Every class is benchmarked on real hardware and capped per machine before it is ever offered for sale.
Issuance, emissions, burns and net supply are published together, in one place, or not at all.
3,631 live nodes, and growing, that have not missed a beat in five months, using about 2.5% of what they can do. Aventus Compute sells the rest as metered, verified work — priced by the unit, paid in fiat or stablecoins, and returned with a receipt you can check yourself.
Aventus Compute is in alpha. Workload classes, CU prices and availability dates are indicative and will change as classes are benchmarked; economic changes affecting AVT are subject to DAO governance. Nothing here is investment advice or a promise of token price, yield or return. Network figures as at 31 August 2026: 3,631 nodes · ~21,800 cores · 1 CU = 1 core-hour.