For builders
Launch an AI business without buying GPUs.
One H100 costs roughly $30,000. Launching an AI network on Tenseuron costs 0.0356 SOL — a single transaction. You define the work; independent miners bring the hardware.
What does it cost to launch an AI network?
One Solana transaction, quoted before you sign. Most of it is not our fee at all — it is on-chain rent that Solana itself charges to store your token and its metadata forever. The protocol fee is the only part that comes to us.
| Component | SOL | Who charges it |
|---|---|---|
| Token mint account rent | — | Solana |
| Token metadata rent | — | Solana |
| Associated token accounts | — | Solana |
| Transaction fees | — | Solana |
| Tenseuron protocol fee | — | Tenseuron |
| Total | 0.0356 |
Quoted live from the same endpoint the create flow uses, so this figure is the one you will actually be asked to approve. There is no subscription, no seat pricing, and no monthly minimum — the cost of launching does not recur.
How is this different from buying or renting GPUs?
The distinction that matters is what you are buying. Cloud and GPU-rental markets sell you machine time — you still choose the model, size the fleet, handle scaling, and pay for capacity while it sits idle. Tenseuron sells you a graded result. You publish a task, miners compete to answer it, validators score the answers, and only accepted work is paid for.
| Buy a GPU | Rent GPUs | Tenseuron | |
|---|---|---|---|
| Upfront | ~$30,000 per H100 | Card on file | 0.0356 SOL, once |
| You pay for | The hardware, idle or not | Machine time, idle or not | Accepted results only |
| You operate | Everything | Model, scaling, uptime | Nothing |
| Quality control | Yours to build | Yours to build | Validator consensus |
| Scales by | Buying more cards | Raising spend | Attracting more miners |
The trade is real in both directions. Renting gives you exact control over which model answers and how fast. A market gives you no fleet to manage and no idle cost, but you are competing for miners rather than commanding machines — which is the honest catch, covered below.
How the network actually works
Four roles. You submit tasks and lock the network's token to fund them. Miners poll for open work and run it on whatever model they choose — a hosted API or a model on their own machine. Validators score the submissions. Settlement happens on Solana.
Launching follows the curve: the token opens on a bonding curve, the SOL raised accumulates as future liquidity, and at graduation that raise converts into a real AMM pool while the network opens for work.
What does it cost to run per task?
Each completed task consumes a small amount of the network token locked against it. The per-task cost is derived from the network's configured budget rather than billed per token of input and output, so it does not move with prompt length.
Every wallet also gets its first three tasks free on any network, enforced from real task history — so clearing your browser does not reset it, and you can evaluate a network before locking anything.
How miners get paid
This is the part worth understanding before you launch, because it decides whether your network attracts supply. The reward for a task is a fraction of the network's remaining reward pool, not a fixed emission:
reward = max(1e-9, poolBalance × 5e-9) × tierMultiplier split = miner 50% · validators 30% · creator 10% · launchpad 10%
Two consequences follow, and both are structural rather than promises. Because the payout is proportional to what remains, it shrinks as the pool drains and the pool cannot be emptied — it decays toward a floor instead. And because the split totals exactly 100% of that reward, the network never pays out more than it takes in. Trading activity refills the pool, so a network with real usage pays its miners better than a dormant one.
Tasks that use tools or run in thinking mode carry a higher multiplier, so running a more capable model is worth more — the market sorts harder work toward stronger models rather than paying every task the same.
Is anything actually built on this?
Yes — ours. WalletAI is a self-custody Solana wallet in the Chrome Web Store whose AI screens tokens for honeypots, proposes trades, and guards open positions. Every one of those is a task on a Tenseuron network.
The extension calls zero AI provider endpoints. There is no OpenAI key, no Anthropic key, no DeepSeek key anywhere in the shipped bundle. Inference goes through the compute API, out to independent miners, and back through validator consensus — the same path any network you launch would use.
We think that matters more than any claim on this page: it is a product handling real money, and it runs on the thing we are asking you to build on. Read how it is wired.
Who this is not for
The network is early, and there are cases where you should use something else. Being straight about them saves you the launch fee:
- You need a latency guarantee. Tasks are answered by independent miners who poll for work. If your product needs a hard sub-second response every time, call a hosted API directly.
- You need to control which model answers. Miners choose their own models. Consensus filters bad answers, but you are specifying a job, not a model version.
- You are handling confidential or regulated data. Tasks are distributed to independent miners, which means your prompt is processed on hardware you do not control. Do not put client data, PHI, or anything under an NDA through a public network — this one or any other.
- You have no workload yet. A network with no tasks pays no miners and attracts none. Launch to serve demand you already have, not to create it.
How to launch, step by step
- 1
Define the network
Name it, describe the task type, and set the token supply and economics. This is the part that decides whether the network is useful: a narrow, well-specified job attracts better answers than a vague one.
- 2
Pay one on-chain fee
Approve a single Solana transaction in your wallet. This mints the token, writes its metadata, and creates the token accounts. The exact amount is quoted before you sign.
- 3
The token launches on a bonding curve
Early buyers price the token as the curve fills. The SOL collected accumulates as the network's future liquidity — it is not a fee, it becomes the pool.
- 4
Graduate to a live market
When the curve fills its threshold the raise converts into a real AMM pool and the network opens for work. Miners can join and start serving tasks.
- 5
Submit work and get graded results
Send tasks through the compute API or the OpenAI-compatible endpoint. Miners answer, validators score, and the accepted submission comes back to you.
Frequently asked questions
Do I need to buy or rent GPUs to launch an AI business?
No. You define the network — what it does, which models qualify, how it pays — and independent miners bring their own hardware or API keys to serve the work. You never buy, rent, or reserve a GPU. Miners are paid from the network's own reward pool in the network's own token, so your costs do not scale with a compute bill.
What does it actually cost to launch an AI network?
A single Solana transaction. Most of it is unavoidable on-chain rent for the token mint, its metadata, and the associated token accounts; the Tenseuron protocol fee is a fixed 0.02 SOL on top. There is no subscription, no seat pricing, and no monthly minimum. The live quote is shown on this page and again in the create flow before you sign.
How is this different from renting GPUs on a cloud or a DePIN network?
Renting GPUs means you still operate the service: you pick the model, size the fleet, handle scaling, and pay for idle capacity. Tenseuron is a market, not a rental. You publish a task, miners compete to answer it, validators score the answers, and only accepted work is paid. You are buying a graded result, not machine time.
What stops a miner from returning garbage?
Every submission is scored 0–100 by independent validators running their own models. The verdict is a stake-weighted median, so no single validator decides the outcome, and validators whose scores drift from the median can be slashed. Only the accepted submission is paid.
Do I need my own users to start?
No, but you do need a reason for the network to exist. A network with no tasks pays no miners and attracts none. The honest sequence is to have a real workload first — even a small one — and launch the network to serve it, rather than launching and hoping demand appears.
Can I use my existing OpenAI client code?
Yes. Alongside the native endpoint there is an OpenAI-compatible surface at /v1/chat/completions and /v1/models, where AI networks appear as model IDs. Existing OpenAI client libraries work unchanged by pointing the base URL at Tenseuron.
Who owns the network after I launch it?
You are the creator and receive a permanent share of every task the network serves, for having launched it. The token trades on its own bonding curve and, after graduation, on an AMM pool. Miners and validators are independent participants, not employees.
What happens if nobody mines my network?
Tasks sit unserved. Miners choose which networks to join, and they can set a cost floor so unprofitable tasks are never shown to them. A network competes for miners on how much it pays, which is a function of its reward pool — so a network with a thin pool and no trading activity will struggle to attract supply.
Launch it
Define the network, approve one transaction, and it is live. If you want to read the protocol first, the docs cover architecture, consensus, and the compute API in full.
Create your AI network