The plan

The power grid
for computing.

A century ago, every factory ran its own generator. Then the grid arrived: you plug in, a meter runs, and you pay for what you use. Computing is still in the generator age. Cirrun is building its grid.

Not a cheaper GPU finder

We're changing how the world uses machines.

Finding the cheapest GPU is where we start. The point is eight shifts that change computing for everyone who uses it, supplies it or pays for it.

  1. Renting machinesBuying finished work
  2. Hourly metersBy the second, nothing for idle
  3. One cloudOne grid of every provider and every spare machine
  4. A few ownersAnyone can supply compute and earn
  5. Hidden pricesAn open market with a public price
  6. People clicking buttonsAgents renting on their own, inside limits
  7. Any power, any timeCompute that follows clean energy
  8. Just computersEvery kind of machine, rentable by the second

Why Cirrun is different

Clouds rent you machines. Cirrun runs your work.

Every compute company sells you time on a machine, and what happens on it is your problem. We sell the finished run. That one change flips everything else.

QuestionMost compute todayCirrun
What you buyHours on a machineYour work, done
When your code stopsThe meter keeps running until you shut it downThe machine turns off that second
When a machine failsYour progress often goes with itYour work was saved the whole time
Which providerOne, and moving is painfulThe cheapest of many, picked for every run
The priceYou find out on the billShown before you start, itemised after
Whose side it's onThe company that owns the machinesYours. We own no machines, so we only win by finding you the best one
AI modelsBring your ownTop-tier Quantavane models, free
AI agentsBuilt for people clicking buttonsBuilt so agents can rent safely, inside hard limits

Some providers already do one or two of these. Nobody does all of them, across every provider, for every kind of machine.

How the grid works

Four parts, like the one in your wall.

  1. Power stationsEvery provider, and anyone with spare machinesBig clouds, GPU clouds, small data centres and gaming PCs through Cirrun Host.
  2. The operatorThe smart layerSends each job to the cheapest machine that will finish it, and moves it if something fails.
  3. The meterBilled by the secondThe price before you start, a receipt when you finish. Nothing for idle time.
  4. The marketA price for the world's computeAn open price index, booking ahead and cover for big runs, all built on real run data.

The mission

End wasted compute.

Machines that are paid for but doing nothing. Runs lost to a crash. Hours billed after the code stopped. Spare GPUs sitting dark while others can't afford one. Every one of these is waste, and the grid removes it.

  • We'll count it in publicFrom the alpha on: money saved, idle hours avoided and runs rescued, from real receipts.
  • We win when you waste lessWe own no machines, so idle time earns us nothing. Our only job is finishing your work for less.
  • Cheaper every yearEvery run makes the grid smarter about price and reliability, so the next run costs less.

Building the grid, level by level

One promise first. Then everything it makes possible.

1

Level 1 · Alpha, November 2026

Never lose a run. Never pay for idle.

The cheapest machine, picked for youAcross five providers to start, with the price and your limit shown first.
Your work saved the whole timeCopied to your own storage while it runs, double-checked before any machine is switched off.
Off the second you're doneNo idle hours, no forgotten machines, no overnight bills.
A receipt for every runTime used, the real machine price and our fee, nothing hidden.
2

Level 2 · Public launch, January 2027

Compute for everyone, wherever they work.

Everywhere you workPython, notebooks, VS Code, GitHub, and AI agents that rent machines safely within your limits.
Pick up where it stoppedIf a machine fails, your work moves to the next cheapest one and carries on.
Copy a run, try ten ideasSplit one run into many versions, each on the cheapest machine, and keep the winner.
Go back in timeReturn to any saved point, change your code and continue from there.
One-command recipes"Fine-tune a model on my data", "transcribe 10,000 files": ready to run, shared by the community.
Free top-tier modelsFrom Quantavane, our sister company, ready to use on day one.
3

Level 3 · 2027

One bill for any machine. Anyone can supply one.

One bill across every cloudPay Cirrun once, run anywhere, with prices that never surprise you.
Cirrun HostGaming PCs, spare servers and small data centres earn by the second doing real work.
Storage by the secondRent space only while you need it, including drives other people aren't using.
Slices of big GPUsA piece of a top data-centre GPU for small models, notebooks and tests.
Put a machine to sleepPause a whole machine, memory and all, and wake it later anywhere, paying only for storage.
Reliability you can seeEvery provider and host scored from real runs; unreliable machines are avoided for you.
4

Level 4 · New products on the same grid

Not one product. A whole platform.

Remote buildsHuge processors for a few minutes, so slow builds and tests finish in a fraction of the time.
Voice and video, close to your usersAI that answers in real time, running near the people using it.
Data pipelinesCollect and clean data on the cheapest processors, paying only while they work.
Security training labsSealed practice environments for security teams, created in seconds and gone afterwards.
Teams and companiesBudgets, rules like "customer data stays in the EU", a sealed private tier and a full record of every run.
Greener by defaultWork moves to cleaner power when it can, with the carbon shown on your receipt.
5

Level 5 · The market

Machines nobody else rents. A price for the world's compute.

Giant clusters for an afternoonBook hundreds of connected GPUs for a few hours, like booking a meeting room.
Enormous memory, by the secondMachines with terabytes of memory, rented for minutes instead of months.
Robots and devicesReal robot arms, drones and edge boards in partner labs, rented by the minute.
Rare and new chipsEarly access to experimental hardware from research partners.
Compute you can book aheadLock in capacity and prices for the future, so big plans have certain costs.
Insurance for big runsCover for long, expensive training runs, priced from real reliability data.
The world's price indexReal prices and speeds from every run, open for anyone to check.
An open receipt standardOne honest receipt format the whole industry can use.
6

Level 6 · The 2030s

The world's compute grid. Running on the world's spare power.

Compute follows energyWork flows to wherever clean power is cheapest right now: a windy night in one country, a sunny afternoon in another.
Every idle machine earnsMillions of homes, offices and data centres supplying the grid, each paid by the second.
Agents rent most of itAI agents buying compute for themselves, safely, inside budgets their owners set.
The reference priceWhen anyone asks what an hour of compute costs, the answer comes from the Cirrun index.

Everything on the grid

If it's a machine, you'll be able to rent it.

One account, one meter, one receipt, for every kind of machine. Partners plug new ones in through an open connector, so the catalogue keeps growing.

Compute
  • GPU slices to giant clusters
  • CPUs and huge-memory machines
  • Macs by the second, for iOS builds
  • ARM servers and FPGAs
  • New AI chips
  • Supercomputer time
  • Quantum computer time
Storage and data
  • Storage by the second
  • Spare drives on Hosts
  • Archives
  • Licensed datasets
AI
  • Free Quantavane models
  • Community fine-tunes
  • Inference endpoints
  • Evaluation runs
Devices
  • Real phones
  • Every browser
  • Smart TVs
  • Hardware boards
Robots and labs
  • Robot arms and drones
  • Automated cloud labs
  • 3D-print and CNC farms
  • Lab instruments
Creative
  • Render farms
  • Video transcoding
  • Game servers by the second
Science and space
  • Telescope time
  • Satellite ground stations
  • Large simulations
Environments
  • Sandboxes
  • Security training labs
  • Full dev environments

New things only a grid can do

Structures that don't exist yet.

The Compute ExchangeLive prices, booking ahead and later futures, so compute is bought like energy, not haggled over.
The Compute PassportOne identity, reputation and history that works across every provider and Host.
Host cooperativesA university, an office or a neighbourhood pools its machines and shares what they earn.
Grid SitesMachines at solar and wind farms, running on power that would otherwise go to waste.
Country gridsCompute that stays inside a country's borders, for governments and regulated work.
The Recipe EconomyCreators earn every time someone runs a recipe they made.
Pay for the resultPost a job and a price. The grid delivers the result, or you don't pay.
Compute allowancesGrants, scholarships and team budgets given as compute, for people and for AI agents.
The startup factoryNew companies built on the grid, paid in compute for a stake, each running on Cirrun and Quantavane models from day one.
Cirrun Founders programmeCompute credits, engineering help, free models and investor demo days for startups anywhere in the world.
The Cirrun AcceleratorCohorts twice a year for the most ambitious founders, building on the grid from day one.
The next generationCompute for students, hackathons, competitions and young founders, so talent never waits for a budget.
Financing for HostsGPUs for small data centres and Hosts, paid back from what they earn on the grid.
An open standardThe job and receipt formats, run by an independent foundation, so every workload can move freely.
The Waste CounterA live public count of money and idle hours the grid has saved.
The Promises LedgerEvery promise we make in public, with its status: kept, in progress or missed, and why.

Why it gets stronger

Every run makes the next one cheaper.

The grid feeds itself. That's what makes it hard to copy: a newcomer starts with no runs, no data and no hosts.

  1. More runsteach us which machines are fast, cheap and reliable
  2. Better routingfinishes work for less and fails less often
  3. Lower pricesbring more people, companies and agents
  4. More demandbrings more providers and hosts, and more runs

The Canli family

Built-in demand. As they grow, we grow.

Every company in the family, and every company Quantavane starts, runs on Cirrun. Their growth fills the grid; a bigger grid makes their compute cheaper. Two engines turning each other.

QuantavaneTop-tier AI models, free for everyone on Cirrun. Every Quantavane company, in every science and sector, trains and runs on the grid.
CanliCapitalFinancial research and data, computed on Cirrun.
TraceAxiomChecks that AI agents did what they claim, run on Cirrun.
GIMBIXThe cheapest way to use AI models, served from Cirrun.

If you use more than one Canli company, you get shared credits and perks.

Who the grid is for

Compute for the next billion builders.

A student, anywhereFine-tunes a free Quantavane model from a borrowed laptop, on a GPU that costs less than lunch, and keeps every result.
A two-person startupShips an AI product without a cloud contract, a DevOps hire or a surprise bill.
A lab without a budgetRuns the experiment it couldn't afford, on machines that would otherwise sit dark.
Someone with a gaming PCEarns while they sleep, by lending it to the grid through Cirrun Host.

The destination

Every machine on Earth,
one grid, one command away.

The cheapest, safest and most honest way to use a computer you don't own, for every person, company and agent on the planet. It starts with the alpha.

Free during the alpha · No card · Your email stays private

Levels 2 to 6, the catalogue and the new structures are our plan, not live features. We will say clearly when each one ships.