ce-net / compute

One cloud made of everyone’s spare GPUs.

Datacenters across Europe have accelerators sitting idle between contracts. CE Compute joins them into one mesh, so an AI team books capacity the way it books a cloud, jobs go only where the residency policy allows, and every job comes back with a usage receipt.

List your spare GPUsWatch a job moveWhat would my idle GPUs earn?
Illustration: datacenters across Europe joined into one mesh. Jobs travel from an AI team to spare GPUs and a receipt travels back. AI team
Datacenter with spare GPUsJobReceiptIllustration with example places.

For data-center operators

We buy compute. And we bring you the industry that runs on it.

ce-net runs physical work from end to end: a request comes in, the part is designed, factories and print farms make it, and the simulation, rendering and design behind every step run on GPUs. We book that capacity in your racks, we send you the factories and studios on our network that need it, and we connect your site to the others so a job lands where its data may live.

A factory simulated on ce-net from its real layout: every machine live, every job tracked.
Real footage from our Shenzhen workshop: the robot arm and its twin make the same pick.
Map of the CE Print network: print farms in Europe, North America and China
The CE Print network: 7 print farms, 57 printers, 4 countries. Live map.
A print running on our farm, seen through the printer camera
Our farm in Shenzhen, through the printer camera.

One network, from request to compute

  1. RequestA company sends a part, a product or a production problem.
  2. DesignCAD, PCB layout and physics checks, rendered and simulated before anything is made.
  3. Factories and print farmsThe job goes to the nearest shop or factory with the right machine, watched by camera.
  4. ComputeSimulation, rendering and training run on GPUs in your data center, under the residency the customer set.

We buy your capacity

We book GPU hours for our own simulation, rendering and design work. You set the net price per GPU-hour and the hours you sell.

We send you customers

Factories, print farms and engineering studios on ce-net book through us. You keep 90% of every booked hour.

We connect data centers

Your site joins one network with the others, so a job runs where its data may live and spills over when you are full.

Offer us capacitySee the print network liveOpen the factory twin

What happens to a job

Every step below is checked by the network before the next one starts.

  1. QuoteThe team asks for GPUs, hours and a residency such as EU only. Only approved, unexpired offers inside the policy answer.
  2. ReserveCapacity is locked in one transaction. Two teams can never book the same GPU.
  3. RunThe datacenter runs the job in an isolated environment on its own hardware.
  4. ReceiptThe datacenter closes the job with a receipt: GPU milliseconds, capped by elapsed time and the booked budget.
  5. StatementBoth sides get a statement that splits the gross into supplier share and platform fee.

Example receipt synthetic numbers

GPU time
3 600 000 ms
Rate
€2.50 per GPU-hour
Gross
€2.50
Datacenter, 90%
€2.25
Platform fee, 10%
€0.25

Every statement is checked against the datacenter's receipts before anything is billed.

Residency is a rule the booking has to pass

A team that picks EU only is never quoted a GPU outside the EU. The filter reads the location the datacenter declared and an operator approved.

A location is not GDPR compliance on its own. Real listings also need processor contracts, access controls, deletion rules and incident handling, agreed per datacenter.

  • Inside the EU: can be quoted
  • Outside the EU: skipped for an EU only job
Map of Europe. EU member states are highlighted. An EU only job connects to example sites inside the EU and skips sites in the United Kingdom, Switzerland and Norway.
Example sites.

Where each euro goes

The platform keeps 10% of gross usage. The datacenter gets the rest, calculated in whole micro-euros and rounded once.

€List price for teams: €1.11 per GPU-hour

The fee is a share of gross, so a €1.00 net price lists at €1.11, not €1.10. Prices exclude tax, storage and data transfer. The 10% is revenue before support, payment and security costs.

List your spare GPUs

You keep control of your hardware. You set the price, declare where it sits, and choose the workloads you accept. Nothing runs until you approve it.

Email us your capacity

Opens an email with the five things we need to know. Or fill in the form below.

We answer within one working day. What you type is stored so we can reply; see privacy.

  1. Tell us what is idleGPU model and count, location, hours or share that sits unused, the net price you want and your interruption terms.
  2. Connect one hostA small provider agent discovers NVIDIA through nvidia-smi, AMD through rocm-smi, Apple GPUs and CPU hosts, and reports over ce-net every minute.
  3. Get approvedWe verify the operator and location, then approve the listing. Offers expire after 24 hours unless your host renews them.
  4. Run the first jobsA reference workload runs first in an isolated environment on your hardware, then customer jobs.

Hardware reporting in right now

Try the booking demo

Inventory reported by provider machines through ce-net. Hardware discovery, operator allocation and readiness to run a workload are tracked separately.

—Online providers
—Detected GPUs
0Bookable devices · execution not enabled

Connecting to the live inventory service…

Providers send heartbeats every minute. After three minutes without a heartbeat, their devices are marked offline. The initial development node reports real hardware but allocates no capacity to customers.

Illustrated integration: provider hardware is discovered locally, reports through CE authorization, and appears in the live inventory

Connect different kinds of compute.

Automatic discovery adapters cover NVIDIA through nvidia-smi, AMD through rocm-smi, Apple GPUs through macOS, and CPU hosts. A validated adapter file supports other accelerators. Drivers and operating-system support determine what each host can report.

Integrate through the CE node.

A provider agent discovers hardware locally and sends capability-authorized heartbeats over CE. The marketplace records inventory and publishes only operator-approved device information to this page.

Talk to us about connecting a provider

Try it live

These run in your browser now. They are the same tools we run our own workshop with.

Factory twinWalk a factory floor, follow one lot, change a station and see the flow move.Open →Production line plannerA robot line simulated week by week: stations, operators, bottleneck, cost per unit.Open →Building twinA real clinic's five consultant models merged into one live 3D twin. Click any element.Open →Huey, a robot you buildA two-legged robot kit: 15 servos, an Arduino UNO Q brain and a face made of 104 pixels. We send the parts, you print the shell.Open →3D mold studioOpen, close and turn a printed casting mold in your browser.Open →

Try the booking flow

Synthetic demo: these offers are examples

Try the workflow with illustrative offers. Prices, hardware and locations below are examples, not available supplier listings.

EU filtering uses declared location. Location alone does not establish GDPR compliance. Prices exclude tax, storage and data transfer; live terms are agreed per supplier.

What you get

Built and tested

  • Booking service with per-action authorization
  • Persistent offers and capacity reservations
  • EU location policy and operator approval
  • Usage receipts and the 10% fee calculation
  • Live hardware discovery and heartbeats

How a job starts

  • Tell us the GPU, hours and where the data may live
  • We match you with a datacenter in the network
  • Contract, data-processing terms and invoice set up per customer