GPU marketplace

Rent a GPU from the TDCC network.

Data centers, micro-clusters, and independent hosts list their GPUs here and set their own hourly price. Pick a machine, send a request, and the host confirms it.

Find a GPUList your GPU
EARLY ACCESSNo card is needed and nothing is charged on this site. When a host accepts your request, you both see each other's email and agree on payment and start time directly. Hosts with a linked TDCC node show as online with hardware reported by the machine itself.
–

GPUs free right now

–

Listings from data centers, clusters, and homes

–

Regions

Just opened

The marketplace is open.

List a GPU in two minutes and set your own hourly price. Listing is free, and you decide which requests to accept.

List your GPU →
GPURegionHostAvailableTrack recordPriceAction
Loading listings…
Exact Core · FPGA cardBit-exact, re-checkable eu-central TDCC Coming exact USDC escrow per job Details

For GPU owners

List your GPU in two minutes.

  1. Fill in the formGPU, how many, region, and your hourly price. You will be asked to create a free account.
  2. Your listing goes liveRenters see it right away. Pause it or change the price any time from your account.
  3. Optional: link your machineRun tdcc market link with the token from your account. Your listing then shows as online with the GPUs the machine actually reports.
  4. Optional: instant API rentalsRun tdcc market api-rentals on and serve with --trust-policy allowlist. Renters you accept then use your models through tdcc for exactly the hours they paid for, with no access to anything else on your machine.
  5. Optional: instant machinesRun tdcc market machines on with Docker installed. Each SSH or Jupyter renter you accept then gets an isolated container with the GPUs they rented, deleted when the rental ends.
  6. Accept the requests you wantYou see each request first. Contact details are shared only after you accept.
your machine

          

        

        

          

            
List a GPU

            
            

              
              
            

            
            

              
              
            

            

              
              
            

            
            
            Free to list. You decide which requests to accept.
          

        

      

    


    

      

        

          
Marketplace FAQ

          
Before you rent.

        

        

          
How does renting work?
Choose a GPU, set how many and for how long, and send the request. The host sees it in their account and accepts or declines. Once accepted, you both see each other's email and the host sends you access details (SSH, Jupyter, or an API endpoint).

          
How do I pay?
By card, through Stripe. When you send a request your card is only authorized. It is charged when the host accepts and released if they decline or do not answer within six days. We hold the money until a day after your rental ends and then pay the host, minus the marketplace fee. If something goes wrong, open a dispute from your account page within 14 days and we will sort it out, including refunds. If you cancel after access started, you get back the hours that had not begun; if the host cancels, you get everything back. Where card payments are not switched on yet, you settle payment with the host directly after they accept.

          
How do I know a host is real?
Listings show the host's star rating from earlier renters, their uptime over the last week, and hardware checked when tdcc on the machine itself reports the GPUs the listing claims. You can message the host from your account page before and during a rental.

          
What is “tdcc API” access?
On listings marked tdcc API, you can skip the back-and-forth: install tdcc, run tdcc auth init once, and paste the Owner ID from tdcc auth status into your request. When the host accepts, your account page shows one command, tdcc client --join …, and the host's models answer at http://localhost:9337/v1 on your computer like any OpenAI-compatible API. Access closes on its own when your hours are up. You use the models the host serves; you do not get a shell on their machine. To keep things fair for everyone on a machine, each renter can have 4 requests running at once by default; your account page shows how much you used.

          
What is an “instant machine”?
On listings marked instant machine, the host's tdcc starts a fresh, isolated container for you as soon as the host accepts: your chosen environment, the GPUs you rented, and only your SSH key let in (or a private Jupyter link). Your account page shows the ssh command when it is ready. Jupyter links use HTTPS with a certificate made for your machine; your account page shows its fingerprint to compare with what your browser reports. Each machine gets its share of the host's memory and CPU and a disk allowance (100 GB unless the host says otherwise). The container and everything in it are deleted when your rental ends, so copy your results off in time.

          
What does “online” mean?
The host linked a running TDCC node to the listing with tdcc market link. The node reports in every minute with the GPUs it actually sees, so you know the machine is real and switched on right now.

          
Is my data safe?
Passwords are stored only as salted hashes, sessions use secure HttpOnly cookies, and your email is never shown publicly. A host or renter sees your email only after a request is accepted. You can delete your account at any time from your account page.

          
What is an Exact Core listing?
Exact Core is TDCC's own accelerator card. Its output can be replayed bit for bit on a laptop, so a renter can check that the work was actually done. Jobs on it will be paid through USDC escrow on Solana.