COMPUTE MEETS MARKET DESIGN
Buy the hours a job needs. Sell the window you can release. Requenta is building one market and execution layer across GPU providers, with prices driven by supply and demand.
One quote. One compute endpoint. One invoice.
A shared execution layer across qualified providers.
01 / TWO SIDES. ONE EXECUTION LAYER.
AI teams need compute for a job. Suppliers have capacity they can release for a window. Requenta standardises the booking and execution across operators.
For AI teams with repeat training, fine-tuning and batch workloads.
Specify GPU type and count, network topology, region and runtime.
See the price before you book. Once accepted, it stays fixed for your booked time.
Run CUDA jobs in an isolated workspace. Retrieve results and pay for delivered compute on one invoice.
For operators and teams with capacity they own or reservations they can resell.
List the machines you can offer and when you need them back.
A booking must clear your net floor after deductions. Unbooked capacity stays yours.
Honour the accepted window. Earn on delivered compute, with no listing fee.
As demand tightens a delivery window, a higher net quote can clear the next supplier’s floor. That capacity enters the market and informs the next quote. Accepted bookings keep their price and availability.
COMPUTE.
WITHOUT THE OVERHEAD.
02 / BUY ON THE MARKET
Prices ease when capacity is plentiful and rise when buyers compete for it. Time flexible workloads for the quieter market.
Training runs, batch inference, experiments. When your job can wait, you can choose a quieter market. Once you book, your rate stays fixed for the accepted hours.
Start with a workload and choose the commercial model that fits.
Flexible workloads.
Better-priced windows.
A compute budget.
Across repeat workloads.
Recurring compute.
A budget you can plan around.
Share your workload to discuss a pilot, hardware fit and commercial terms.
03 / FOR SELLERS
Make your unused GPU computing power available through Requenta. Set a minimum net price and a firm availability window. The market connects that capacity with compatible workloads.
Choose your terms. Your minimum earnings, when bookings can start and when you need your machines back.
Get verified. Connect your GPU machines and complete hardware and performance checks.
Earn on delivered hours. The fee model charges 5.5% on completed third-party transactions. Listing and quoting carry no fee.
AI teams with repeat training, fine-tuning, batch evaluation and inference workloads. On the supply side: professional operators and holders of GPU reservations they can resell, with firm capacity windows.
Prices ease when available capacity exceeds demand. Flexible workloads can choose those windows. As demand tightens, a higher quote can attract more supply by clearing the next seller’s net floor.
The booking model fixes your rate for the hours you accept. Changes in the market affect new quotes. Extra hours require a new quote and an available window.
The initial market design centres on dedicated 8×H100 SXM nodes in Europe. A compatible offer specifies GPU configuration, network topology, region, service class and a firm delivery window.
Yes, if you can resell it and commit the full delivery window. Supplier qualification covers hardware, performance and isolation, as well as your right to offer that capacity.
✳ JOIN EITHER SIDE OF THE MARKET.
Let’s discuss your workload, capacity or investment thesis.