If you have tried to rent a new flagship GPU, you have seen it: quotas, waitlists, “contact sales”, or a price that is listed but cannot be deployed anywhere. Here is why, and what we do differently.
GPUs arrive in batches
Data-center GPUs are bought in fixed batches, delivered as complete servers that must be racked, powered, cooled, burned in and imaged before anyone can use them. Each batch is an allocation: a known number of GPUs that becomes available on a known date.
Why clouds oversell or hide stock
Hourly clouds want to show as much capacity as possible, so availability is often hidden behind quotas or approximations. The result for customers is uncertainty: you do not know whether the GPU will be there when your run starts.
What we do instead
- We sell what we have racked. The counts on our site are the GPUs actually free in the current allocation.
- We publish the next date. When a model sells out, the page shows when the next allocation opens.
- Waitlist first. People on the waitlist get the email before units return to public sale.
- Monthly, not hourly. Renting by the month lets us plan allocations precisely, which is part of how we stay at least 25% below the market median.
How to get GPUs before they sell out
- Watch the live stock on the GPU pages or poll the inventory API.
- Join the waitlist for sold-out models from any GPU page.
- For more than 8 GPUs, ask for a private allocation before the batch is announced.