There is less of a GPU shortage than you have been led to believe. Data centers have underutilized capacity, but it comes in a form that most orchestration software is not capable of using: a few minutes here, a few hours there.

Once those unused minutes have passed, they can never be reclaimed. Like electricity that cannot be stored, or produce that will spoil, compute capacity has an expiration date after which its value drops to zero. We function as a spot market for this perishable compute, ensuring that otherwise wasted capacity finds productive use before it expires.

To accomplish this, we built custom scheduling and orchestration software that aggregates these small chunks across data centers to run AI models. As a buyer of last resort for this otherwise wasted capacity, we secure steep discounts from data centers and pass these savings directly to you.

Why?

Inference.net is the marketplace where humanity discovers the moment-to-moment price of turning energy into intelligence. We believe efficient markets for AI inference will drive the widespread proliferation of artificial intelligence over the next decades, leading to unprecedented human flourishing on Earth and beyond.

We aim to accelerate this process.

— The Inference Team