Meta launched a service to sell spare AI compute capacity, reported alongside an expanded deal with an accelerator manufacturer and the launch of its model line.
Selling surplus in-house capacity is how the market's largest cloud businesses began: infrastructure built for internal use that, once it exceeded need, became a product. The pattern repeating suggests build scale has overtaken predictable internal demand.
For buyers, another seller of capacity is good news in a tight market, particularly one that isn't among the three traditional providers.
One point deserves careful reading in any arrangement like this: spare capacity is, by definition, what's left after internal use. Priority at peak belongs to whoever built it, which matters to anyone depending on the service to operate.
The move also joins the company's efforts on other fronts in the same period, including publishing model weights and launching a coding tool. Three fronts with one shared objective: contesting position without charging for model access.