Meta shipped its model line, expanded a deal with an accelerator manufacturer, and created a service to sell spare AI compute capacity.
Selling surplus is how the largest cloud businesses began: infrastructure built for internal use that, once it exceeded need, became a product. The pattern repeating suggests build scale overtook predictable internal demand.
For buyers, another seller of capacity is good news in a tight market, particularly outside the three traditional providers.
The point to read carefully in any such arrangement: spare capacity is what remains after internal use, and priority at peak belongs to whoever built it.
The move joins the company's other fronts in the same period, including publishing weights and launching a coding tool. Three routes with one shared objective: contesting position without charging for model access.