AI inference

Meta Details System for Using Idle AI Inference in Ads
Meta outlines an engineering system that uses idle inference capacity to improve ad recommendations through two elasticity levers.

Nebius acquires Inferize to speed AI inference cold starts
The AI cloud company said Inferize's technology shortens the time large models need to load, cutting idle GPU costs inside its Token Factory inference platform.

Volantis raises $88M to attack AI memory wall with photonics
The semiconductor startup says its photonic interconnect links compute chips to pooled memory, with first inference engines promised to customers in 2027.