Nebius has folded Inferize, a stealthy Israeli team founded only in January, into its Token Factory inference business in a deal reported between $100M and $150M.
Inferize’s specialty is the gap between a model being ready and a customer being served. Loading a new model, spinning up extra capacity for a demand spike, or refreshing weights mid-run all leave assigned GPUs sitting idle, a cost its founders call the idle GPU tax.
Their technology shortens those gaps so capacity tracks actual usage more closely, which improves both utilization and the economics of serving a token. Nebius says Token Factory customers should see faster time-to-launch and better unit costs for large models.
The hire is really an acquisition of a team. Founded in January 2026, Inferize had a working prototype within three months. Its engineers now work across Token Factory, starting with integrating their own technology.
Inferize is the latest piece Nebius has bolted onto that platform, after Eigen AI added model, kernel and system-level optimization and Clarifai’s core team brought orchestration.
The deal adds to a busy stretch of consolidation in AI infrastructure, where cloud providers keep buying small teams with narrow technical edges rather than build the capability in house.
