Nebius, the artificial intelligence cloud provider listed on Nasdaq as NBIS, has acquired AI startup Inferize. The company said the deal is intended to use Inferize’s technology to shorten the time required to launch and scale large AI models and make inference workloads more flexible.
The deal targets cold-start delays in large-scale AI deployments
In production-grade deployments of large AI models, so-called cold-start latency from model loading and startup can prevent hardware resources from being used right away. According to the report, when systems spin up new instances to handle traffic surges, or when model weights are updated during reinforcement learning, provisioned GPUs often remain idle while waiting for data to load.
To keep services stable, companies usually have to maintain extra standby compute capacity. That leaves part of their GPU allocation unused before real demand arrives and pushes infrastructure costs higher.
Inferize built technology to reduce startup and scaling time
Inferize developed technology aimed at cutting the time needed for model startup and dynamic scaling. Nebius said that can reduce the cost burden created by idle compute capacity and allow infrastructure to track actual demand more closely.
As part of the acquisition, Inferize’s technology and engineering team will be integrated into Nebius Token Factory, the company’s managed inference platform.
Nebius says Token Factory will gain broader optimization layers
Nebius Chief Technology Officer Shtan said efficient inference services depend on how quickly system architecture can react to shifting demand. With Inferize’s engineers and technology joining Nebius Token Factory, the company said it will be able to optimize its AI computing stack across multiple layers.
Nebius said Eigen AI had already provided model- and core-level tuning for the platform, while Clarifai added system-level inference and compute orchestration. After integrating Inferize’s technology, Nebius said it expects to supply available hardware capacity more quickly when compute demand changes.
Inferize team to join future development work
Inferize CEO Bortnikov said the team’s goal has been to improve the utilization of every GPU by removing the cost of standby capacity, which he said was the company’s founding purpose.
According to the report, Inferize was founded in January 2026 and completed its technology prototype within three months of launch. Nebius said Inferize engineers have deep GPU engineering experience and will take part in development work across Nebius Token Factory as the two sides move toward full technical integration.
The report also noted that Nebius had previously formed an alliance with Nvidia and invested $2 billion to deploy a full-stack artificial intelligence infrastructure.
The specific financial terms of the acquisition have not been disclosed.


