The Pulse
Nebius Buys Inferize to Reduce AI Model Cold Starts
Nebius says Inferize’s technology is designed to reduce the time and GPU capacity tied up when large AI models start or scale. The company announced on October 1, 2026, that Inferize’s technology and team had joined Token Factory, with tech
AI.info Team · Written with AI ·
Nebius says Inferize’s technology can reduce the time large AI models spend waiting to serve requests—and the GPU capacity left idle while they start. On October 1, 2026, the company announced it had acquired Inferize and that its technology and team had joined Nebius Token Factory, Nebius’s managed platform for production inference. Integration of the technology is starting, the announcement says. Nebius did not disclose the deal’s financial terms or quantify the savings it expects.
Cold starts leave GPUs waiting
A cold start is the time a model needs to load before serving a request, Nebius says. When demand spikes and new instances start, assigned GPUs can sit idle during the launch process. The company also points to weight updates during a run, including in reinforcement learning, as a situation that can leave GPUs waiting. Platforms may hold spare capacity to meet service-level targets, even when that capacity is not serving requests.
Nebius says Inferize’s technology cuts this idle time, allowing capacity to scale more closely with actual usage. The company says that can improve capacity utilization and token economics. Its announcement does not provide benchmarks, a measured reduction in startup time or a quantified change in GPU utilization, so the size of any operational savings is unspecified. It describes the intended effect but does not give a customer release date or details of how the integration will be rolled out.
Inferize joins Token Factory’s engineering team
Inferize’s technology and team have joined Token Factory, according to Nebius. The company says Inferize’s engineers will work across the platform, starting with integration of their technology. Inferize was founded in January 2026 and had a working prototype within three months, the announcement says. Nebius has not provided a schedule for completing the integration.
Nebius Chief Technology Officer Danila Shtan said the work involves how quickly new capacity can respond when demand changes. He described the acquisition as bringing both technology and GPU-systems expertise into Token Factory.
“Running inference well takes more than fast GPUs and optimized models. The whole system needs to respond when demand changes, including how quickly additional capacity is ready to serve customers.”
Shtan said the team’s contribution would extend beyond the first integration. The announcement does not specify which later projects or product changes that work may involve.
A growing stack, without disclosed deal terms
Nebius presents Inferize as another addition to Token Factory’s production inference capabilities. It says Eigen AI brought optimization at the model, kernel and system levels, while Clarifai’s core team and licensed technology added system-level inference and compute orchestration. Inferize’s technology is aimed at the time and capacity involved in bringing models online as workloads change.
Inferize co-founder and CEO Guy Bortnikov characterized spare capacity as a cost of preparing for demand. He said the team would work across the stack with the aim of serving more customer demand from each GPU. Nebius has not disclosed how much it paid, when customers will receive the integration or whether changes in GPU utilization will affect Token Factory pricing.
“Keeping spare GPUs running is the price of being ready for demand. Removing that cost is what we built Inferize to do, and Nebius is where it can go straight into the platform.”