Nebius, the Nasdaq-listed AI cloud company based in Amsterdam, said on October 1, 2026, that it had acquired Inferize, a Tel Aviv startup whose software shortens the time it takes to launch and scale large AI models. Inferize's technology and team have joined Nebius Token Factory, the company's managed inference platform for production AI.
Nebius did not disclose the terms. Israeli business publication Calcalist, through its English-language site CTech, estimated the deal at $100-150 million.
Inferize's engineers will work across Token Factory, starting with integrating their own technology. Nebius said the team's role would extend beyond that first project.
What Inferize built
Inferize targets cold starts, the delay while a model's weights load before it can serve its first request. Nebius said this leaves assigned GPUs idle when new instances start, when demand spikes and when weights are updated mid-run, for example during reinforcement learning, and pushes platforms to keep spare capacity running to meet service-level targets.
Nebius calls that cost an idle GPU tax. It said Inferize's technology lets capacity follow actual usage more closely, which raises utilisation and improves the economics of each token served. Those performance benefits are the companies' own claims and have not been independently measured.
Inferize was founded in January 2026 and had a working prototype within three months. It operated in stealth and was about 10 months old at the time of the sale, CTech reported.
In CTech's account, the problem shows up both when a model is first loaded and when its weights change during workloads such as reinforcement learning. In either case the GPUs assigned to the job are paid for but not yet serving requests. Letting inference capacity rise and fall with demand is meant to raise GPU utilisation and lower the cost of each AI request.
“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. Our team will work across the stack with one objective: serving more customer demand from every GPU.”
The founders and the funding
The company was founded by chief executive Guy Bortnikov and Lior Gorbonos, who were both on the founding team of Granulate, an Israeli infrastructure-optimisation company that Intel bought in 2022, according to CTech.
Inferize raised a seed round led by TLV Partners with participation from angel investors; the amount was not disclosed. It had 17 employees at its Tel Aviv office, CTech reported. Shahar Tzafrir, managing partner at TLV Partners, told CTech that the firm already knew the team from its seed investment in Granulate.
Tzafrir described the team as exceptionally rare in its ability to solve hard technical problems. Shtan, for his part, said Inferize brought both technology and a team with deep expertise in GPU systems, and Nebius said it is adding systems-engineering talent to its inference team.
Why Nebius is buying
Nebius has been adding software and inference expertise on top of its GPU cloud. It said Inferize adds another layer to how it runs production inference, after Eigen AI brought model, kernel and system-level optimisation into Token Factory and Clarifai's core team and licensed technology added inference and compute orchestration.
CTech noted that Nebius also agreed in February to acquire Israeli startup Tavily for an initial $275 million, making Inferize at least its second Israeli acquisition this year.
Chief technology officer Danila Shtan said running inference well requires the whole system to respond when demand changes, including how quickly extra capacity becomes ready, and that Nebius would use Inferize to make Token Factory more responsive and get more work out of its infrastructure.
Nebius is led by Arkady Volozh. CTech reported that not every approach has turned into a deal: talks between Nebius and Israeli AI company AI21 ended without an agreement, and AI21 later cut more than 60% of its workforce.
What happens next
Bortnikov said the team's goal inside Nebius would be to serve more customer demand from every GPU. Nebius did not say when the cold-start technology would be generally available to Token Factory customers or whether Inferize would keep a separate presence in Tel Aviv.
For Nebius, the purchase is a bet that the next gains in inference cost will come from the software layer around the GPUs rather than from the hardware itself. Shtan's comments put the emphasis on how quickly extra capacity can be made ready, not on raw chip performance.
