Nebius buys Inferize for up to $150M to cut cold starts

Nebius, a company that rents out computing power for AI, has bought Inferize, an Israeli startup that cuts the time GPUs sit idle while AI models start up. The deal was announced on October 1, 2026, and is reported at $100 million to $150 million, Parameter reports. Inferize was founded in January, so it sold less than nine months after it started.

What Inferize does

Inferize targets cold starts. A cold start is the delay between launching an AI model and the moment it is ready to answer requests. GuruFocus describes the technology as limiting the periods when GPUs sit unused as workloads shift.

That idle time costs real money. A cloud provider has to keep GPUs ready for spikes in demand, and a GPU that is waiting for a model to load earns nothing. Co-founder Guy Bortnikov calls idle GPU capacity the price of being prepared for swings in demand, Parameter reports.

"Removing that cost is what we built Inferize to do," Bortnikov said, as quoted by Startup Fortune. He added that Nebius is where the technology can plug directly into operations.

Who built it

Inferize is a second act for its founders. Guy Bortnikov, the CEO, and Lior Gorbonos previously founded Granulate, which Intel bought in 2022, all three sources say. They started Inferize in January 2026. TLV Partners led its seed round, Parameter reports, though the amount was not disclosed.

The whole company is 17 people. All of them join Nebius.

Detail	Value
Buyer	Nebius (Nasdaq: NBIS)
Startup	Inferize, Israel
Announced	October 1, 2026
Reported price	$100 million to $150 million
Founded	January 2026
Team size	17
Seed lead investor	TLV Partners
Founders	Guy Bortnikov (CEO), Lior Gorbonos

GuruFocus says the price range is an estimate first reported by Calcalist, an Israeli business outlet.

Where the team goes inside Nebius

The Inferize team and technology move into Nebius Token Factory, the company's managed inference platform, Startup Fortune reports. Inference is the work of running a trained model to answer requests, as opposed to training it. A managed platform runs the models for customers, who pay per use instead of managing GPUs themselves.

Nebius is led by Arkady Volozh. It has been buying to build out this platform. Parameter says Inferize complements its earlier purchases of Eigen AI and Clarifai. GuruFocus and Startup Fortune put Eigen AI's valuation at $643 million. Parameter also cites Nebius's earlier deal for Tavily, at $275 million upfront and up to $400 million with performance payments.

Investors were not impressed on the day. Nebius stock fell 1.5% in Thursday morning trading after the announcement, GuruFocus reports.

What this means for developers

If you serve AI models in production, cold starts are both a cost and a speed problem. Every minute a GPU spends loading a model is a minute you pay for without serving anyone. This deal is a sign that inference providers now compete on that waste, not only on GPU prices.

Measure your own cold starts first. Time how long each model takes from launch to its first answer, and how often you scale up from zero. Those two numbers show whether faster loading would save you money or only a few seconds.

If you use Nebius Token Factory, watch for changes in how fast new capacity comes online and in how idle time is billed. Neither company has announced dates or prices tied to the Inferize technology, so do not plan around it yet.

If you run your own GPUs, the trade-off stays the same. Keeping models loaded and warm costs money all the time. Scaling down to save money means users wait during the next cold start. Tools that shorten loading change where that balance sits.

For founders, the timeline is the story. A focused team of 17, with a track record from one earlier sale, solved one costly problem and sold within nine months.