Nebius acquires Israeli inference startup Inferize in a reported $100-150M deal

Nebius (Nasdaq: NBIS), the AI cloud company, said on October 1 it had acquired Israeli startup Inferize, whose technology shortens the time to launch and scale large AI models. Financial terms were not disclosed: CTech estimates the acquisition at $100-150 million, while Globes reported a $100-130 million range.

Inferize was founded in January 2026 by Guy Bortnikov, its CEO, and Lior Gorbonos, both from the founding team of Granulate, an infrastructure-optimization company Intel bought in 2022. The startup, operating in stealth, employed 17 people in Tel Aviv and raised a seed round led by TLV Partners. Its technology and team join Nebius Token Factory, the managed inference platform for production AI.

The cold start problem

To answer a request, a model's weights must first load — a cold start that leaves GPUs idle while new instances come online. The same inefficiency appears when weights are updated mid-run, as during reinforcement learning. Platforms face a trade-off: hold spare capacity for demand spikes and pay for GPUs that often do nothing, or run close to demand and risk slow responses.

Inferize's technology is designed to let inference capacity follow real usage much more closely, raising GPU utilization and reducing the cost of serving each AI request — what Nebius calls the "idle GPU tax". A working prototype followed within three months.

What the companies say

Danila Shtan, Nebius's chief technology officer, said running inference well takes more than fast GPUs and optimized models: the whole system has to respond when demand changes, including how fast extra capacity is ready. Inferize brings technology that accelerates that process and a team with deep GPU expertise, he said; both go into Token Factory. "The team's contribution will extend well beyond this first integration," he said.

Guy Bortnikov, Inferize's co-founder and CEO, said 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," he said, adding that his team will work across the stack to serve more demand from every GPU.

Part of a wider build-out

The deal extends Nebius's strategy of wrapping its computing capacity in software. Eigen AI brought model- and system-level optimization; Clarifai's core team and licensed technology added inference and orchestration. In February, Nebius agreed to buy Tavily, a search layer for AI agents, for an initial $275 million. Talks to buy AI21 later ended without a deal and became a commercial partnership.

Nebius is also expanding in Israel, signing data center agreements and being selected to establish a national AI supercomputer, with planned facilities in Modi'in, Masmiyya and Beit Shemesh. The moves track a market shift: as companies move from training models to deploying them, inference cost and speed matter more.

Sources: CTech · Nebius Newsroom