Nebius Buys 17-Person Inferize; CTech Estimates $100 Million to $150 Million

Nebius Group announced on October 1 that it has acquired Inferize, a Tel Aviv inference-optimisation startup founded in January 2026, and disclosed no price. Calcalist's CTech put the deal at $100 million to $150 million for a company with 17 employees, which works out to between $5.9 million and $8.8 million a head.

The terms are an estimate from one outlet. Nebius's filing on Form 6-K says terms were not disclosed, and the documents read for this story do not say whether the transaction has closed or only been agreed.

What Nebius is buying

CTech reported that Inferize was founded by Guy Bortnikov, its chief executive, and Lior Gorbonos, both from the founding team at Granulate, and that a seed round led by TLV Partners funded it. The seed amount has not been published. The filing says the company had a working prototype within three months of its founding.

The problem it targets is the cold start. When a model's weights have to load before it can answer, the GPU sits idle. CTech quotes Bortnikov: "Keeping spare GPUs running is the price of being ready for demand. Removing that cost is what we built Inferize to do." The team joins Nebius Token Factory, the managed inference platform, and Nebius chief technology officer Danila Shtan is quoted in the filing saying the contribution would "extend well beyond this first integration."

The claim with no ship date

Beri's reading of the announcement found that the only performance figure comes from Inferize, not Nebius. A single vendor-published test shows DeepSeek V4 Pro restoring in 20 seconds against roughly 23 minutes cold, about 70 times faster. Nothing independent confirms it, the Nebius announcement carries no benchmark, and Inferize's own homepage is a waitlist.

The same piece notes that Nebius's billing documentation warns released capacity "may not be immediately available again." A fast restore does not help if there is no GPU to restore onto, which is why the documentation recommends higher minimum replicas when cold starts are unacceptable. The buyer has to keep paying for spare capacity either way.

Price list and acquisition list

Beri also reports that Nebius raised GPU rates on October 1, the day of the announcement. Its figure for the H200 is a rise from $4.50 to $5.40 an hour, which is exactly 20 percent. That is a single-source number and the rise has no link to the acquisition in anything published; the two events share only a date.

1. Before October 14.5 $ per hour
2. From October 15.4 $ per hour

Source: Beri, accessed 2026-10-02

Acquisition	Capability	Reported price
Eigen AI	Quantisation	About $643 million
Clarifai team and IP	Inference tooling	About $97.4 million
Inferize	Cold-start restore	$100-150 million, CTech estimate

Per Beri, Inferize is the third inference-stack purchase in five months. Each purchase buys one piece of the serving path. For other money moving through AI infrastructure, see Amazon's Nvidia chip leaseback and Snorkel AI's Series E.

The next observable event is Token Factory pricing. If restore times reach customers, a published change in minimum-replica guidance would be the first sign, and no date has been given for it.