Nebius buys Inferize in stealth GPU efficiency play

Nebius has acquired Israeli AI startup Inferize for $100–150 million, adding a team of 17 engineers and technology designed to minimize GPU downtime during model initialization and usage spikes. The deal, first reported by CalcalisTech and later confirmed by MoneyCheck, was completed in early October but kept under wraps until now. Neither company has disclosed the exact structure, but the price tag suggests Nebius sees real value in Inferize’s approach—one that could reshape how AI infrastructure providers compete beyond raw GPU supply.

Most providers focus on scaling up hardware or optimizing software stacks, but Inferize’s team—former executives from a performance-optimization startup previously acquired by a major tech company—built tools to reduce idle time during model loading and handle sudden demand fluctuations. That’s a niche, but in a market where GPU costs can make or break a startup’s economics, it’s a lucrative one. Nebius, a cloud provider spun out of a larger tech firm, has been building out its AI infrastructure play, and this acquisition suggests it’s betting on efficiency as a differentiator, not just capacity.

The timing is telling. When we covered CoreWeave’s shift toward AI inference last month, the focus was on capturing demand for AI workloads beyond training. That’s a different angle, one that could appeal to startups and enterprises struggling with the cost of running AI models, not just those scaling them. If Nebius can integrate Inferize’s tools into its platform, it might carve out a space between the hyperscalers and the pure-play GPU rentals, offering a more cost-effective alternative for customers who care about utilization, not just raw power.

The acquisition also highlights the growing importance of Israeli engineering talent in AI infrastructure. Upwind’s recent purchase of Aegis, another Israeli AI security startup, suggests that the country’s expertise in performance optimization and cybersecurity is becoming a key asset for companies looking to build proprietary AI stacks. Inferize’s team, with its background in a performance-optimization startup, fits neatly into that trend. What’s less clear is how Nebius plans to monetize the acquisition. Will it keep Inferize’s technology as a proprietary advantage, or will it productize it for broader use? The lack of public details suggests the former, at least for now.

What to watch next: whether Nebius starts pitching GPU efficiency as a core selling point, or if this remains a quiet internal play. If it works, expect competitors to scramble for similar talent—or to acquire the next Inferize before it’s too late.

Sources: moneycheck.com