Nebius Acquires Israeli AI Startup Inferize in Deal Valued at $150 Million
Israeli cloud infrastructure company Nebius has acquired AI startup Inferize in a deal estimated at up to $150 million, just eight months after the startup's founding.
The cold start problem is one of the most painful and well-known issues in running large language models in production. It refers to the time it takes for a model to load before it can process and serve a user's first request, creating a significant bottleneck. During this waiting time, powerful and expensive GPUs sit virtually idle. This phenomenon occurs mainly during sudden demand spikes that require new resource allocation or when model weights update on the fly. As a result, AI companies must maintain large reserves of computing power just to meet performance targets. This is precisely the issue addressed by the Israeli startup Inferize, which is being acquired shortly after its founding.
Nebius Acquires Inferize
Cloud company Nebius, traded on Nasdaq and specializing in AI infrastructure, announced the acquisition of Inferize. While financial terms were not officially disclosed, estimates suggest the deal could reach up to $150 million, including performance-based grants. Founded in January 2026 by Guy Bortnikov and Lior Gorbunos, former founders of Granulate which was acquired by Intel, Inferize developed optimization technology for AI inference and model execution.
"We knew what they were capable of. Reaching such an exit just eight months after founding is an exceptional achievement," said Shahar Tzafrir, Managing Partner at TLV Partners.
Integration into Nebius Token Factory
Inferize's technology drastically shortens the time required to launch and scale large AI models, making workloads more elastic and improving token economics. Following the acquisition, the team and technology will be integrated into the Nebius Token Factory platform. Inferize employs 17 people at its offices in Tel Aviv and previously raised a seed round led by TLV Partners and angel investors.
