Nebius, a Nasdaq-listed AI cloud enterprise led by Arkady Volozh, has completed the acquisition of the Israeli startup Inferize. This strategic buy aims to enhance Nebius’s capabilities by integrating technology focused on minimizing underutilization of costly GPU resources when launching AI models or responding to fluctuating demand.
Although the specific financial details of the acquisition remain undisclosed, the transaction is estimated to fall within a range of $100 to $150 million. Inferize, which was established in early 2026 by founders Guy Bortnikov and Lior Gorbonos—both of whom were integral to the team behind the Israeli infrastructure optimization firm Granulate—had been operating in stealth mode and was only around ten months old at the time of acquisition. The startup successfully secured a Seed investment round led by TLV Partners, along with contributions from esteemed angel investors, and employed 17 individuals in its Tel Aviv offices.
Bortnikov, the CEO of Inferize, remarked, “Keeping excess GPUs operational incurs costs necessary for meeting demand. Our solution aims to eliminate this expense, and with Nebius, we can seamlessly integrate it into their platform.” He emphasized the focus of the Inferize team on optimizing GPU service delivery to meet customer demand efficiently.
The acquisition addresses a subtle yet crucial concern in AI infrastructure: the fate of expensive GPU resources when demand varies unexpectedly. Before an AI model can process requests, its weights must be loaded, a step that introduces latency known as “cold starts.” This scenario often results in GPUs being underused while new instances are initiated. Similar inefficient patterns can materialize when model weights are modified during processes like reinforcement learning.
For companies managing AI services at scale, achieving an optimal balance proves challenging. Maintaining additional computing capacity ensures readiness for increased demand but incurs costs for GPUs that may remain inactive for extended periods. On the other hand, operating closer to precise demand levels can jeopardize the ability to respond promptly to sudden spikes in workload.
Inferize's technology is engineered to align AI inference capacity more closely with actual demand, thereby enhancing GPU utilization and potentially lowering costs per AI request.
Within three months of its inception, Inferize had produced a working prototype. Bortnikov’s background includes collaboration with Gorbonos at Granulate, a firm that specialized in optimizing computing infrastructure and was acquired by Intel in 2022.
This acquisition further solidifies Nebius’s commitment to developing the software layer surrounding its growing AI computing infrastructure. The company has embraced a broader strategy that combines large-scale computing capacities with targeted software and technology acquisitions. Earlier in the year, Nebius agreed to acquire the Israeli startup Tavily for an initial sum of $275 million, with the possibility of reaching $400 million based on performance metrics. Tavily has developed a search layer for AI agents, which Nebius intends to integrate into its cloud services.
Additionally, Nebius previously explored the possibility of acquiring AI21, another Israeli AI firm, but those negotiations were ultimately redefined into a commercial partnership. Following those discussions, AI21 implemented significant workforce reductions, trimming over 60% of its staff, while refocusing on AI agent optimization technology with its Maestro platform.
In conjunction with its acquisition strategy, Nebius is expanding its operational footprint in Israel. The company has secured agreements for data center capacity and has been chosen to spearhead the establishment of a national AI supercomputer, with further plans for infrastructure development in locations such as Modi’in, Masmiyya, and Beit Shemesh.
This expansion signifies a broader transformation in the AI infrastructure sector, moving from the training of models to their deployment in customer services, with cost-efficiency and speed of inference becoming paramount. Nebius asserted that Inferize will contribute significantly to enhancing Token Factory’s capability to respond adeptly to changes in customer demand.
Inferize adds to a portfolio of technologies already being integrated into the Nebius platform. Among these, Eigen AI provides optimization across model, kernel, and system aspects, while the acquisition of Clarifai's core team and licensed technology has contributed to advancements in system-level inference and compute orchestration.
