AI cloud infrastructure service provider Nebius (NBIS.US) has announced the acquisition of Israeli artificial intelligence startup Inferize, further strengthening its AI inference infrastructure capabilities. Inferize specializes in reducing idle time for graphics processing units (GPUs) and accelerating the deployment speed of large AI models. According to estimates from Israeli tech media Calcalist, the transaction value ranges between $100 million and $150 million, though Nebius did not disclose specific financial terms. The core of this acquisition lies in improving GPU resource utilization efficiency, particularly amid rapidly fluctuating AI inference demand, helping Nebius more quickly allocate computing resources and reduce idle time for expensive computing equipment.

Targeting AI Inference Efficiency Bottlenecks to Reduce GPU Idle Time

Inferize's core technology primarily addresses the "cold start" problem in the AI model deployment process. Cold start refers to the process that AI models must complete before they begin processing user requests, including model loading and related resource initialization. When a new inference instance starts, the GPU may need to wait for model loading to complete before it can formally begin computation, resulting in brief periods of resource idleness. This problem is particularly prominent when AI inference demand suddenly increases. As user requests surge, cloud service providers need to rapidly launch more inference instances, but if model loading takes too long, the newly added GPU resources cannot be immediately put into use, affecting service response speed and overall computing utilization efficiency. Inferize's technology aims to shorten this waiting process, enabling newly added inference resources to be brought online more quickly and making computing expansion more closely aligned with actual demand changes. For Nebius, this means the company is expected to more flexibly meet customer needs without having to maintain large amounts of idle GPUs as backup resources over the long term, improving the usage efficiency of its existing infrastructure. Nebius Chief Technology Officer Danila Shtan stated that efficiently operating AI inference services requires not only faster GPUs and optimized models, but the entire system must also be able to respond to demand changes in a timely manner, including rapidly providing additional computing capacity when customer demand increases. He noted that Inferize not only brings technology to accelerate this process but also possesses an engineering team with extensive experience in GPU systems. Nebius plans to incorporate Inferize's engineers into its inference business team and integrate the related technology into its AI inference platform Token Factory to enhance the platform's ability to respond to customer needs and enable existing infrastructure to handle more actual computing tasks. Shtan also stated that the Inferize team's future contributions will extend beyond this initial technology integration.

Lowering Backup Computing Costs to Improve AI Infrastructure Operational Efficiency

Inferize co-founder and CEO Guy Bortnikov stated that to respond to customer demand at any time, cloud service providers typically need to maintain a certain scale of backup GPU resources, and these standby computing devices themselves represent additional costs. He pointed out that Inferize was founded precisely to reduce this portion of costs, and joining Nebius will enable the related technology to be directly applied to an operational AI cloud platform. In the AI inference business, the scheduling efficiency of computing resources directly relates to infrastructure operating costs. Since GPU equipment is expensive, if enterprises need to retain large amounts of idle computing power over the long term to cope with potential demand peaks, it could affect overall investment returns. By shortening model startup time and accelerating the deployment of computing resources, Nebius is expected to improve service elasticity while reducing dependence on backup GPU capacity. It is worth noting that Bortnikov previously co-founded Granulate, a computing infrastructure optimization software company. That company primarily developed technology to improve computing resource operational efficiency and was acquired by Intel (INTC.US) in 2022. This background also reflects that the Inferize team has relevant experience in the field of computing infrastructure optimization.

Successive Acquisitions of AI Technology Companies as Nebius Continues to Strengthen Its Inference Platform

This acquisition of Inferize is one of a series of recent acquisitions by Nebius focused on AI inference and model optimization. Previously, Nebius completed the acquisition of Eigen AI for a transaction value of $643 million. In addition, the company recently acquired AI technology company Clarifai but did not disclose the specific amount of that transaction. These acquisitions are all aimed at enhancing Nebius's AI inference and model optimization capabilities, further improving its AI infrastructure services. Unlike infrastructure investments that mainly focus on computing power needed for AI model training, AI inference places greater emphasis on operational efficiency after models are put into actual use, including request processing speed, computing resource scheduling, model deployment, and cost per unit of computing power. As Nebius continues to integrate related technologies, its business layout is also extending from providing GPU computing resources to technologies and services that improve the actual operational efficiency of AI models. This acquisition of Inferize will further supplement Nebius's technical capabilities in dynamic computing scheduling and rapid model deployment. However, the extent of cost savings and efficiency improvements that the related technology integration can bring still needs to be verified through subsequent operational performance.