The latest

Nebius has acquired Tel Aviv-based Inferize just 10 months after the startup’s creation, bringing its GPU-efficiency technology and 17-person team into the Token Factory platform. The Nasdaq-listed AI cloud company announced the purchase on October 1 without disclosing financial terms. Globes valued the transaction at between $100 million and $130 million, citing sources familiar with the deal. Inferize had developed a working prototype within three months but had not commercially launched its products before the acquisition.

Details

• Cold-start challenge: Inferize developed technology to shorten the time required to launch and scale large AI models. When demand requires additional capacity, model weights must be loaded into GPU memory before a new instance can process requests. This cold-start delay can leave costly processors idle precisely as traffic increases.
• Capacity economics: AI infrastructure providers often maintain spare GPU capacity to protect performance and availability during traffic spikes, production model updates and reinforcement-learning workflows. Inferize aims to reduce that idle period, allowing infrastructure to respond faster and helping Nebius generate more useful computing work from the same hardware.
• Team integration: Inferize’s employees will join Token Factory, Nebius’s managed platform for running AI models in production. Co-founder and CEO Guy Bortnikov said the team would work across the technology stack with the objective of serving more customer demand from every GPU, reducing the cost of reserved capacity.
• Inference expansion: Nebius launched Token Factory in 2025 to deploy and scale open-source and custom models through integrated inference, post-training and infrastructure management. It later completed the roughly $643 million acquisition of Eigen AI, added Clarifai’s core engineering and research team, licensed its orchestration technology and acquired Tavily for agentic search.
• Israeli startup: Inferize was founded in January 2026 by Bortnikov and Lior Gorbonos, who previously worked on infrastructure optimization at Granulate. TLV Partners was its sole institutional investor. The transaction turns a pre-launch Israeli company into a technology component of Nebius’s vertically integrated AI infrastructure business.

Background

AI cloud economics increasingly depend on GPU utilization, inference latency and cost per token, alongside access to advanced processors. A powered-on GPU waiting for a model to load or for demand to arrive represents infrastructure spending that is not producing revenue. Nebius describes its strategy as optimizing inference across the full stack, from hardware to model and system-level software, as competition shifts from simply securing GPUs toward converting each processor into productive AI capacity.

What’s next

The next measurable indicator is Nebius’s progress toward deploying more than 5 gigawatts of Nvidia systems by the end of 2030 under their March partnership. Nvidia also announced a $2 billion investment, making efficient GPU use increasingly important as Nebius expands capacity.