Nebius Acquires Eigen AI for $643M to Boost AI Inference Capabilities

European AI infrastructure provider Nebius has announced the acquisition of San Francisco-based startup Eigen AI in a transaction valued at approximately $643 million in cash and stock. The deal aims to enhance Nebius’s ability to deliver high-performance AI compute and inference solutions to enterprise clients.

Nebius, headquartered in Amsterdam, operates advanced data centers equipped with large numbers of GPUs. These facilities provide essential computational resources for businesses seeking to deploy and operate artificial intelligence applications at scale.

Eigen AI specializes in optimizing the performance of open-source AI models, including those from leading developers such as Meta, OpenAI, and Alibaba. The company focuses on improving model inference—the process by which trained AI models are applied to real-world data to generate predictions or insights. Efficient inference is crucial, as it determines how quickly and cost-effectively AI systems can be deployed in practice.

Industry analysts expect inference workloads to comprise nearly two-thirds of AI compute demand this year, highlighting the acquisition’s strategic alignment with market needs. Nebius stated that Eigen AI's technology boosts the efficiency of token processing, the fundamental units of data that AI models consume. This advancement has the potential to reduce operational costs for enterprise customers while accelerating deployment.

The deal also includes the integration of Eigen AI's 20-member research and engineering team, recognized by Nebius as “elite inference research talent.” Co-founders Ryan Hanrui Wang and Wei-Chen Wang, who are both alumni of MIT’s HAN Lab under Professor Song Han, will establish a new engineering and research base in the San Francisco Bay Area. This move is expected to further internationalize Nebius’s research footprint and access to talent.

According to Nebius, the addition of Eigen AI's optimization technology will be incorporated directly into the Nebius Token Factory, the company’s product offering for token management and inference lifecycle. The integration promises to address bottlenecks in memory, routing, and compute resources, delivering higher throughput and lower per-inference costs without extra engineering effort for clients.

Consequently, Nebius’s customers can expect faster time to production, improved unit economics, and greater agility in adopting new models as they emerge.

This latest acquisition demonstrates Nebius’s growing role as a European provider in the race for AI infrastructure. Through expansion into the US talent pool and direct enhancements to its technical platform, the company aims to strengthen its competitive edge in a rapidly evolving global AI ecosystem.

tech.eu