Nebius has announced its acquisition of Eigen AI, a startup specializing in AI model optimization, for approximately $643 million.
This move is a decisive step to win in the increasingly important field of AI inference. As the AI industry matures, the focus is shifting from simply training massive models to running them efficiently in real-world applications. This is the 'inference' stage, where success is measured not just by speed, but by cost-effectiveness—specifically, metrics like tokens per second (TPS) and the cost per million tokens. To gain a true competitive edge here, a company needs both massive infrastructure and a highly optimized software stack. This is precisely why Nebius acquired Eigen AI: to vertically integrate its cutting-edge optimization technology directly into its 'Token Factory' platform.
Looking back, this acquisition was a logical progression of events. First, Nebius demonstrated strong financial health and market demand, with its Q4 2025 earnings showing a sold-out capacity and $3.7 billion in cash. This provided the financial muscle for strategic moves. Second, NVIDIA's landmark $2 billion investment in March 2026 was a massive vote of confidence, solidifying Nebius's strategy of combining large-scale compute power with sophisticated software. This set the stage for acquiring a key software piece like Eigen AI.
Furthermore, the technical validation was already in place. A partnership between Nebius and Eigen AI, announced in March, had already demonstrated superior performance on public benchmarks like Artificial Analysis. This successful trial run provided concrete evidence that a full integration would yield significant benefits, making the acquisition a well-calculated decision rather than a speculative bet. The regulatory environment also became more favorable, with streamlined HSR filing processes allowing Nebius to guide for a quick closing within weeks.
Ultimately, this deal is about more than just technology; it's a strategic acquisition of world-class talent and a gateway to the American market. Eigen AI's team, with roots in MIT's Han Lab, is a leader in model optimization. By bringing them in-house, Nebius not only enhances its product but also establishes a crucial R&D hub in Silicon Valley, which is vital for attracting top engineering talent and securing large US enterprise customers. This is a classic vertical integration play aimed at controlling the entire inference value chain, from hardware to software, to deliver superior performance at a lower cost.
- Glossary
- Inference: The process of using a trained AI model to make predictions or generate outputs based on new, unseen data. It's the 'live' or 'production' phase of an AI model's lifecycle.
- Token Factory: Nebius's proprietary platform designed to serve large-scale AI models for inference tasks, focusing on performance, reliability, and cost-efficiency.
- TPS (Tokens Per Second): A key performance metric for large language models, measuring how many tokens (pieces of words) the model can process or generate in one second.