Nebius Acquires Eigen AI for $643 Million as Inference Optimisation Gains AI Infrastructure Priority
May 1, 2026 – 4:29 pm
TL;DR:
Nebius, the Dutch neocloud that split from Yandex in 2024, has agreed to acquire Eigen AI for approximately $643 million in stock and cash. The deal involves a 20-person startup founded by MIT alumni. The acquisition is driven by the growing importance of inference optimisation—maximising tokens per Nvidia GPU—in AI infrastructure.
Nebius Group, a Dutch cloud computing company that split from Russian internet provider Yandex in 2024, has reached an agreement to acquire Eigen AI for roughly $643 million. This acquisition involves a small team of 20 individuals. While the number may seem modest compared to other tech deals, the value per person is significant at approximately $32 million.
The Explanation: Inference
Eigen AI’s technology optimises token generation on Nvidia GPUs, allowing for more tokens per chip. Roman Chernin, Nebius co-founder and chief business officer, likens it to an Olympic sport: "who can extract more tokens for the same price?"
The Economics:
While training large language models is expensive, running them (inference) is even costlier. Every percentage point of efficiency gained in inference translates directly into lower costs or higher margins for AI service providers. Eigen AI specialises in optimising open-source models from major players like OpenAI, Alibaba, Meta, and Nvidia, using techniques such as activation-aware weight quantisation (awarded the MLSys 2024 Best Paper Award).
This optimisation allows models to run faster and more efficiently on GPUs, changing the economic dynamics for cloud providers like Nebius who invest heavily in GPU infrastructure.
The Acquirer:
Nebius is known as a hub of EU tech innovation.