Netherlands-based AI cloud firm Nebius has agreed to accumulate Eigen AI, a agency specialising in inference and mannequin optimisation firm, in a money and inventory transaction valued at roughly $643m.
With this acquisition, Nebius plans to combine Eigen AI’s inference and post-training optimisation applied sciences immediately into Nebius Token Manufacturing facility, its managed AI manufacturing platform.
This integration goals to help enterprise-grade deployment, together with autoscaling endpoints and fine-tuning for a broad vary of open-source fashions.
Nebius, which is listed on Nasdaq, and Eigen AI have beforehand collaborated on implementations that acquired excessive efficiency rankings in Synthetic Evaluation assessments.
The transfer additionally marks Nebius’s growth into the US, with Eigen AI’s founding researchers becoming a member of the previous to ascertain an engineering and analysis hub within the San Francisco Bay Space.
Key members of Eigen AI’s group embody co-founder Ryan Hanrui Wang, identified for analysis on Sparse Consideration, and co-founder Wei-Chen Wang, who acquired recognition for work on Activation-aware Weight Quantisation. Co-founder Di Jin has contributed to Meta’s giant language fashions and reinforcement studying frameworks.
The founding group beforehand undertook analysis on the Massachusetts Institute of Expertise’s HAN Lab and CSAIL.
Wang mentioned: “We’re proud to affix Nebius and work alongside the Token Manufacturing facility group to push the boundaries of inference efficiency.
“Collectively, we’re eradicating the friction of AI mannequin customisation and deployment so builders can run fashions reliably in manufacturing with out managing the underlying infrastructure.”
Trade traits point out that inference at present represents the fastest-growing phase of AI, with forecasts estimating it’ll account for about two-thirds of compute demand this yr.
The rising adoption of open-source fashions in manufacturing environments has made the optimisation of inference a precedence, significantly as newer mannequin architectures introduce further calls for on reminiscence and compute assets.
Eigen AI’s know-how goals to deal with optimisation wants throughout the whole mannequin lifecycle, from post-training and fine-tuning to manufacturing deployment. It helps a variety of widespread open-source fashions, together with GPT-OSS, Qwen, Gemma, Nemotron, Llama, GLM, DeepSeek, Kimi, and MiniMax.
The mixing of Eigen AI’s stack into Nebius Token Manufacturing facility is meant to enhance {hardware} effectivity and throughput, whereas decreasing the operational overhead for purchasers.
Nebius co-founder and chief enterprise officer Roman Chernin mentioned: “The mixing of Eigen AI’s optimisation capabilities and founding group will set up Nebius Token Manufacturing facility on the frontier of inference, providing prospects market-leading mannequin efficiency and unit economics with large compute capability to again it at scale.”
