The deal combines Eigen AI’s advanced inference stack with Nebius’s global AI infrastructure to enhance performance, scalability and production efficiency.
The AI cloud company Nebius has announced an agreement to acquire Eigen AI, a leading inference and model optimisation company.
The acquisition will strengthen Nebius Token Factory as a frontier managed inference platform for production AI, combining a battle-tested optimisation stack with Nebius’s global compute capacity and AI cloud platform, and will add elite inference research talent to the company’s established in-house AI R&D capabilities.
Following close, Eigen AI’s inference and post-training optimisation layers will be integrated directly into Nebius Token Factory, which provides enterprise-grade autoscaling endpoints and fine-tuning pipelines across all major open-source models. The two companies have already delivered jointly optimised implementations of leading open source models that ranked among the fastest on Artificial Analysis.
The acquisition also accelerates Amsterdam-based Nebius’s expansion in the US. Eigen AI’s founding team – researchers who have developed optimisation techniques and tools the industry runs on – will join Nebius to establish a Nebius engineering and research presence in the San Francisco Bay Area.
Roman Chernin, co-founder and Chief Business Officer, Nebius, said: “We are operating in a capacity-scarcity world where AI builders need optimised inference and infrastructure scale. The integration of Eigen AI’s optimisation capabilities and founding team will establish Nebius Token Factory at the frontier of inference, offering customers market-leading model performance and unit economics with massive compute capacity to back it at scale.”
New York based Eigen AI’s founding team brings deep expertise from research that shapes how the industry deploys inference today. Co-founders Ryan Hanrui Wang and Wei-Chen Wang are alumni of MIT’s HAN Lab, led by Professor Song Han, a pioneering researcher in AI computing and model efficiency.
Ryan’s pioneering Sparse Attention (SpAtten) work is the most-cited HPCA paper since 2020, while Wei-Chen received the MLSys 2024 Best Paper Award for Activation-aware Weight Quantisation (AWQ) quantisation – now the standard for 4-bit model serving in production deployments. Co-founder Di Jin, an MIT CSAIL PhD, brings deep expertise in post-training and large-scale model alignment, having contributed to Meta’s Llama 3 and Llama 4 post-training and co-authored the CGPO RLHF framework.
Ryan Hanrui Wang, co-founder and CEO, Eigen AI, said: “We’re proud to join Nebius and work alongside the Token Factory team to push the boundaries of inference performance. Nebius has built a world-class AI cloud with a deep engineering culture that perfectly aligns with our own. Together, we are removing the friction of AI model customisation and deployment so developers can run models reliably in production without managing the underlying infrastructure.”
Inference is now the fastest-growing segment of AI, forecast to account for about two-thirds of compute demand this year. Open-source model usage is rising alongside it. With more workloads moving into production, the system optimisation layer is becoming critical infrastructure.
Running inference efficiently in production is inherently complex and requires deep expertise across the entire execution stack, from how models are represented, to how GPU kernels execute them, to how workloads are scheduled in real time.
Open-source models typically ship unoptimised, and newer architectures such as Mixture-of-Experts (MoE), Compressed Sparse Attention (CSA), reasoning and long-context models introduce additional challenges around memory, routing and compute efficiency. Most teams do not have the capacity to solve these problems in-house.
Eigen AI addresses this challenge with a full-stack optimisation approach that spans the entire model lifecycle. From post-training and fine-tuning to production inference optimisation, across all major open-source models in production demand, including GPT-OSS, Gemma, Qwen, Llama, Nemotron, DeepSeek, GLM, Kimi and MiniMax.
By integrating Eigen AI’s optimisation layer directly into Nebius Token Factory, Nebius removes this bottleneck across the lifecycle. The system, model- and kernel-level techniques developed by the Eigen team are designed to extract materially better performance from hardware out of the box, delivering higher throughput and lower cost per inference without additional engineering overhead.
As a result, Nebius Token Factory customers will benefit from faster time to production, significantly better unit economics and the ability to adopt new models more quickly. Existing Eigen AI customers will gain access to Nebius’s global AI infrastructure and platform capabilities.
The deal consideration will be paid in a combination of cash and Nebius’s Class A shares with aggregate value as of signing, based on Nebius’s 30-day weighted average stock price, of approximately US$643 million, subject to adjustments. The transaction is expected to close in the coming weeks – subject to certain customary conditions, including antitrust clearance.
