Nebius, the Amsterdam-based AI cloud company listed on Nasdaq, agreed on 1 May 2026 to acquire Eigen AI, a young company that tunes open-source AI models so they run faster and more cheaply in production. Nebius said the deal had an aggregate value of about $643 million at signing.
The price is being paid in a mix of cash and Nebius Class A shares. The share portion was valued on Nebius's 30-day weighted average stock price, and the total is subject to adjustments. Nebius did not say how the amount was divided between cash and stock.
Eigen's optimisation software is being built into Nebius Token Factory, the company's managed service for running models, and Eigen's founders are setting up a Nebius engineering and research office in the San Francisco Bay Area. Nebius said on 16 June that the acquisition had closed on 10 June after the required regulatory approvals.
What Eigen AI built
Eigen works on the part of the AI stack that sits between a trained model and the GPUs that serve it. According to Nebius, its tools cover post-training, fine-tuning and production inference, and work with the main open-source model families, including GPT-OSS, Gemma, Qwen, Llama, Nemotron, DeepSeek, GLM, Kimi and MiniMax. Techzine reported that the methods include post-training quantisation, KV-cache optimisation and custom CUDA kernels.
The aim, as Implicator described it, is to get more tokens out of each Nvidia chip. The two companies had already worked together before the deal: in March 2026 they began offering Eigen-optimised open-weight models inside Token Factory, and those endpoints ranked among the fastest measured by the benchmarking firm Artificial Analysis. Implicator cited 911 output tokens per second for GPT-OSS-120B.
Eigen is small. Bloomberg reported that it had 20 employees, according to Implicator. Neither company disclosed when Eigen was founded, how much money it had raised or who its investors were. Nebius said Eigen's existing customers would get access to its global infrastructure after closing.
“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.”
The founders
Chief executive Ryan Hanrui Wang and co-founder Wei-Chen Wang both came out of the HAN Lab at the Massachusetts Institute of Technology, which is run by Professor Song Han and focuses on efficient AI computing. Nebius said Ryan Wang's SpAtten paper on sparse attention is the most cited paper at the HPCA computer architecture conference since 2020.
Wei-Chen Wang won the MLSys 2024 best paper award for Activation-aware Weight Quantization (AWQ), a method for compressing model weights to four bits that Nebius described as the standard for serving 4-bit models. The third co-founder, Di Jin, holds a PhD from MIT's Computer Science and Artificial Intelligence Laboratory and worked on post-training for Meta's Llama 3 and Llama 4 models. He also co-authored CGPO, a framework for reinforcement learning from human feedback.
Why Nebius is buying
Nebius rents GPU capacity and sells software on top of it. Token Factory, launched in November, offers autoscaling endpoints, fine-tuning pipelines and access controls for open and custom models. Nebius said inference is the fastest-growing part of AI and pointed to forecasts that it will make up about two-thirds of compute demand this year.
Its argument is that open-source models usually ship without optimisation and that newer designs, such as mixture-of-experts and long-context models, are harder to run efficiently. Most customers lack the specialists to fix that themselves. Roman Chernin, Nebius co-founder and chief business officer, said customers needed both optimised inference and infrastructure at scale, and that Eigen's team would put Token Factory at the front of the inference market.
The deal also gives Nebius its first engineering and research base in the Bay Area, which Verdict described as part of an expansion into the United States.
A second deal in three months
Eigen was Nebius's second acquisition agreement in three months. In February it agreed to buy Tavily, which provides real-time web search for AI agents. CTech reported a starting price of $275 million for Tavily, rising to $400 million if milestones are met, according to Implicator.
Nebius has been expanding quickly. Techzine noted that Nvidia invested $2 billion in the company in March and that Nebius has a five-year, $27 billion AI cloud capacity contract with Meta. Implicator reported that Chernin told Bloomberg the company wants products closer to direct customer use, and that Nebius competes with other specialised compute providers such as CoreWeave and Nscale.
