Nebius Pays $643M for Eigen AI

Nebius Group has agreed to acquire Eigen AI for about $643 million, a move the Amsterdam-based neocloud says will accelerate its push into inference software — the model‑serving and optimization layer that determines per‑token costs for AI services.

The deal was reported in early May 2026 and the headline figure — roughly $643 million in cash and stock — was confirmed by multiple outlets that covered the acquisition announcement.

The timing comes on the heels of a strong first quarter for Nebius. The company reported nearly eightfold revenue growth in Q1 2026, a jump analysts tied to rising demand for AI infrastructure and cloud services, and the stock moved higher on the news.

Eigen AI is a small, specialized startup focused on inference and model optimization. Industry coverage describes Eigen as an engineer‑dense team that squeezes extra tokens out of existing compute by improving how models are served and how kernels use silicon.

That software layer is increasingly valuable to neocloud providers because inference — the repeated serving of models to users — is where much of the ongoing revenue and cost pressure now sits. Providers that control inference software can raise effective capacity without buying more GPUs.

Nebius has framed the acquisition as plug‑and‑play for its Nebius Token Factory, the company’s inference platform that routes and bills model usage. Nebius says Eigen’s IP and engineers will be folded into Token Factory to improve token efficiency across the cloud offering.

Market analysts and industry commentators call the transaction an example of “inference economics” — the shift in competition from raw training capacity to optimizing the repeatable, per‑token work that drives margins for AI clouds. The Eigen deal is being positioned as a strategic purchase of per‑token margin.

Bloomberg and other reporters noted the acquisition buys Nebius not just code but talent and system‑level optimization techniques that can make the same GPUs run more workloads for the same cost. That capability is central to claims of improving gross margins on inference.

Nebius also disclosed in recent company posts and filings that the Eigen integration will be accompanied by other hires and licenses aimed at inference orchestration, including a separate agreement to bring in engineering teams and IP from other inference players. The company says the combined moves strengthen its full‑stack inference offering.

Integration is likely to take quarters rather than weeks. Independent reporting and market commentary estimate that embedding deep optimization software into a large cloud stack typically takes two to three quarters before customer‑facing gains appear. Investors will watch upcoming earnings and margin commentary for signs of progress.

For Nebius customers the promise is lower real‑world cost per token and potentially faster response times, with Nebius pitching this as a way to reduce customers’ AI bills without sacrificing model fidelity. For rivals, the deal raises the bar for software differentiation versus simply adding capacity.

What to watch next: whether Nebius can translate Eigen’s optimizations into measurable token savings at scale, how quickly those savings show up in margins, and whether other neoclouds respond by buying similar software teams or doubling down on capacity. The near‑term test will be the company’s next quarterly update and published performance metrics from Token Factory.