Nebius Agrees to Acquire Eigen AI, Strengthening Nebius Token Factory as a Frontier Inference Platform

Key Terms

inference technical

Inference is the process of drawing a conclusion from available evidence or data, like a detective piecing together clues to form a likely story. For investors it matters because these judgments turn raw reports, test results, or market signals into expectations about future performance, risk, or regulatory outcomes—so how someone infers from the same facts can change investment decisions and valuation.

autoscaling endpoints technical

Autoscaling endpoints are service access points—places where users or other systems connect to a cloud-based feature—that automatically adjust computing resources up or down based on demand. Think of a restaurant that summons more servers during a rush and scales back when it’s quiet; for investors, autoscaling endpoints signal that a company can handle traffic spikes without manual intervention, helping control costs, maintain performance and uptime, and support growth without large upfront infrastructure spending.

mixture-of-experts (moe) technical

A mixture-of-experts (MoE) is a machine‑learning design that uses many specialist submodels and routes each task to the few experts best suited to handle it, rather than running one large model for everything. For investors, MoE can deliver faster, cheaper and more scalable AI performance—like calling a team of specialists instead of hiring one person who knows a little about everything—so it can lower operating costs and improve product competitiveness.

compressed sparse attention (csa) technical

Compressed sparse attention is a technique used in large language and AI models that lets the system focus only on the most important pieces of information instead of comparing every item to every other item. Think of it as skimming a long book by highlighting the key sentences rather than re-reading every page; it reduces computing time and memory use. For investors, this matters because it can lower operating costs, speed up product performance, and enable models to scale to larger datasets or real‑time applications, affecting profitability and competitive advantage.

sparse attention (spatten) technical

Sparse attention (often called 'spatten') is a way for machine-learning models to focus only on a limited, relevant subset of information instead of comparing every piece to every other piece at once — like skimming only the important paragraphs in a long report rather than reading every line. For investors, sparse attention matters because it cuts computing time and costs and lets companies run bigger or faster AI services with less infrastructure, which can improve product performance, margins, and scalability while sometimes trading off a bit of accuracy.

activation-aware weight quantization (awq) technical

A method for shrinking a trained neural network so its numeric parameters take up less space and compute power, by converting those parameters into smaller, simpler numbers while paying attention to the range of values the network produces during use. For investors, this matters because it lowers the hardware, energy and latency costs of deploying AI models—like reducing a high-resolution photo to a smaller file without losing the faces—making AI products cheaper and faster to scale.

rlhf technical

Reinforcement learning from human feedback (RLHF) is a way of training artificial intelligence by having people review the system’s outputs and using those reviews to guide the AI toward preferred behavior, much like teaching a pet with praise and corrections. For investors, RLHF matters because it can make AI products more reliable, safer and more aligned with customer expectations, which affects user adoption, regulatory exposure and the company’s long‑term value.

gpu kernels technical

A GPU kernel is the small program or recipe that tells a graphics processor how to perform a specific calculation many times in parallel, like a single instruction a kitchen hands to dozens of cooks working on identical dishes. For investors, kernels matter because their efficiency and flexibility determine how fast and cheaply a company can run AI models, render graphics, or process large datasets, which affects product performance, operating costs, and competitive advantage.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

• Combines Eigen AI's industry-leading inference stack with Nebius's global capacity
• Jointly optimized endpoints achieved top rankings on Artificial Analysis across multiple models
• Eigen AI's founding team, including MIT HAN Lab researchers, will establish Nebius’s Bay Area engineering and research presence

AMSTERDAM--(BUSINESS WIRE)-- Nebius (NASDAQ: NBIS), the AI cloud company, today announced an agreement to acquire Eigen AI, a leading inference and model optimization company.

The acquisition will strengthen Nebius Token Factory as a frontier managed inference platform for production AI, combining a battle-tested optimization 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 optimization 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 optimized implementations of leading open source models that ranked among the fastest on Artificial Analysis.

The acquisition also accelerates Nebius’s expansion in the US. Eigen AI’s founding team – researchers who have developed optimization 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 of Nebius, said:

“We are operating in a capacity-scarcity world where AI builders need optimized inference and infrastructure scale. The integration of Eigen AI’s optimization 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.”

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 Quantization (AWQ) – 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 of 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 customization 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 optimization 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 unoptimized, 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 optimization approach that spans the entire model lifecycle – from post-training and fine-tuning to production inference optimization, 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 optimization 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 $643 million, subject to adjustments. The transaction is expected to close in the coming weeks, subject to certain customary conditions, including antitrust clearance.

About Nebius

Nebius, the AI cloud company, is building the full-stack platform for developers and companies to take charge of their AI future – from data and model training to production deployment. Founded on deep in-house technological expertise and operating at scale with a rapidly expanding global footprint, Nebius serves startups and enterprises building AI products, agents and services worldwide.

Nebius is listed on Nasdaq (NASDAQ: NBIS) and headquartered in Amsterdam. For more information please visit www.nebius.com.

Disclaimer

Forward-looking statements

This press release contains forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995, which involve risks and uncertainties. All statements contained in this press release other than statements of historical fact, including, without limitation, statements regarding our ability to complete the Eigen AI acquisition and our ability to integrate the Eigen AI team and to achieve the synergies and other benefits anticipated, are forward-looking statements. The words "anticipate," "believe," "continue," "estimate," "expect," "guide," "intend," "likely," "may," "will" and similar expressions and their negatives are intended to identify forward-looking statements.

These forward-looking statements are subject to risks, uncertainties and assumptions, some of which are beyond our control. Actual results may differ materially from the results predicted or implied by such statements, and our reported results should not be considered as an indication of future performance. The potential risks and uncertainties that could cause actual results to differ from the results predicted or implied by such statements include, among others: risks associated with acquisitions and the integration of businesses and teams; market, macroeconomic and geopolitical conditions; technological developments; our ability to secure and retain clients; our ability to secure additional capital to enable the growth of the business; as well as those risks and uncertainties related to our continuing businesses included under the captions "Risk Factors" and "Operating and Financial Review and Prospects" in our Annual Report on Form 20-F for the year ended December 31, 2024, filed with the Securities and Exchange Commission on April 30, 2025.

All information in this press release is as of the date hereof (unless stated otherwise). Except as required by law, we undertake no obligation to update or revise publicly any forward-looking statements, whether as a result of new information, future events or otherwise, after the date on which the statements are made or to reflect the occurrence of unanticipated events.

In addition, statements that "we believe" and similar statements reflect our beliefs and opinions on the relevant subject. These statements are based upon information available to us as of the date hereof and, while we believe such information forms a reasonable basis for such statements, such information may be limited or incomplete, and our statements should not be read to indicate that we have conducted an exhaustive inquiry into, or review of, all potentially available relevant information. These statements are inherently uncertain, and investors are cautioned not to unduly rely upon these statements.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260501259620/en/

Media relations: media@nebius.com
Investor relations: askIR@nebius.com

Source: Nebius