•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
On May 1, 2026, Nebius (NBIS) said it intends to purchase Eigen AI in a transaction valued at approximately $643 million. The acquisition will be funded through a mix of cash and Nebius Class A shares, with the share component calculated using the company’s 30-day volume-weighted average share price at the time of signing. Following the announcement, NBIS shares rose 8.51% to $150.00.
Nebius said the purchase is designed to bring Eigen AI’s inference optimization and model optimization technology into its Token Factory platform. The goal is to improve model performance and unit economics at scale.
The transaction is expected to close in the coming weeks. Eigen AI focuses on inference optimization and enhancing model performance, enabling AI development teams to deploy open-source models more efficiently and cost-effectively in live environments without needing to build custom optimization infrastructure.
Nebius plans to integrate Eigen AI’s technology into Token Factory, its managed inference offering. Token Factory provides autoscaling API endpoints and fine-tuning capabilities for major open-source models, including Llama, DeepSeek, Qwen, Gemma, and additional architectures.
Nebius said the two companies had already been working together prior to the acquisition announcement. Before the deal was announced, they collaborated on optimized model deployments that achieved leading positions on Artificial Analysis, an AI performance evaluation platform.
Eigen AI was formed from MIT’s HAN Lab research group. The company’s co-founders, Ryan Hanrui Wang and Wei-Chen Wang, developed two methodologies described as influential in production AI infrastructure.
The third co-founder, Di Jin, earned his doctorate from MIT CSAIL and is described as having played a direct role in developing Meta’s Llama 3 and Llama 4 post-training processes. The article also notes that his work includes co-authoring CGPO, a reinforcement learning from human feedback methodology.
After the transaction closes, Eigen AI’s team will establish operations in the San Francisco Bay Area, which Nebius said will be its first American engineering and research center.
Nebius said inference has become the fastest-growing segment in the AI compute landscape, with projections indicating it will represent about two-thirds of overall AI computational requirements throughout 2026.
The article highlights that efficient inference deployment involves multiple technical challenges, including model representation, GPU kernel optimization, and dynamic workload management—capabilities that many organizations do not have internally.
It also notes that open-source models are often released without optimization, and that newer architectures such as Mixture-of-Experts and compressed sparse attention add further constraints related to memory use and computational efficiency, requiring specialized expertise.
Eigen AI’s optimization methodology is described as covering post-training refinement, fine-tuning, and production-grade inference across leading open-source model families. The article states that Eigen AI’s kernel-level and model-specific techniques are designed to maximize performance on existing hardware infrastructure without requiring additional development resources.
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…
