Nebius’ co-founder and Chief Business Officer said the integration will deliver market-leading model performance and unit economics backed by massive compute capacity at scale.

Nebius has agreed to acquire Eigen AI to strengthen its Token Factory as a frontier managed inference platform, aiming to deliver top-tier model performance and economics at scale.

Eigen AI’s founding team will establish a Nebius engineering and research presence in the San Francisco Bay Area, expanding the company’s U.S. footprint.

Eigen AI provides full-stack optimization—from post-training and fine-tuning to production inference—designed to boost performance on existing hardware without extra engineering effort.

The deal, valued at $615 million in stock and cash, is expected to close in the coming weeks subject to antitrust clearance and standard closing conditions.

Joint deployments have already delivered leading performance rankings on Artificial Analysis, demonstrating accelerated inference for major models.

MIT-aligned Eigen AI technology has demonstrated optimized versions of models such as DeepSeek, Llama, and Qwen, achieving benchmark-leading speeds.

Eigen AI was founded by MIT HAN Lab alumni, with co-founders Ryan Hanrui Wang, Wei-Chen Wang, and Di Jin, who have contributed to Sparse Attention, AWQ, and post-training for Meta’s Llama models.

The company brings techniques like Sparse Attention and Activation-aware Weight Quantization to boost model deployment efficiency.

Token Factory, Nebius’ production platform for open-source AI models, has been growing to offer scalable options for users.

Supported open-source models include GPT-OSS, Gemma, Qwen, Llama, DeepSeek, and others, with Eigen AI spanning from post-training to production inference.

Nebius clients will gain access to Eigen AI and Nebius’s global infrastructure, with emphasis on faster time-to-production and improved unit economics.