Nebius announced today an agreement to acquire Eigen AI for approximately $643 million. The deal combines Eigen AI’s inference optimization with Nebius’s Token Factory platform. The founders of Eigen AI, who come from MIT’s HAN Lab, are establishing a new Nebius office in San Francisco.
Nebius, officially a Dutch company but originating from Russia’s Yandex, is paying for the acquisition with a combination of cash and stock. The deal is expected to close within the coming weeks.
Token Factory was launched earlier this year as Nebius’ production platform for open-source AI models, offering various options for scaling up and down during use. Eigen AI optimizes those same models for inference using techniques such as post-training quantization (a well-known compression technique for AI), KV-cache optimization, and custom CUDA kernels. Following the acquisition, those optimization layers will be directly integrated into Token Factory.
MIT expertise behind the technology
The founders of Eigen AI are alumni of MIT’s HAN Lab, which focuses on AI development. Co-founder Ryan Hanrui Wang is the author of the most well-known paper on Sparse Attention, a crucial technique that current LLMs use to function. Meanwhile, Wei-Chen Wang developed Activation-aware Weight Quantization (AWQ), a method for efficiently reducing the size of AI models while maintaining quality. The third founder, Di Jin, contributed to the post-training of Meta’s Llama 3 and Llama 4. Together, the two companies have already published optimized versions of popular models such as DeepSeek, Llama, and Qwen that rank at the top of Artificial Intelligence benchmarks, with output speeds of up to 911 tokens per second.
Nebius is rapidly scaling up
Nebius is thus further expanding its capabilities in the field of AI infrastructure. In March, Nvidia also invested $2 billion in the company to expand hyperscale AI cloud capacity to over 5 gigawatts by the end of 2030. Previously, Nebius had already signed a $27 billion contract with Meta for five years of AI cloud capacity.
According to Nebius, inference is the fastest-growing segment of AI and is expected to account for two-thirds of total compute demand this year. Existing Eigen AI customers will gain access to Nebius’s global infrastructure following the closing.