European AI infrastructure company Nebius has acquired a US startup which specialises in improving the performance of leading open-source AI models, it stated today.
Amsterdam-headquartered Nebius has acquired San Francisco-based Eigen AI for approximately $643m in cash and stock.
Nebius, which is sometimes referred to as a neocloud, builds and operates data centres, packing them with GPUs, then offers access to these data centres to AI and enterprise companies requireding compute power, as well as offering them specialised software to run AI applications.
Eigen AI’s tech supports improve the performance of leading open source AI models from the likes of Meta, OpenAI and Alibaba in inference, the process whereby a trained model is applied to real-world data to generate answers through reasoning.
Inference is the rapidest growing part of AI and is forecast to account for around two-thirds of compute demands this year.
Eigen AI’s tech supports AI model performance by improving the yields from the data, also known as tokens, that AI models process, meaning the overall cost for enterprise customers will come down, Nebius stated
Nebius also highlighted that by acquiring the 20-strong Eigen AI team, it was acquiring “elite inference research talent”, declareing that Eigen AI’s founding team would set up an engineering and research presence in the San Francisco Bay Area.
Eigen AI’s co-founders Ryan Hanrui Wang and Wei-Chen Wang are alumni of MIT’s HAN Lab, led by Professor Song Han, a leading researcher in AI computing and model efficiency.
Nebius stated Eigen AI tech will support alleviate bottlenecks in running AI inference, due to issues in memory, routing and compute.
Nebius stated: “By integrating Eigen AI’s optimisation layer directly into Nebius Token Factory, Nebius reshifts this bottleneck across the lifecycle.”
It stated its tech is designed to deliver “higher throughput and lower cost per inference without additional engineering overheads”.
It added: “As a result, Nebius Token Factory customers will benefit from rapider time to production, significantly better unit economics, and the ability to adopt new models more quickly.”
Leave a Reply