🧠 Food for thought

Implications, context, and why it matters.

The acquisition brings model-serving software to Nebius’s AI cloud push

• Nebius is building an AI cloud and plans to grow from 20,000 graphics processing units, or GPUs, to 240,000 GPUs by 2027 1.
• The company also signed a long-term AI infrastructure supply agreement with Meta, the parent company of Facebook and Instagram. The contract could reach about US$27 billion over five years, including US$12 billion of dedicated capacity from early 2027 2.
• Buying Eigen AI adds software that helps run AI models more efficiently after training. Nebius plans to add it to Token Factory, its managed inference service, and expects the deal to close in the coming weeks 3.
• Eigen works on the full process of running AI models and says its technology can make inference up to 10 times faster and cheaper 3.
• The purchase gives Token Factory a stronger managed inference service in a market with heavy competition and tight capacity 4.

Efficiency grows more important in the AI cloud race

• The contest in AI infrastructure now goes beyond raw computing power.
• As large GPU clusters spread, providers like Nebius may compete more on workload efficiency. Pricing will matter too, alongside scale.
• The deal fits a broader move toward vertical integration, where infrastructure companies buy specialized software so their computing capacity is less likely to become a commodity 3.
• That could squeeze standalone inference startups. It could also push large cloud providers to prove their own optimization tools improve performance.
• Companies that cut the cost of running AI models may gain an edge. Inference is set to account for about two-thirds of AI computing demand this year, Deloitte forecast.