Microsoft-Backed Nebius Just Fired a Shot at Amazon and Google in the AI Cloud War

This article first appeared on GuruFocus.

Nebius Group NV (NASDAQ:NBIS) is stepping deeper into the AI infrastructure race with the launch of Token Factorya new platform built to give developers direct access to open source models and the computing muscle to run them. The product zeroes in on inference workloads, where trained AI models go live. Clients can now tap into models such as OpenAI's GPT-oss, Meta's Llama, and China's DeepSeek, choosing the mix that fits their needs while securing Nebius's high-performance compute clusters to power deployment. The move positions Nebius to capture a larger slice of the booming AI demand now spreading across industries.

Warning! GuruFocus has detected 3 Warning Signs with NBIS. Is NBIS fairly valued? Test your thesis with our free DCF calculator.

The timing could be strategic. Nebius, which spun off from Yandex last year, has been scaling fast across the US, Europe, and Israel, where it recently opened one of the country's first public clusters using Nvidia's newest chips. Microsoft's (NASDAQ:MSFT) multibillion-dollar dealworth up to $19.4 billionfor Nebius computing capacity underscores how even the largest players see value in partnering with emerging neocloud providers. For Nebius, Token Factory could be more than just a software add-on. Co-founder Roman Chernin said the company's ambition isn't merely to lift margins but to expand its reach by offering a broader portfolio to enterprise clients seeking flexibility.

Chernin suggested that developers are beginning to rethink their reliance on closed, proprietary AI ecosystems, which can restrict customization and raise costs. Token Factory, he said, was built for that shifta platform designed to scale, adapt, and let clients pivot between models as their needs evolve. Early users include Prosus NV and Higgsfield, while HuggingFace has already integrated the product into its inference marketplace. For investors, Nebius could be positioning itself as the open-source alternative in a cloud market still dominated by hyperscalers, where the next phase of AI growth might depend as much on accessibility as on raw compute power.

View Comments