By Adam Clark

A private cloud-computing firm backed by Super Micro Computer has clinched artificial-intelligence infrastructure deals worth $7.8 billion. It's a sign of the continuing strong demand for Nvidia chips and the growth of AI outside the U.S.

Argentum AI told Barron's it has signed two major long-term AI infrastructure deals to deploy a total of 47,000 Nvidia GB300 chips at a 300 megawatt data center in Poland. The deals have not been announced previously. One of the deals is worth $4.1 billion over five years, with the other worth $3.7 billion, with computing capacity set to come online beginning this year.

Argentum is backed by server maker Super Micro Computer, which has invested $100 million in the company via a convertible note.

Super Micro stock has been on a roll recently -- rising 16% on Monday after it showcased new AI infrastructure designs for Nvidia chips -- although shares were falling 6% in premarket trading on Tuesday amid a widespread tech selloff.

The Argentum deals are notable for the continued demand for Nvidia graphics-processing units for AI models in Europe. Nvidia is targeting the European market for growth, announcing on Monday that there were a record 35 AI supercomputers powered by its chips in development across the continent.

"These [deals] are probably...one of the top three largest GPU deployments in Europe today," Argentum AI CEO Andrew Sobko told Barron's in an interview.

AI skeptics have questioned whether cloud computing companies such as Argentum -- and publicly listed peers including CoreWeave and Nebius -- are accurately reflecting the long-term economic value of their hardware. However, Sobko argued Nvidia chips can continue to generate revenue long after the original leasing contracts.

"We believe that after five years, the value of the GPUs will be at least 25% of the cost of the capex originally and will continue being used over many years after that," Sobko said. "We're seeing massive demand for older chips today."

Another question dogging Nvidia has been whether it will lose its dominance as companies shift from training AI models toward inference -- the process of running the models. However, Sobko said the Nvidia GB300 systems are optimized both for model development and high-performance inference workloads.

Write to Adam Clark at adam.clark@barrons.com

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

June 23, 2026 07:21 ET (11:21 GMT)

Copyright (c) 2026 Dow Jones & Company, Inc.