The artificial intelligence race shows no sign of slowing, and every major tech company is working on chips and infrastructure that could prove transformative for the entire industry. Nvidia sits squarely in that ecosystem, having positioned itself from the very beginning as one of the brands most committed to playing a central role in the technology’s growth.

For months, attention has focused on Kyber NVL144, a new artificial intelligence infrastructure whose development appeared to be well on track. According to a report that surfaced on July 7, 2026, the company has run into trouble caused by a manufacturing defect — and the launch is now likely to slip to at least 2028.

What happened with Nvidia’s Kyber

The Kyber NVL144 is a system designed to integrate up to 144 GPUs (graphics processing units) based on the Rubin Ultra architecture into a single server rack. In practice, the goal is to turn the entire configuration into one unified platform for training and running inference on artificial intelligence models.

All of this would increase computing power and reduce latency between graphics processes. A report released in recent hours, however, flags serious problems that have slowed the project’s development. The main obstacle appears to lie in the production of the midplane PCB — the multilayer printed circuit board responsible for the interconnections between the system’s modules.

The high complexity of that single component has slowed large-scale production. The resulting delay would push the company to wait until at least 2028 before launching, and it could also affect the roadmap of the NVL576, another project that calls for interconnecting multiple racks through optical links.

Nvidia’s plans from here

For now, the only near-certainty is that Kyber NVL144 is in a holding pattern, and several more months will be needed before development work is complete. Other rumors suggest the company has abandoned alternative plans, including one that involved joining two existing racks together — a route considered impractical given the high costs and complexity of the solution.

There is good news for the brand as well. Despite the difficulties that have emerged, GPUs based on the Rubin architecture remain in production, and shipments to major partners are expected to begin as early as this fall. Those partners include Amazon Web Services (AWS), Microsoft Azure, and Google Cloud.

All that remains now is to wait for official statements on the postponement of Kyber NVL144. For the moment Nvidia has stayed silent, but the report is fairly confident on one point: the manufacturing defect forces a delay of at least a year and a half.

Editor’s note

This article was originally published in Italian on money.it by Pasquale Conte on July 07, 2026 as «NVIDIA ritarda il nuovo super sistema basato sull’intelligenza artificiale al 2028». It has been translated and adapted for an international audience by the Money.it International desk.