Nvidia is opening its AI server platform to chip rivals

Amid the gold rush to build data centers and AI models, Nvidia has sold the most sought-after tools, driving triple-digit annual sales growth and turning the company into a $3 trillion market value giant.

For the most part, Nvidia sells a “full-stack AI solution” that includes computer processors, or CPUs, networking equipment, and the AI accelerator, also known as the GPU. It’s inside an Nvidia-designed server rack. Together, these components helped to train and serve new AI models.

But now Nvidia is taking a significant step toward opening its platform, allowing customers to bring their own CPU or AI chip. It offers the option for future Nvidia racks with, say, Nvidia and Qualcomm chips inside.

Nvidia CEO Jensen Huang announced the move on Monday at the COMPUTEX trade show in Taiwan. The new system is called NVLink Fusion.

“A tectonic shift is under way: for the first time in decades, data centers must be fundamentally rearchitected—AI is being fused into every computing platform,” Huang said. “NVLink Fusion opens Nvidia’s AI platform and rich ecosystem for partners to build specialized AI infrastructures.”

Nvidia’s current top-of-the-line GB200 NVL72 AI server rack system includes Nvidia Grace CPUs and Nvidia Blackwell GPUs.

Initial NVLink Fusion AI chip partners include MediaTek, Marvell, and AIchip. Fujitsu and Qualcomm are the initial CPU partners.

“With the ability to connect our custom processors to Nvidia’s rack scale architecture, we’re advancing our vision of high-performance, energy-efficient computing to the data center,” Qualcomm Technologies CEO Cristiano Amon said in a press release.

The open system means large cloud providers, which were locked out of Nvidia’s ecosystem when using their own custom chips, can now mix their custom chips with Nvidia’s sought-after technology.

For Nvidia, it means the potential to fill more data centers with its server platform and to drive more revenue.

There’s one catch: Broadcom, which makes many of the custom AI chips for large tech companies, wasn’t on Nvidia’s initial partner list, though the company said more partners could be added in the future.