NVIDIA is opening up its AI server platform to bargain with competitors

In the gold rush that builds data centers and AI models, NVIDIA sold its most popular tools, driving triple-digit annual sales growth and turning the company into a $3 trillion market value giant.
In most cases, NVIDIA sells “full heap AI solutions” that include computer processors or CPUs, network devices, and AI accelerators (also known as GPUs). It is in a server rack designed by NVIDIA. Together, these components help train and provide new AI models.
But now, NVIDIA has taken an important step towards opening its platform, allowing customers to bring their own CPU or AI chips. It offers options for future Nvidia shelves, such as the in-house Nvidia and Qualcomm chips.
NVIDIA CEO Jensen Huang announced the move at the Computex trade show in Taiwan on Monday. The new system is called NVLINK fusion.
“The construction shift is underway: Data centers have to be fundamentally re-study for decades – AI is fused into every computing platform,” Huang said. “NVLink Fusion opens NVIDIA’s AI platform and rich ecosystem for partners to build professional AI infrastructure.”
NVIDIA’s current top-level GB200 NVL72 AI server rack system includes Nvidia Grace CPU and NVIDIA Blackwell GPU.
The initial NVLink Fusion AI chip partners include Mediatek, Marvell and Aichip. Fujitsu and Qualcomm were the original CPU partners.
“With a rack-scale architecture that can connect our custom processors to NVIDIA, we are pushing our vision of high-performance, energy-efficient computing to the data center,” Cristiano Amon, CEO of Qualcomm Technology, said in a press release.
Open systems are large cloud providers locked into the NVIDIA ecosystem when using their own custom chips, which can now mix their custom chips with Nvidia’s highly sought-after technology.
For Nvidia, this means it is possible to populate more data centers with its server platform and earn more revenue.
One takeaway: Broadcom is Broadcom that makes many custom AI chips for large tech companies, but is not on Nvidia’s initial partner list, although the company says it will add more partners in the future.