Nvidia (NVDA, Financials) is loosening its grip. At Computex 2025, the chipmaker launched NVLink Fusion, a new effort to let outside CPUs and ASICs work with its GPUs—part of a broader bid to remain indispensable in AI infrastructure, even when competitors build the rest.
NVLink used to be Nvidia-only. Not anymore. Fusion cracks open the ecosystem, making room for third-party processors to plug into Nvidia's platform. Early partners include MediaTek, Marvell Technology (MRVL, Financials), Synopsys (SNPS, Financials), Cadence Design Systems (CDNS, Financials), and Alchip. Customers like Qualcomm (QCOM, Financials) now have more room to mix and match.
That flexibility matters. Nvidia already dominates general-purpose AI chips, but rivals like Amazon and Google are pouring money into custom silicon. Analysts say Fusion helps Nvidia stay in the room—even if it doesn't own all the furniture.
There's a tradeoff. Fusion could reduce demand for Nvidia's own CPUs, but makes its GPUs harder to replace in hybrid systems.
Broadcom (AVGO, Financials), Advanced Micro Devices (AMD, Financials), and Intel (INTC, Financials) are not part of this first wave.
Nvidia also introduced DGX Cloud Lepton, a kind of GPU marketplace to help AI developers find cloud horsepower. And it teased the GB300, its next Grace Blackwell system, coming in Q3 2025.
The company is also going bigger in Taiwan, opening a new office and working with Foxconn on a local AI supercomputer.
If Fusion catches on, Nvidia could tighten its hold on the AI data center—one link at a time.
