Nvidia is loosening the boundaries around one of its most important chip technologies. With NVLink Fusion, announced at Computex 2025 in Taiwan, the company is allowing non-Nvidia CPUs and GPUs to connect with Nvidia hardware through the NVLink interface in AI data centers.
The change matters because NVLink had previously been reserved exclusively for Nvidia chips. By opening that connection layer, Nvidia is not simply adding another product option. It is positioning NVLink as a shared foundation for AI systems that may mix Nvidia processors with outside CPUs, GPUs, and application-specific chips.
What NVLink Fusion Changes
NVLink Fusion is designed to let companies build semi-custom AI infrastructure around Nvidia’s high-speed NVLink platform. Instead of requiring an AI data center design to stay entirely inside Nvidia’s chip ecosystem, the initiative allows selected partners and customers to connect other processors to Nvidia hardware.
At the event, Nvidia CEO Jensen Huang described the goal directly:
"NV link fusion is so that you can build semi-custom AI infrastructure, not just semi-custom chips,"
That distinction is important. A semi-custom chip is only one component. Semi-custom AI infrastructure points to a broader system-level approach, where enterprises can combine Nvidia processors with CPUs or ASICs they choose while still using the NVLink platform and its wider ecosystem.
In practical terms, the announcement gives companies more flexibility in how they design AI data center systems. They can keep Nvidia’s interconnect technology at the center while bringing in processors from other suppliers where those chips fit their plans.
The First Partners And Participants
Nvidia named several early partners for NVLink Fusion: MediaTek, Marvell, Alchip, Astera Labs, Synopsys, and Cadence. The company also said Fujitsu and Qualcomm will be able to connect their own processors to Nvidia GPUs in data centers.
That list shows the initiative is not limited to a single kind of company. It includes chip companies and firms connected to the broader semiconductor design and infrastructure ecosystem. The shared point is access to Nvidia’s interconnect layer for AI data center hardware.
At the same time, some major competitors are not currently part of the NVLink Fusion ecosystem. Broadcom, AMD, and Intel are not yet included, according to the source article.
That absence keeps the scope of the announcement clear. Nvidia is opening up NVLink, but it is doing so through a defined ecosystem rather than turning every major rival into an immediate participant.
Why Nvidia Benefits From Opening Up
On the surface, opening a formerly closed technology can look like Nvidia is giving up control. The strategic logic runs the other way. If NVLink Fusion is adopted widely, Nvidia’s technology could become a central backbone for future AI infrastructure, including systems that are not built entirely with Nvidia chips.
That would let Nvidia extend its influence beyond the sale of its own processors. If companies rely on NVLink to connect mixed AI hardware, Nvidia remains embedded in the design of the data center even when other chips are part of the system.
The source article frames this around future AI "factories." In that context, Nvidia’s opportunity is to make its platform central to how those facilities are built. The chips inside a system may vary, but the connection layer can still carry Nvidia’s architecture and ecosystem into the design.
This is also why NVLink Fusion is more than a compatibility announcement. It gives enterprises a path to combine different processors while keeping Nvidia’s high-speed interface as the organizing technology. For buyers and builders of AI infrastructure, that could make procurement and system design less tied to a single chip lineup while still anchored in Nvidia’s platform.
DGX Cloud Lepton And Taiwan Expansion
NVLink Fusion was not the only announcement connected to Nvidia’s Computex 2025 presence. The company also introduced NVIDIA DGX Cloud Lepton, a cloud platform intended to give developers access to tens of thousands of GPUs across a global network of cloud providers.
Nvidia describes Lepton as a marketplace for reliable, high-performance GPU resources across the Nvidia compute ecosystem. According to Nvidia, the platform addresses a key bottleneck in AI development.
The logic is straightforward: AI development depends heavily on access to GPU resources. By presenting Lepton as a marketplace across cloud providers, Nvidia is trying to make that access easier to find within its own compute ecosystem.
Huang also announced a new office in Taiwan. Nvidia plans to work with Foxconn on a local AI supercomputer project, with support for other companies such as TSMC.
Taken together, the announcements show Nvidia widening its role in AI infrastructure from several directions at once. NVLink Fusion targets the physical design of AI data centers. DGX Cloud Lepton focuses on developer access to GPU capacity. The Taiwan office and supercomputer project add a local expansion effort tied to major technology partners.
The Bigger Picture
The clearest takeaway is that Nvidia is opening part of its ecosystem without stepping away from the center of it. NVLink Fusion lets other CPUs, GPUs, and ASICs connect into Nvidia-based AI data center systems, but it also strengthens the importance of Nvidia’s interconnect platform.
For enterprises, the appeal is flexibility. For Nvidia, the advantage is influence. If future AI infrastructure uses a mix of processors but depends on NVLink to connect them, Nvidia remains a core part of the architecture.
That is the strategic bet behind NVLink Fusion: AI data centers may become more heterogeneous, but Nvidia wants its platform to remain the connective tissue.