After tech giants and Gulf tycoons, how does NVIDIA maintain high growth?

NVIDIA's super growth story is facing a critical turning point.

As the multi-billion dollar deals from tech giants and Gulf countries gradually saturate, this AI chip giant is seeking new growth drivers.

Perhaps the model of relying on huge orders from a few super customers for growth is destined to be unsustainable. Seaport Research analyst Jay Goldberg pointed out:

"Will every country announce $10 billion or $50 billion data center projects like Saudi Arabia? Of course not. Their obvious trading opportunities are drying up."

In the face of growth pressure, NVIDIA is betting on NVLink Fusion and RTX Pro enterprise servers at this year's Computex conference, hoping to expand its control in the AI market through technological innovation. However, when recently asked by the media how NVIDIA plans to cope with the slowdown in AI spending, CEO Jensen Huang's response seemed somewhat evasive:

"AI infrastructure is being built everywhere—that's one of the reasons I'm traveling around the world. AI infrastructure will become part of society."

Big orders hard to replicate, can new strategy take over?

In the face of growth pressure, NVIDIA showcased a new growth path at the Computex conference that does not rely on sovereign infrastructure big orders: expanding its control in the AI market through technological innovation.

NVIDIA refers to its core technology as NVLink Fusion, which will help build AI infrastructure, not just a single, specific super chip. It is said to be able to integrate all companies' custom CPUs and needs, "directly into the ecosystem," thereby integrating NVIDIA GPUs.

Marvell Technology Vice President Nick Kucharewski explained:

"Companies do not need to build the entire equipment rack themselves but can innovate or differentiate on the custom chips themselves."

The logic of this strategy is clear: by attracting companies to build hardware utilizing the NVIDIA Fusion platform, it aims to drive demand for NVIDIA's underlying AI network and data center components. But the question is whether this indirect revenue model can match the explosive growth brought by previous direct big orders.

In addition to NVLink Fusion, NVIDIA has also begun to enter the enterprise market. This week, the company launched the RTX Pro enterprise server, supporting traditional x86, Hypervisor, Windows, and other IT workloads. It can also run graphical AI agents (AI Agents), and even run the Crysis game.

At the launch event, Jensen Huang's pitch was very enticing: these servers open up a multi-billion dollar market because customers can use them for "everything," such as graphics, virtual machines, and AI applications

However, Goldberg from Seaport Research poured cold water on it: The enterprise market is large, but it is difficult to penetrate. Compared to sovereign data centers, enterprise transactions are often smaller in scale, and the cost and time to secure transactions are higher:

"My feeling is that we are hitting a ceiling in expanding our customer base."