• NVIDIA's Chief Financial Officer indicates the vast majority of new AI chips are powering brand-new data center infrastructure, not replacing older systems.
  • The remarks, made at a UBS conference, underscore the explosive, capacity-driven nature of the current AI investment cycle.
  • This dynamic suggests a larger and more sustained total addressable market for NVIDIA's systems as enterprises build dedicated 'AI factories.'

NVIDIA's financial chief offered a telling glimpse into the engine of the company's historic growth during a recent appearance, framing the AI boom as a massive infrastructure build-out rather than a simple upgrade cycle.

Speaking at a UBS conference, the CFO stated that most of the advanced AI chips the company is currently shipping are being deployed to add entirely new data center capacity, not to replace an installed base. This distinction is crucial for investors trying to gauge the longevity and scale of the current demand surge. It implies that enterprises are not merely swapping out old servers for new ones but are constructing dedicated computational facilities—what NVIDIA leadership often calls "AI factories"—from the ground up.

This perspective aligns with a series of recent, massive infrastructure announcements. The company's landmark strategic partnership with OpenAI, for instance, aims to deploy at least 10 gigawatts of NVIDIA systems for next-generation AI infrastructure, with the first phase targeted for the second half of 2026. Similarly, initiatives like "Stargate UK" with data center operator Nscale plan to deliver tens of thousands of NVIDIA GPUs across new UK sites. Microsoft's recently announced 315-acre AI data center campus in Wisconsin is also designed to host "hundreds of thousands" of the latest NVIDIA GB200/GB300 GPUs.

"What we are seeing is the creation of a new asset class," said one analyst who attended the UBS session, speaking on condition of anonymity. "The CFO's comments confirm that this is a greenfield expansion story. Customers are building new data centers specifically architected for AI workloads that simply didn't exist at this scale two years ago."

The build-out is so significant that NVIDIA is establishing American-made manufacturing capacity for the first time, commissioning over a million square feet of space in Arizona and Texas to build and test its Blackwell chips and AI supercomputers. This move to onshore critical production underscores the strategic nature of the infrastructure being deployed.

Efforts to reach NVIDIA for further comment on the CFO's remarks were not immediately successful. However, the company's recent product unveilings, like the GB300 node which delivers roughly 40 petaflops—effectively replacing entire prior-generation supercomputers in a single package—are engineered for this new-build paradigm. The industry-wide transition to 800 VDC power infrastructure to support megawatt-scale racks further highlights that this is a foundational shift in how data centers are designed.

For the market, the implication is clear: the demand environment may be more resilient and expansive than a typical tech refresh cycle. If the majority of shipments are for net-new capacity, it suggests a longer runway for growth as global enterprises and cloud providers continue to lay the physical groundwork for AI. The risk, of course, is that capital expenditure cycles can be volatile. But for now, the message from NVIDIA's top finance executive is that the industry is still in the intense construction phase.

Correction: An earlier version of this article misstated the power infrastructure transition; it is to 800 VDC, not 800W.