NVIDIA’s chief financial officer, Colette Kress, is pushing back on the growing chatter that today’s AI boom is an “AI bubble,” arguing instead that the tech world is in the middle of a major platform shift that’s still gaining momentum.
Speaking at the UBS Global Technology and AI Conference, Kress framed the current surge in AI infrastructure spending as part of a longer-term transition away from CPU-dominant computing and toward GPU-accelerated computing. In her view, the demand isn’t being driven by hype alone—it’s being driven by the reality that traditional CPU-based approaches are no longer delivering the kind of performance improvements the industry needs for modern AI workloads. That’s why, she says, moving more of the compute stack to GPUs isn’t optional; it’s becoming necessary.
Competition is also heating up, especially from custom ASICs designed for specific AI tasks. But Kress suggested that focusing only on a single chip built for a narrow use case misses what many AI customers actually require. She emphasized that NVIDIA’s strategy covers the full AI lifecycle, from training massive models to running inference at scale. According to her, NVIDIA’s advantage isn’t about one standalone processor—it’s about an entire accelerated computing environment where multiple chips and technologies work together as a coordinated platform. She contrasted this with the more “single product lineup” nature of typical ASIC approaches, implying that a broader, tightly integrated platform can be a stronger foundation for real-world AI deployment.
Another pillar of NVIDIA’s argument is ecosystem strength. Kress pointed to CUDA as a key reason customers continue to rely on NVIDIA for AI development, highlighting how ongoing CUDA and library improvements can translate into significant performance gains. The message is clear: for many organizations building and deploying AI, performance isn’t just about raw hardware—it’s also about the software stack, tools, and libraries that make models run faster and more efficiently.
Kress also offered an update on NVIDIA’s next-generation Rubin architecture, one of the most closely watched upcoming platforms in the AI hardware market. She confirmed that Vera Rubin has been taped out, meaning the design stage has progressed to the point where chips exist and work is underway to prepare for launch in the second half of next year. Alongside the chips, the associated networking infrastructure for the Rubin lineup has also reached the tape-out milestone, reinforcing the sense that NVIDIA is moving through its roadmap on schedule.
With Rubin progressing toward production and NVIDIA continuing to defend its platform approach against ASIC competition, the company’s stance is that the AI buildout is not peaking—it’s evolving. And if Kress’s outlook proves accurate, the next phase of AI growth will be defined less by speculation and more by an industry-wide shift in how computing is built and scaled for the GPU era.
