At the opening keynote of GTC 2026, Nvidia CEO Jensen Huang invoked the 20th anniversary of CUDA not as a celebration of a technical milestone, but as a strategic reframing of the company's position in the global technology landscape.
For Huang, the two decades since CUDA's launch represent a complete transformation. NVIDIA has evolved from a graphics processing unit (GPU) vendor into a powerhouse controlling a software platform, system design capabilities, and the architectural foundation of modern AI factories.
Analysts and industry observers who still view Nvidia primarily as a chip supplier, Huang suggested, are relying on an outdated narrative.
CUDA as the foundation, not the finale, Huang opened the keynote by reminding the audience that GTC remains, first and foremost, a technology conference. He outlined Nvidia's three current pillars: a software platform tied to CUDA-X, a systems platform, and a new AI factories platform.
The framing was deliberate. CUDA is no longer the sole protagonist in Nvidia's narrative; it is the bedrock of a sprawling structure that extends from developer tools into data centers, networking, power infrastructure, cooling systems, and full-scale factory deployment.
Huang returned to CUDA's anniversary because it remains the hardest layer of Nvidia's competitive moat to replicate. A massive global installed base now runs on CUDA, continuously attracting developers and generating new algorithms. This creates a self-reinforcing flywheel: An installed base draws developers, whose breakthroughs unlock new markets, which in turn strengthen the platform.
From Chips to Systems
Huang addressed a second layer of competitive advantage: Nvidia no longer sells individual chips, but complete systems.
This shift explains why Huang refrained from holding up a single chip for the audience when discussing the new Grace Blackwell and Vera Rubin architectures. While earlier discussions of the Hopper architecture centered on a high-performance processor, Vera Rubin was presented as an integrated platform combining CPUs, GPUs, NVLink interconnects, networking, and data center architecture.
The competitive threshold, Huang indicated, has officially moved from component-level performance to full system integration capabilities.
Rise of the AI Factory
Above the hardware sits what Huang described as a third layer: the operational capacity of the AI factory itself.
He argued that future chief executives will monitor AI factory metrics — token output, throughput, cost, and revenue — as closely as traditional business indicators. The underlying claim is that Nvidia is extending its competitive position from computing hardware into an operating system for computing factories.
The contest among enterprises and cloud operators will no longer be over who acquires the most GPUs, but over who can most efficiently coordinate hardware and software, optimize workload scheduling, and reduce the cost of producing AI services at scale.
Looking Ahead
Taken together, Huang's references to CUDA's anniversary were less about retrospection than establishing a trajectory. The underlying message at GTC 2026 was straightforward: CUDA's 20th anniversary marks a starting point, not a summit.
The larger commercial opportunity is only now taking shape in AI factories. For cloud operators and enterprise technology decision-makers, the most consequential subtext of the conference is clear: Future competition will be determined not only by who holds the most powerful chips, but by who commands the entire AI factory. （Related： TSMC Operations Unaffected by Fire at Hsinchu Science Park Substation ｜ Latest ）
