NVIDIA founder and chief executive Jensen Huang declared at GTC Taiwan on Monday, 1 June, that artificial intelligence has crossed a threshold — moving from a technology that demonstrates capability into one that generates measurable economic returns, a shift he framed as a turning point for the entire global technology industry.
"Compute is revenue. Compute is profit," Huang told the audience at Taipei Popular Music Center. As AI systems produce tokens that can be packaged into sellable services, he argued, computing power is no longer an overhead cost but a direct engine of revenue. The world, he said, is now locked in a race to build AI factories — and the infrastructure surge driving that race will push Taiwan's supply chain into a new phase of growth.
From Data Centers to AI Factories: A Structural Shift in How the Industry Thinks
Huang drew a pointed distinction between the data centers of the past decade and the AI factories now under construction worldwide. Traditional data centers were built to store, process, and move information. AI factories, he said, are designed to do something fundamentally different: to continuously convert raw data into intelligence, reasoning, and action — and to price and sell that output as tokens.
Tokens, in his framing, have become a profitable unit of production in their own right. Because AI companies want to produce more of them, they will build more facilities optimized for that output. That demand, Huang argued, is the underlying reason appetite for AI servers, rack systems, thermal management, power supplies, and networking equipment keeps intensifying — and why Taiwan's manufacturers are running at full capacity.
He described the global buildout of AI factories as one of the largest infrastructure deployments in human history. Every facility requires chips, cabinets, power, cooling, networking, storage, security, and software management systems operating in lockstep. There is no tolerance for a weak link.
NVIDIA's DSX Platform: An Operating System Built for AI Infrastructure
To address the operational demands of AI factories, NVIDIA unveiled its DSX architecture at the event. Huang explained the product hierarchy plainly: RTX is NVIDIA's GPU platform, DGX its systems platform, and DSX the new layer aimed at AI infrastructure at scale.
DSX integrates with Omniverse to build digital twins of planned facilities, allowing operators to model power loads, cooling performance, network topology, and system integration before breaking ground. Once a factory goes live, DSX OS takes over — handling configuration, real-time monitoring, and fault remediation to keep the facility performing efficiently.
Huang pointed to a structural inefficiency that DSX is designed to fix: AI factories today routinely over-provision power capacity by as much as 40 percent. Through DSX Max LPS and dynamic power allocation, he said operators can deploy more GPUs within the same power budget, redirecting that wasted headroom toward revenue-generating compute.
Why Taiwan's Role Goes Well Beyond Chip Manufacturing
For Taiwan's industry, Huang's argument carries implications that extend far beyond wafer fabrication and advanced packaging. As AI factories become the defining infrastructure layer of the next technology era, the companies positioned to benefit span the full hardware stack: AI server assembly, cabinet integration, thermal management, power supply units, networking equipment, data center engineering, and end-to-end systems delivery.
Speaking directly to Taiwan's supply chain, Huang said factories and teams are running at full tilt because the world has recognized that AI now carries genuine profit-generating capability — making compute the scarcest resource in the global economy.
The competitive logic of the AI industry, he argued, has fundamentally shifted. The race is no longer about who holds the most powerful chip. It is about who can most rapidly and efficiently integrate chips, systems, and infrastructure into a functioning AI factory. That shift means Taiwan's future competitiveness will rest not on manufacturing scale alone, but on its capacity to deliver complete, end-to-end systems — from silicon and advanced packaging through rack assembly, cooling, and full system integration. （Related： Jensen Huang Unveils 'RTX Spark': The AI Agent PC That Reinvents Computing ｜ Latest ）
