Internet Data Centers (IDCs) have long served as the backbone of the digital economy, evolving from simple server rooms in the 1990s to sprawling hyperscale facilities that power cloud computing, e-commerce, and global connectivity today. Yet in just the past few years, a profound transformation has taken place. Traditional data centers—once optimized for storage and general-purpose computing—are being reimagined as “AI factories” that produce intelligence itself. At the forefront of this shift is NVIDIA CEO Jensen Huang, whose data-driven insights and bold predictions highlight how AI is driving the largest infrastructure buildout in human history.
Huang has repeatedly described modern data centers not as warehouses for files, but as advanced factories that convert electricity into “tokens”—the fundamental units of AI output. “Future data centers will no longer be warehouses for storing files, but rather ‘factories’ producing Tokens,” he emphasized in his GTC 2026 keynote. This redefinition marks a pivotal inflection point: the arrival of the “inference era,” where AI systems generate responses, agents, and insights at massive scale, 24/7. Traditional IDC architectures, built around CPUs and storage arrays, are giving way to GPU-accelerated clusters capable of handling exponential compute demands.
The numbers underscore Huang’s vision. At NVIDIA’s GTC 2026, he projected at least $1 trillion in AI hardware demand (covering Blackwell and next-generation Vera Rubin platforms) through 2027—doubling an earlier $500 billion forecast made just months prior. NVIDIA itself reported record fiscal 2025 revenue of $215.9 billion, with data center revenue exploding as customers race to build these AI factories. Huang attributes this surge to two simultaneous exponentials: exploding AI capability and insatiable demand. “Computing demand is growing exponentially,” he noted. “Our customers are racing to invest in AI compute—the factories powering the AI industrial revolution and their future growth.”
Power has emerged as the central constraint—and Huang’s most urgent warning. “Every data center of the future will be power-limited,” he has stated, underscoring that physics imposes hard limits. “A 1-gigawatt (1GW) factory will never become 2 gigawatts. That’s the law of physics.” Under fixed power budgets, competitiveness hinges on efficiency. Huang has elevated “Tokens per Watt” as the new key performance indicator for AI factories, arguing that the lowest cost per token will determine winners in the intelligence economy. Rack power densities are already pushing 100–300 kW+, demanding liquid cooling, advanced thermal management, and co-designed energy systems.
Huang frames the entire AI ecosystem as a “five-layer cake”: energy at the base, followed by chips and computing infrastructure, cloud data centers, AI models, and finally applications that deliver economic value. All layers must scale together. At the World Economic Forum in Davos in January 2026, he told BlackRock CEO Larry Fink that the world is witnessing “the largest infrastructure buildout in human history.” Hundreds of billions have already been invested, but “trillions of dollars of infrastructure still need to be built.” A McKinsey estimate cited in related discussions projects cumulative global data center investment reaching $6.7 trillion by 2030 to meet AI demand.
This buildout extends far beyond hyperscalers in the U.S. and Europe. Huang urges every nation to treat AI infrastructure as core national infrastructure, akin to electricity or the internet. “You should have AI as part of your infrastructure,” he said in Davos. “I really believe that every country should get involved to build AI infrastructure, build your own AI, take advantage of your fundamental natural resource, which is your language and culture.” Sovereign AI initiatives are proliferating as countries recognize that local models grounded in native languages and data will deliver the greatest economic and cultural benefit.
Beyond economics, Huang highlights job creation. The AI infrastructure boom—spanning chip factories, data center construction, and energy projects—will generate millions of high-paying roles. “We’re talking about six-figure salaries for people who are building chip factories or computer factories or AI factories,” he observed. AI factories can even act as flexible grid assets, ramping power consumption up or down to ease peak demand on electrical networks.
NVIDIA’s technological roadmap underpins this future. Platforms like Blackwell, Vera Rubin, and the MGX ecosystem are engineered for multi-gigawatt AI factories, with supply chains now capable of delivering thousands of systems weekly. Innovations in networking, software (such as NemoClaw for agentic AI), and energy-efficient architectures aim to maximize tokens per watt while minimizing waste. Huang has even explored futuristic concepts, such as space-based data centers, though he acknowledges the timeline: “It’ll take years. It’s OK. I got plenty of time.”
Looking ahead, Huang is unequivocal: the $700 billion already committed to AI infrastructure is merely the opening act. “We have only just begun this buildout,” he wrote. “Trillions of dollars of infrastructure still need to be built.” As inference and agentic AI accelerate, IDCs will evolve into intelligent production engines that not only store and process data but generate economic value at unprecedented scale.
In summary, the development of Internet Data Centers is no longer about capacity—it is about intelligence. Jensen Huang’s data and viewpoints paint a clear picture: AI factories powered by NVIDIA technology will define the next era of computing, constrained only by power and imagination. Nations, enterprises, and developers that embrace this shift today will lead tomorrow’s intelligence economy. The infrastructure race is on, and the factories of the future are already under construction.