NVIDIA founder and CEO Jensen Huang used his opening keynote at the GTC 2026 conference on March 16 not to spotlight a single new product, but to rewrite the artificial intelligence industry's investment narrative.
More than a product launch, Huang's address served as a declaration that the competitive axis of AI is shifting decisively. The era of model training is giving way to inference deployment and the rapid expansion of what Huang called "AI factories," cementing AI as the foundation of next-generation global infrastructure.
The 'inference inflection point'
The remark that drew the most attention from the industry was Huang's blunt assertion that an inference inflection point has arrived.
In Huang's framing, every instance of an AI model thinking, reasoning, and generating a response constitutes an inference operation. Over the past two years, he noted, global demand for AI computing has surged 1 million-fold. This figure reflects the growing scale of AI models, the acceleration of generative AI usage, deepening reasoning capabilities, and widespread commercial deployment.
（Related：
When a Minister Puts Herself Above the Legislature
｜
Latest
）
For NVIDIA, this surge represents a strategic opportunity to transition from the dominant player of the training era to the central infrastructure provider for the inference era.
A Trillion Dollar Market
Speaking from the stage, Huang significantly extended his market outlook, estimating that the addressable opportunity in AI computing infrastructure could reach at least $1 trillion by 2027.
This projection is notably higher than the $500 billion benchmark commonly cited by industry analysts. The statement signals Huang's view that the build-out phase for AI infrastructure is not decelerating, but rather entering a new, massive cycle of expansion.
This financial projection underpinned a recurring theme in the keynote: Companies must stop evaluating AI infrastructure by the scale of traditional data center construction. Instead, they should measure facilities by their ability to continuously produce tokens, sustain inference workloads, and convert output into revenue. Facilities are no longer simple repositories for data and computing tasks; they are full-scale "AI factories."
The real competition is no longer over who holds the most graphics processing units (GPUs). It is over who can achieve higher efficiency and lower cost per inference output within the constraints of available power, rack space, and time.
Large-scale monetization
Notably, Huang did not frame the narrative simply around Nvidia selling more chips. Instead, he repositioned the company as a comprehensive infrastructure provider spanning platforms, systems, and AI factories.
From an opening retrospective marking the 20th anniversary of CUDA to discussions of open models, agentic systems, and physical AI, the address served a single overarching argument: The next wave of opportunity lies in large-scale deployment, operation, and monetization.
For investors, cloud providers, enterprise IT departments, and government policymakers, Huang's signal was clear. The next contest in AI will not be determined solely by who trains the most powerful model, but by who can most rapidly convert models into services, services into tokens, and tokens into revenue.
（Related：
When a Minister Puts Herself Above the Legislature
｜
Latest
）
You've read it. Now let's talk. Follow us on X. Editor: Chase Bodiford
