Jensen Huang walked into GTC 2026 with 3,500 people in the room representing a combined $40 trillion in market cap — and someone in the audience told him NVIDIA had just posted what might be the single best earnings print in recorded human history. His response: “It must be only recorded humanity. I’m sure somebody had better returns.” That’s the energy of a man who isn’t surprised by his own numbers anymore. What followed was one of the clearest articulations yet of why Huang believes the demand for compute is not cyclical, not speculative, and not close to peaking. He laid out five interlocking theses that, taken together, describe a structural shift in how the global economy actually runs. This article unpacks each one.
Thesis 1: Compute Is Now a Revenue Line Item for Every Company
Huang’s first thesis is blunt: “Every single company will need compute for revenues.” He’s not talking about IT infrastructure in the old sense — servers running payroll software or hosting a company website. He’s describing a specific causal chain: compute generates intelligence, intelligence enables a digital workforce, and a digital workforce generates revenue. The sequence matters because it reframes what compute is for a business.
For most of the last twenty years, compute was a cost center. You spent money on servers or cloud credits to run your business. Now Huang is arguing it’s the opposite — that compute is the input to a production line that outputs economic value. An AI agent that handles customer onboarding, a model that qualifies sales leads, a system that dynamically prices inventory — all of those are compute directly converting to revenue. Not indirectly. Directly.
This is why he said what he said about NVIDIA’s stock, which was down 30 cents on the day of the interview: “You can’t hold the stock back. You can’t hold it back.” He wasn’t being glib. He was making a structural argument. If every company in every industry needs compute to generate revenue, the total addressable market isn’t a sector — it’s the economy. That’s a very different ceiling than “companies that buy chips.”
Thesis 2: Compute Equals GDP — Every Nation Becomes a Buyer
The second thesis is where things get geopolitical fast. Huang’s direct quote: “Compute equals GDP. Therefore, every country will have it.” The logic is tight. If intelligence is what enables productivity, growth, and competitive advantage in the coming economy, and if intelligence requires compute to run, then compute becomes as foundational as energy infrastructure or transportation networks. You don’t opt out of electricity. You don’t opt out of roads. Huang’s argument is that you won’t opt out of compute either.
His framing on this was unambiguous: “Not one country in the future will say ‘we’re going to opt out on intelligence.’” Think about what that means for NVIDIA’s customer base. It’s not just hyperscalers and Fortune 500 companies. It’s ministries of finance building sovereign AI infrastructure. It’s national health systems running inference on population-scale datasets. It’s militaries, central banks, energy grids. The demand signal Huang is describing is sovereign, not just commercial.
This also explains why the geopolitics of chip export controls became so intense over the last two years. If you believe Huang’s thesis — and the governments of the US, China, the EU, India, and the UAE all appear to believe some version of it, given their respective AI investment programs — then restricting compute access is equivalent to restricting a country’s future GDP capacity. That’s not a trade dispute. That’s an existential policy question.
Thesis 3: The Entire Internet Industry Already Proved the ROI
This is the thesis that should matter most to anyone who still thinks we’re in the “pilot project” phase of enterprise AI. Huang’s claim: every major cloud service provider has already taken their entire capital expenditure and converted it to generative and agentic AI. Not a portion of it. All of it. And critically, they did this because it works.
He named the companies: Meta, Google, AWS. And the reason is straightforward — AI makes the internet’s core revenue engines better. Search improves. Shopping recommendations improve. Ad targeting improves. Social feeds improve. These are not experimental features. They are the primary products of the most profitable companies on earth, and they are now AI-native. Huang’s quote here is worth sitting with: “The entire internet industry could take 100% of their capex and make it AI because it’s better. We’ve proven it to be better.”
That word — proven — is doing a lot of work. This is Huang explicitly saying we are past the hypothesis stage. Meta has demonstrated it. Google has demonstrated it. The ROI is in the income statements, not in a whitepaper. The implication is that any company still treating AI as a cost-center experiment is operating on outdated assumptions. The companies that are furthest along have already made the full commitment, and the results are showing up in their numbers.
Thesis 4: Every Software Company Is About to Become Token-Driven
This is arguably the most consequential thesis for anyone working in or investing in software — and it’s the one that gets the least attention. Huang’s claim: “The entire software industry will be token driven.” Here’s what that means in practice.
Tokens are the output unit of AI inference. When a model generates a response, summarizes a document, writes code, or takes an action as an agent, it produces tokens. Tokens cost compute to generate. And Huang’s argument is that every software company will either be producing tokens directly (which means they need their own compute) or reselling tokens produced by someone else (which means they are dependent on compute infrastructure). Either way, every software company becomes a compute buyer or a compute reseller. There is no third path.
Think through what this means for the traditional SaaS model. Salesforce’s CRM becomes a system that produces intelligent summaries, auto-drafts outreach, and runs agentic workflows — all token-generating operations. SAP’s ERP starts optimizing supply chain decisions through continuous inference. ServiceNow’s workflows become autonomous agents. Oracle’s database products start answering questions in natural language. Huang put it directly: “For the first time, the entire IT industry will have to be fueled by compute.” And he offered a challenge: “You pick your favorite software company and I can show you exactly how they’re going to be token driven.”
This is a significant structural claim about software economics. The per-seat SaaS model, where you charge a flat monthly fee for access to a feature set, starts to break down when the marginal cost of each user interaction is real and variable. Pricing models shift toward consumption. Infrastructure requirements shift toward inference. The software companies that figure this out early have an edge. The ones that don