Jensen Huang declared that Artificial General Intelligence already exists. He said it in a conversation that circulated widely, with the confidence of someone who isn't announcing a discovery but signing a decree. The problem isn't that he's wrong about the technology. It's something subtler: he didn't make a technical announcement. He redefined a concept.

The definition Huang used boils down to this: AGI is any system capable of generating a billion dollars in economic value. By that metric, we've indeed arrived. But that definition is a far cry from what the scientific community debated for decades under that term. AGI, in its original sense, refers to systems with general cognitive capacity comparable to humans: abstract reasoning, knowledge transfer across domains, genuine adaptation to novel situations. What Huang described is something else entirely. He took a word loaded with technical and philosophical meaning, hollowed it out, and refilled it with a financial metric conveniently suited to whoever sells the infrastructure holding up the whole scaffolding.

This matters because language shapes realities. When the CEO of the most valuable graphics chip company declares that AGI has arrived, that statement moves concrete things: public policy, regulation, investment decisions, corporate contracts, government budgets. Imprecise language generates very precise institutional consequences.

I recognize this pattern in other contexts. It's not new. It happened with "the cloud," when companies that sold servers rebranded the service and built dependencies at a global scale. It happened with blockchain, where the promise of radical decentralization ended up concentrated in a handful of exchanges and venture capital funds. And with the metaverse, which lasted about one attention cycle before revealing that nobody actually wanted to live wearing a virtual reality headset. In each case, the structure was the same: new tool → narrative of radical transformation → infrastructure controlled by the usual players. The name changes. Who collects the check doesn't.

One detail that went unnoticed in much of the coverage: in that same conversation, someone asked whether a hundred thousand AI agents could build Nvidia. Huang said no. That admission deflates his opening statement. If AGI exists and generates massive economic value, but can't replicate the very company that supports it, we're not talking about general intelligence. We're talking about tools that are capable in specific domains—faster, and more expensive to run, than what came before.

Whoever defines the terms controls the narrative. This idea connects to a recurring thesis: new technological tools don't automatically redistribute power. They consolidate it. If Nvidia defines what AGI is, it decides when we've arrived, what's needed to sustain it, and, by extension, who to buy it from. The definition isn't a semantic nuance. It's a mechanism of control.

A historical parallel is worth considering. In the 19th century, the expansion of railroads promised to democratize commerce and access to markets. In part, it did. But it also concentrated economic and political power in the hands of those who controlled routes, rates, and hubs. Farmers, dependent on those rails to sell their produce, learned that the liberating tool could just as easily strangle them with price hikes. The technology was real. The democratization narrative had a basis in fact. But the structural power remained in place—or migrated to new players who soon behaved exactly the same way.

What's curious is how these cycles keep speeding up. The metaverse took years to deflate as a narrative. Enthusiasm for blockchain covered an entire market cycle. The idea that AGI has already arrived, with its convenient redefinition, might last even less time. More people read between the lines: researchers, engineers, and specialized journalists have been flagging this dynamic for a while now. Plenty of people are saying it. The headlines just travel faster than the corrections.

Rather than accepting imposed definitions, open source communities could forge their own metrics for AGI, grounded in verifiable and accessible capabilities rather than profits alone. That doesn't solve the structural problem, but it opens the door to more distributed governance.

That's the real cost of swallowing the headline whole. The institutions that regulate, the governments that set AI policy, the funds that allocate resources don't follow skeptical threads on social media. They pay attention to statements from CEOs who have every incentive to present transformation as a done deal. Declaring AGI arrived shifts the debate from "should we build this?" to "how do we adapt?" It's a powerful rhetorical maneuver.

It's still not clear how to solve the structural problem: those who create the tools also create the vocabulary to describe them, and that vocabulary translates into policy before any auditing takes place. What is clear is that the first line of defense is simple: read the whole thing, not just the headline. Huang's admission was in the very same talk. Hardly anyone quoted it.

There's something hopeful in that, paradoxical as it may seem. If the deception works through superficial attention, the remedy doesn't require alternative technology or global regulations. It requires a habit. Asking who defines the terms, and why now. Recognizing that when the person selling picks and shovels declares gold has been found, the question isn't how much gold there is. It's how many picks and shovels they're about to sell.

Stones don't lie, but historians sometimes do.

Sources:

1. Statements by Jensen Huang in a public interview (various coverage, January 2025)

2. Bostrom, N. Superintelligence: Paths, Dangers, Strategies (2014) — original framework for the AGI debate

3. Doctorow, C. The Internet Con: How to Seize the Means of Computation (2023) — consolidation of power across technology cycles

4. Cortada, J. W. The Digital Flood: The Diffusion of Information Technology Across the U.S., Europe, and Asia (2012) — historical parallels in technology diffusion

5. Marcus, G. & Davis, E. Rebooting AI: Building Artificial Intelligence We Can Trust (2019) — technical critique of general AI narratives