Huang drops the agi bomb: it's already here and can mint billion-dollar startups
Jensen Huang just moved the finish line. The Nvidia CEO told Lex Fridman's microphone that artificial general intelligence is no longer a horizon event—it walks among us, seeding companies that could hit a ten-digit valuation before the next presidential cycle closes.
The wolf is inside the house
Fridman pinned him: could an AI agent alone found, scale and operate a billion-dollar tech shop within five-to-twenty years? Huang answered in one breath—yes—then warned the room to brace for a graveyard of viral toys that flame out after two months of hype. Translation: the same intelligence that can reason through a supply-chain simulation still can't tell a Tamagotchi feeder from the next Nvidia.
Zero drama in his voice, just the flat arithmetic of someone who ships 80 % of the world's AI silicon. Huang's engineers already burn $250 k a year in inference tokens per half-million salary; he expects that ratio to double as models learn to 'think' before they speak. Inference is the new training, and the power bill is moving from the data-center row to the CFO's nightmare column.
Co-design or die
He buried another lede: classical Moore's Law is dead, and the only defibrillator left is co-design at birth. Software, silicon, network, liquid cooling—one blueprint, one team, one massive rack. If your GPU tray wasn't sketched next to your PyTorch kernel, you're already late. Huang's bet is that future models will devote half their transistor budget to self-verification loops, turning chatbots into cautious attorneys who cross-examine themselves before every answer.
Meanwhile, Microsoft and Nvidia are quietly redrawing nuclear plant blueprints to unplug the power choke point. When your megawatt appetite competes with small cities, friendship bracelets between chip and reactor vendors become survival strategy.
No meetings, no mercy
governance footnote feels almost quaint: Huang runs a $3 trillion company without a single 1:1 staff meeting. Information travels in open Slack channels and war-room whiteboards; hierarchy is replaced by latency. If your update can't fit in a 140-character DM, it probably isn't ready for silicon.
The takeaway isn't that AGI will spare us. It's that the survivors will be the ones who treat compute, capital and curiosity as the same raw material—melted together under the same heat sink. Everyone else gets the Tamagotchi graveyard.