The Crowd Is Wrong About AI's Death. It's Just Getting a Second Wind.
By Viktor Volkov | Against the Grain
Everyone seems convinced the AI trade is either a bursting bubble or a spent force. The sentiment across Reddit is a nervous oscillation between "crash incoming" and "don't worry, it always comes back." The magnificent seven are down 15-30%, fears of a "compute crunch" and "AI airline economics" are proliferating, and Michael Saylor is selling Bitcoin at a 50% loss—a classic capitulation signal for the crypto-adjacent tech mania. The consensus is clear: the easy money has been made, the narrative is fraying, and it's time to rotate into "safe" value or hide in cash.
I'm here to tell you the crowd is missing the forest for the dying trees. They're looking at the speculative froth receding from the shores of Nvidia and OpenAI and calling it an ocean-wide drought. They're not seeing the industrial mobilization happening beneath the surface. The real AI trade isn't about which LLM chatbot wins; it's about who builds, powers, and connects the physical infrastructure required to run them. The panic over "artificial earnings" is obscuring a very real, multi-trillion-dollar capital expenditure cycle that is only in its second inning. While retail debates if AI is a bubble, South Korea is announcing a national strategy to build an entire AI industrial ecosystem. While they fret over Saylor's margin calls, the JPM collar roll is mechanically forcing billions into the market. The signal isn't in the fear; it's in the forced flows and the geopolitical chess moves.
Look at the evidence buried in today's chatter. A highly-upvoted post details South Korea's "long-term national strategy centered on AI infrastructure, semiconductor manufacturing, advanced manufacturing, and industrial infrastructure." This isn't a company earnings call; this is a nation-state committing to a decades-long industrial policy. It’s the modern equivalent of the Marshall Plan for compute. The money will flow to the picks and shovels—the ASMLs, the Applied Materials, the power grid engineers, and the construction firms. The panic over AI software becoming a low-margin commodity (the "airline industry" analogy) is valid for pure-play software. But it makes the case for owning the hardware layer even stronger. You don't care if the airlines are profitable if you own Boeing and Airbus.
Even the fear is misplaced. The post asking "Reduced chances of crash?" correctly notes that the Mag7 have already corrected meaningfully. This isn't 1999 where the bubble pops at the peak; this is 2004, where a mid-cycle valuation reset clears the way for the next leg driven by actual earnings and capex. The real risk isn't that AI fails; it's that it succeeds beyond our current grid capacity, creating a durable, multi-year bottleneck in power and silicon that benefits a completely different set of companies. The crowd is selling the narrative; the contrarian move is to buy the bottleneck.
What If I'm Wrong?
If I'm wrong, then the AI skeptics are right: the demand surge is purely speculative, the capex plans will be slashed at the first sign of economic weakness, and the entire sector faces a decade of overcapacity and crushed margins. In that scenario, the only safe haven is long-duration Treasury bonds and companies that sell discounted beef jerky.
Methodology Note: Analysis based on approximately 200+ posts and 2,000+ comments from Reddit's investing communities over the past 24 hours. The contrarian take here feels necessary because the bearishness on AI has become a reflexive, almost performative stance. The evidence for a structural, government-backed infrastructure build-out is being drowned out by noise about bubbles. Confidence: 70%.