AI Overspending Is the Strategy

Everyone keeps asking whether AI spending is a bubble. Wrong question. The bubble isn’t a bug in venture capital, it is part of the development of the product.

Here’s the uncomfortable truth about how VC actually works: the industry doesn’t just fund innovation, it manufactures the conditions for innovation by engineering mass spending and mass attention. Capital floods a sector until the infrastructure gets overbuilt, and controversy floods the media until the market gets educated. Both look wasteful. Both are the point.

Venture returns follow a Power Law. One winner pays for dozens of losers. But you can’t know the winner in advance so the rational strategy is to fund the entire search space at once. Overspending isn’t a failure of discipline; it is parallel processing for markets. The faster you burn through the bad ideas, the faster you find the one that matters.

The spending side has a twin, and it’s less understood: controversy is a distribution strategy.

Start with the economist Herbert Simon, who saw this coming in 1971: in a world flooded with information, the scarce resource isn’t information it is attention. Whoever captures attention captures the market’s imagination, and everything downstream of it: talent, capital, customers, regulation. Media theorists later formalized the mechanism as agenda-setting theory: the press doesn’t tell people what to think, it tells them what to think about. A technology that dominates the news cycle even negatively has already won the only war that matters at this stage, the war over what the conversation is about.

Psychology explains why negative coverage still converts. In Everett Rogers’ diffusion-of-innovations framework, the bottleneck for any new technology is the awareness stage. Most innovations die not because people rejected them, but because people never seriously considered them or even heard about them. Controversy blows through that bottleneck for free. “AI will take your job” and “AI is fake” are opposing headlines with the identical effect: they force every reader to form a position on AI. Once you have a position, you’re in the market.

Here are some amazing historical examples I found.

The War of the Currents (1880s–90s). Edison ran a smear campaign against Westinghouse’s alternating current by creating staged electrocutions, lurid press coverage, lobbying for the electric chair. It was ugly, and it put electricity on the front page of every newspaper in America for a decade. The controversy didn’t slow electrification; it was electrification’s marketing department. AC won anyway, and the public came out the other side understanding and wanting electric power.

Video game moral panics (1993). Mortal Kombat (my favorite character was Sub-Zero) triggered congressional hearings about violence in games. The industry’s response was having the ESRB rating system created. This legitimized gaming as a real medium, and the hearings themselves told every teenager in America exactly which game to buy. The panic was the launch campaign.

There’s a serious academic framework behind the spending side. The economist Carlota Perez, in Technological Revolutions and Financial Capital, showed that every major technological revolution like canals, railways, steel, oil, or information technology follows the same two-act structure. First comes the installation phase, driven by financial capital, which always culminates in a speculative frenzy and a crash. Then comes the deployment phase, when the surviving infrastructure gets absorbed into the real economy and produces the actual golden age. The bubble isn’t a malfunction. Instead it is the only mechanism societies have ever found for funding infrastructure whose payoff is too distant and too uncertain for rational capital.

Here are some great patterns that support this theory.

Fiber (late 90s). Telecoms and dot-com capital laid millions of miles of fiber. Most companies funding it went bankrupt. But the dark fiber they left behind made bandwidth nearly free and that cheap bandwidth is why YouTube, Netflix, and cloud computing became possible a decade later. The losers built the roads; the winners drove on them.

The internet (1995–2000). Pets.com was a punchline, but the dot-com frenzy funded the browsers, payment rails, logistics experiments, and the public debate that taught a generation to trust buying things online. Amazon survived the crash because the crash had already paid for its market education.

Mobile apps (2008–2014). After the iPhone, VCs sprayed capital at tens of thousands of apps. The vast majority died. But the spending built app store infrastructure, mobile payments, and push-notification behavior and the “there’s an app for that” media cycle trained billions of people to live on their phones. Instagram and WhatsApp were the power-law payoffs.

Today’s GPU buildout, foundation model burn rates, and existential-risk discourse aren’t anomalies. They’re Perez’s installation phase. The big four hyperscalers alone are guiding to roughly $725 billion in 2026 capex — up 77% from an already-record $410 billion in 2025. The capital is building compute infrastructure that will outlive most of the companies spending it. The controversy seen in doom threads, congressional hearings, “AI is a bubble” op-eds is legitimation running at internet speed, educating every knowledge worker on the planet for free.

Most of this money will be lost. That’s fine. It was never the money’s job to survive, instead it was the money’s job to build the infrastructure, educate the market, and find the winner. Three jobs, one method. AI overspending isn’t the bug everyone’s warning you about. It’s the strategy. It always was.