Is the artificial intelligence boom over before it really began? The question sounds absurd — and yet it’s precisely the kind of headline that’s begun creeping back into public discourse. However, talk of an “AI winter” is premature, and perhaps even a categorical error.
This is a classic instance of the so-called Amara Law, which states that we tend to overestimate the short-term impact of a new technology and underestimate its long-term impact. Deja vu with the internet bubble, the telecom bubble, the crypto bubble. These were ebbs and flows of financial instruments and public interests, and yet the underlying technology continued its course and eventually had a great impact in the real world, which was eventually recognized again by investors. The Godfather of value investing, Benjamin Graham, also stated that financial markets were a voting machine in the short term but a weighing machine in the long term.
What we are witnessing in AI isn’t a collapse, but a consolidation: the inevitable narrowing of experimentation into durable architectures, concentrated capital, and new power dynamics. We also knew that despite the brisk pace of progress, technology diffusion takes more time, as companies need to restructure around the new technology. It took 40 years for electricity, 20 years for computers, 10 years for the internet… it will not take more than five years for AI, and the countdown started a good 18 months ago.
DAVID HARSANYI: CHARLIE KIRK EPITOMIZED THE BEST OF OUR CIVIC TRADITION
If the early AI boom was characterized by demo-chasing and a frothy proliferation of startups, the present moment reflects the hard gravitational pull of platform economics. Valuations are not evaporating across the board, but coalescing around a handful of incumbents — Nvidia, OpenAI, Anthropic — who now function less like scrappy innovators and more like infrastructural monopolies. Meanwhile, many rank-and-file AI workers are being left behind, victims of a fraying social contract in Silicon Valley where acqui-hires replace genuine exits, and loyalty offers no equity. This is concerning, and will get worse in one year when the first AI researcher (AI agents performing machine learning research) shakes the market.
That shift is not evidence of decline. It is what technological maturation always looks like: Creative destruction giving way to oligopoly, the Wild West succumbing to a corporate rail map. But it does raise real questions: Not about whether AI is overhyped, but about where the value — and the risk — will now accrue. Hardware currently is where value accrues: parts of the current stack may be over-owned by incumbents (raising the question of Nvidia’s remaining upside), and there will be a shift toward on-device AI as a coming locus of capability and margin. Our huge data centers look like the digital highways of the 1990s, but economics teaches us that they will one day be rivaled by decentralized, on-device systems.
POLITICAL VIOLENCE ON THE RISE IN THE US: A TIMELINE OF KEY INCIDENTS
Hype today clusters around headline models; but under-discussed opportunity sits in practical deployment layers, edge inference, and products that compress cost-to-serve rather than chase benchmarks. Near-term traction is expected in applied domains — finance, robotics, biology, and other “atoms + AI” areas — where software advances meet real-world constraints. The reason that we so often talk about hype, bubbles, and corrections in the investment and corporate world these days is because of the extreme liquidity of capital: 20 years ago, when the internet stocks crashed, only public equity markets were really liquid.
Today, private and credit markets are on par with stocks when it comes to hubris, volatility and liquidity. When the main players from these markets recurrently talk about an AI slowdown or bubble, usually this is much ado about nothing.
Sebastien Laye is an economist and AI entrepreneur.