When will the AI trend die?
The hype will die; the technology won't. AI already works well enough to be permanent — the same way the web survived the dotcom crash. What can collapse is the financing: hyperscalers are committing $1 trillion-plus in 2025–26 capex that outpaces their free cash flow, and the BIS compares this to railway mania. Expect a funding bust, not a disappearance.
Why — the first-principles explanation
The word "trend" is doing a lot of work in this question, and separating its two meanings answers it.
A fad dies because it never solved a problem. NFTs, fidget spinners, metaverse real estate — remove the excitement and nothing remains, because there was never a job being done. A general-purpose technology doesn't die, because the enthusiasm was never what was holding it up. Electricity, the internal combustion engine, the web: each had a hype phase, each had a crash, and none of them went away. The test is simple and you can apply it yourself: if everyone stopped talking about it tomorrow, would people keep using it? Millions of people use AI daily for translation, code, drafting, and search. That usage isn't sustained by press coverage. It survives silence. That's the signature of infrastructure, not a fad.
So the accurate prediction is that the conversation dies, and this is normal — it's what "maturing" looks like. Nobody says "e-commerce company" anymore; they say "store." Nobody markets a "cloud-enabled" product; cloud just became how software works. AI is heading for the same fate: the label disappears because it won, not because it lost. When a technology is genuinely everywhere, naming it becomes redundant.
What can genuinely collapse is the financing layer, and this is where the real answer lives. Per the BIS's June 2026 annual report, the five largest hyperscalers are set to spend over $1 trillion on AI capex across 2025–2026 — spending that outpaces their earnings and free cash flow, pushing some toward debt. Meanwhile MIT found roughly 95% of 300+ enterprise AI initiatives showed no measurable P&L impact. A gap that wide between spending and returns historically closes violently. The BIS names the precedents: canal mania, railway mania, 1920s electrification, dotcom — genuine breakthroughs that drew capital beyond what returns could justify.
Notice what survived each of those. Britain still has the railways. The 1999 fiber carried YouTube in 2006. The investors died; the infrastructure didn't. That's the realistic forecast: a lot of AI startups fail, valuations reset hard, the word "AI" quietly drops out of product marketing — and the technology becomes so ordinary you stop noticing it. Timing? Unknowable, for the same reflexive reason bubbles are never timeable.
An example that makes it click
Think about the word "electric."
In 1900 it was the hottest word in advertising. Electric corsets. Electric hairbrushes. Electric belts that cured everything. Investors poured money into hundreds of electrification companies, and in the late 1920s that exuberance collapsed spectacularly — the BIS lists it alongside railway mania as a textbook case.
So did the electricity trend die? Look around the room you're in. Every single thing plugged into a wall says no. What died was the word. Nobody sells you an "electric toaster" now — it's a toaster, obviously it's electric, what else would it be? The trend didn't end. It won so completely that mentioning it became pointless. That's where AI is headed, and the confusing part is that it looks identical to dying from the outside: the buzz fades, the startups fold, the label vanishes. Then you notice it's in everything.
Key facts
- The BIS Annual Economic Report (June 2026) reports the five largest hyperscalers are set to spend over $1 trillion on AI capex across 2025-2026, outpacing earnings and free cash flow, with some issuing debt.
- An MIT study of 300+ enterprise AI initiatives (July 2025) found roughly 95% delivered no measurable P&L impact on $30-40 billion in enterprise generative AI spending.
- The BIS compares the AI boom to canal mania (1830s), British railway mania (1840s), 1920s electrification, and the 1990s dotcom boom — all real breakthroughs followed by investment reversals and recessions.
- The IEA projects data centre electricity roughly doubling from ~415 TWh (2024) to ~945 TWh by 2030 — multi-year physical infrastructure that persists regardless of market sentiment.
- Historical pattern: after the dotcom crash wiped out investors, the fiber infrastructure laid in the late 1990s enabled streaming video in the mid-2000s at a fraction of its build cost.
- Prior general-purpose technologies (electricity, the automobile, the web) each had hype cycles and crashes without the underlying technology disappearing.
▶ The 60-second explainer (script)
When will the AI trend die? The hype will die. The technology won't. And separating those two is the entire answer. The word trend is hiding two different things. A fad dies because it never solved a problem — NFTs, fidget spinners, metaverse real estate. Take away the excitement and nothing's left, because no job was being done. A general-purpose technology doesn't die, because enthusiasm was never what held it up. Here's the test you can run yourself: if everyone stopped talking about it tomorrow, would people keep using it? Millions of people use AI every day for translation, code, drafting, search. That usage isn't sustained by press coverage. It survives silence. That's infrastructure, not a fad. So what actually dies is the conversation — and that's just what maturing looks like. Nobody says e-commerce company anymore. They say store. Nobody markets a cloud-enabled product; cloud is just how software works now. AI's label will vanish because it won, not because it lost. What can genuinely collapse is the money. The BIS says the five biggest hyperscalers will spend over a trillion dollars across 2025 and 2026 — more than their free cash flow, some of it borrowed. Meanwhile MIT found ninety-five percent of enterprise AI projects showed no measurable profit. That gap closes violently. But look at what survived every previous version of this. Britain still has railways. The fiber laid in 1999 carried YouTube in 2006. The investors died. The infrastructure didn't. Expect that: startups fold, valuations reset, the word quietly disappears from marketing — and the technology becomes so ordinary you stop noticing it.
What authoritative sources say
People also ask
Is AI a fad like NFTs?
No, by the usage test. Remove the hype from NFTs and nothing remains — no job was being done. Remove the hype from AI and millions still use it daily for translation, code, and drafting. Sustained use without attention is the difference.
So will people stop talking about AI?
Almost certainly, and that's a sign of success rather than failure. Nobody says 'e-commerce company' or 'cloud-enabled software' anymore — those labels vanished because the technologies became the default. AI is heading the same way.
What part of AI can actually collapse?
The financing. Hyperscaler capex now exceeds free cash flow with debt entering the picture, while MIT found ~95% of enterprise pilots showed no measurable profit. That gap has historically closed through a violent valuation reset.
If there's a crash, does AI stop working?
No. A crash destroys investors and startups, not capabilities. The models keep running, the data centres keep standing, and someone buys the infrastructure cheap — exactly what happened with railways and dotcom-era fiber.
Can anyone predict when the hype ends?
No. Bubbles pop when enough people simultaneously decide they will, which makes timing reflexive rather than physical. Any reliable date model would be traded on and would move the date — the prediction destroys itself.