Another AI Winter Is Coming—But This One Will Be Different

Everyone’s drunk on AI right now. NVIDIA is printing money, every CEO says “AI” five times per earnings call, and investors are acting like productivity gains will show up any minute now.

But what if they don’t? What if the technology’s payoff takes longer than the hype cycle can tolerate? What if we’re heading into (another) AI winter?

It’s a decades-old term every insider knows all too well. Each one arrives when expectations outpace reality, when the story moves faster than the science, and when investors run out of patience.

While we are indeed in an AI summer right now, a revolution unlike any before—it’s important to understand history. Just as we study stock charts to spot patterns, we can study the history of AI to see how each boom eventually cools before the next ascent. The AI industry will keep climbing, but in fits and spurts, not in a straight line.

A Quick History of AI Booms and Busts

Hard to believe, but for about seventy years AI has moved in waves, each one bigger, faster, and more expensive than the last.

• 1956 to 1970: The Dawn.
  • Optimism from the first AI conference at Dartmouth. AI scores early wins in logic and game-playing computers. Then the limits of hardware and failed translation projects triggered the first chill.
• 1970s: The Lighthill Winter.
  • The UK’s Lighthill Report kills government funding and the US quietly pulls back.
• 1980s: Expert Systems Boom.
  • Corporate America builds rule-based “intelligent” software. It works for a while, then maintenance costs explode and the funding collapses.
• 1990s to 2000s: The Quiet Years.
  • AI hides under new labels like “machine learning,” “data mining,” and “neural networks.” Slow, steady progress under the radar.
• 2010s to 2020s: The Deep Learning Spring.
  • GPUs, data, and cloud power drive breakthroughs in speech, vision, and language. Suddenly AI works, at least in demos.
• 2023 to 2025: The Generative Euphoria.
  • ChatGPT, Midjourney, Claude, Gemini. Hype hits escape velocity. The race is on and so are valuations.

In every previous cycle, the crash came quietly. AI was smaller, academic, and mostly invisible to the public markets. This time is different.

Now, AI is the market.

When the next chill comes, it will not just be research labs and startups tightening belts. It will hit the portfolios of everyday investors who hold the ETFs, the chip stocks, the cloud providers, the megacaps funding it all.

The last AI winters were whispered among scientists. The next one will be broadcast live on CNBC.

So the question is not if an AI winter comes—but what happens to Main Street investors when it does.

1. Demo vs. Product

Andrej Karpathy, former director of AI at Tesla and one of the field’s most respected engineering voices, said it best:

“Self driving cars have been demo-ready since the 1980s… but making them reliable enough for the real world takes decades.”

It is the same story with AI. The demos are dazzling, we see glimpses of intelligence that feel world-changing, but turning those demos into durable, profitable products is another matter. The “March of 9s,” moving from 90 percent to 99.999 percent reliability, is where the real work happens, and it is slow, expensive, and often unglamorous.

Karpathy recently added another telling line:

“Debugging AI-generated code can take longer than writing it yourself. We’re not there yet, and it could take longer than people think.”

Technology always looks finished on YouTube. Then it hits the real world, with regulation, human behavior, and edge cases that break everything.

2. Productivity Takes Time

Everyone is counting on massive productivity gains, the kind that rewrite GDP charts and justify today’s valuations. But those gains do not just appear overnight. Companies have to rewire workflows, retrain staff, clean data, and rethink entire processes.

That kind of transformation unfolds over years, not quarters.

When results do not show up on schedule, investors start to lose faith. The headlines shift from “AI Revolution” to “Where’s the ROI?” and suddenly the air gets colder.

3. Mini AGI / ASI Winter

Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI) may one day emerge, but right now they remain more marketing construct than measurable progress. Until then, the mini winter around these grand narratives will cool some of the heat, as investors realize intelligence at scale is still bounded by physics, compute, and patience.

In early 2025, the paper “AI 2027” by Daniel Kokotajlo and Scott Alexander reignited the hype cycle, predicting a near term leap to AGI through self replicating agent networks, agents teaching agents, building robots, building factories, building more robots. The vision was cinematic. The timing was not.

I believe the theme, but the timeline is crazy. These claims came from former OpenAI insiders, smart people, but often the most heavily invested in their own mythology. Big money was raised on the concept, but the reality is nowhere in sight.

OpenAI suggested that AGI was on the way with ChatGPT 5, and instead investors got an AI content generated social network called Sora. What are those investors thinking today?

4. Winners and Losers

That is when the market begins to sort itself out. The strong, well capitalized players with data moats and distribution survive. The rest disappear into the snow.

In an AI winter, inventors and investors alike often flee to safety, back to the familiar warmth of big tech. The cycle always repeats: retreat, regroup, then reemerge when the next breakthrough reignites belief.

There is a greater than zero chance that one of today’s AI darlings, Anthropic, Perplexity, even OpenAI, could be gone in five years. Not because the vision was wrong, but because scaling that vision into a sustainable business is brutally hard.

Rather, we are entering a mini winter, a cooling off period where exuberance gives way to realism and valuations finally reflect execution risk.

Conclusion

If the impending cooling arrives in a market built on leverage, with hyperscalers financing data centers on debt and chipmakers booking years of preorders, the shakeout could spread quickly.

A few counterparties fail. Funding tightens. The same investors who called it a revolution start calling it a bubble.

But even if that happens, I remain long AI. The progress is real, and the endgame will be transformative. The big productivity gains will come, just later than people expect.

The frost will come, but so will the thaw.

Because every AI winter has always ended the same way, with spring.

The early-rate deadline for the 2026 Inc. Regionals Awards is Friday, November 14, at 11:59 p.m. PT. Apply now.