A lot of people are coming out now saying, “The AI bubble just burst. I told you so. This whole thing was a fad.”

And you know what? They might be right. Maybe today is the day the bubble bursts, NVIDIA’s stock crashes, OpenAI becomes a relic of the past, and Microsoft has to lay off a bunch of workers because AI is done. Apparently, this Chinese company—the one that made DeepSeek with just a few million dollars—is triggering the AI apocalypse.

Here’s the thing: I’ve seen this story before. Every time there’s a big downturn, people panic. Bears come out of the woodwork to say, “I told you so.” But long term, things tend to stabilize and stocks go back up. I’m not saying recessions don’t happen—they do. I’ve lived through a few myself: the dot-com crash of 2000, the Great Recession of 2007, the COVID-triggered downturn five years ago, and 2022, which was an incredibly hard year for me.

But if you look at the long-term trend, betting on stocks falling indefinitely is a bad call. Historically, the stock market averages about 10% growth year over year. Sure, some years you get 20–30% gains, and others you see 20–40% losses. But when you smooth it out over the long haul, it’s been 10% growth for a century.

Now, back to DeepSeek. For years, I’ve been saying AI is bound to get more efficient—it has to. That’s just how computing works. A hundred years ago, computers ran on vacuum tubes. Now, the computing power of those room-sized machines fits on a nanometer, or even smaller. Computing used to be massively resource-intensive, but over time, we’ve learned to do more with less. The same will happen with AI.

It’s a bad bet to think AI-driven data servers will expand indefinitely, requiring more energy until we’re building nuclear reactors to power them. That doesn’t make sense. At some point, you hit diminishing returns. People have been warning about this for years.

Take audio as an example. We hit the peak of audio reproduction decades ago with the CD, introduced in 1982. The Red Book standard defined it, and since then, every so-called improvement has been marginal. Sure, we’ve had SACDs, DVD audio, Blu-ray audio—but none of them caught on because the benefits over CDs were nominal. Yes, SACDs can do surround sound, but even CDs can handle that, just not as well. It all comes back to diminishing returns. Audio didn’t get better—it just got more convenient. In fact, in some ways, it got worse. Spotify, for example, has a lower bitrate than CDs. But streaming is more accessible, so people trade quality for convenience.

The same principle applies to AI. We’ve already had the big breakthrough: machine learning. We can take a low-resolution video in 240p and upscale it to 4K. We can translate languages in real-time. That’s the breakthrough. Everything from here on out is incremental improvement. It’s not about building massive data centers and boiling oceans to keep them running. It’s about making AI more cost-effective and efficient.

If the reports about DeepSeek are true—that they pulled this off with $7 million and without the latest NVIDIA chips—then the hundreds of billions spent on AI over the past two years have been massively inefficient. Is that bad news for NVIDIA or OpenAI? Maybe.

But it also means we’re heading toward cheaper, more efficient AI. The next big leap isn’t going to be massive data centers; it’ll be resource-efficient AI running on smartphones and embedded systems. Imagine AI on a $30 Raspberry Pi. That’s the future.

The real fear in the stock market isn’t that the AI bubble has burst—it’s that AI has advanced so quickly that much of what’s been built is already obsolete. That’s bad news for investors but great news for builders who couldn’t previously afford a data center’s worth of compute power.

Now, think big picture. With AI built into your smartphone, you could shoot a video, edit it in minutes, and make quick changes on the fly. Tasks that took hours are now done in moments. And all this hype about AGI or the industrial revolution of AI, where jobs disappear and humans are replaced? It’s just that—hype. The tech industry does this every few years. They promise the moon, there’s a breakthrough, and then the improvements become incremental, like with the CD.

Does this mean the tech industry will collapse if investments in AI don’t pay off? No. The tech industry is already hyping the next big thing—robotics.