AI Hype Is Crashing Into Reality. That’s Good for the Tech’s Future.

A market correction. A wake-up call. A great digestion. Call it what you want: AI is going through it.

Two things appear to be happening in tandem. Businesses are starting to finally grasp what AI can — and importantly, can’t — do to boost their bottom lines. And the sky-high expectations that have been partly inflated and overhyped by AI firms over the past few years are finally coming down to Earth.

In short, it’s increasingly looking like both the AI doomers and boomers were both wrong. AI’s trajectory is starting to look less like a time machine or space elevator and more akin computers, smartphones, televisions: The technology will get better, it will almost certainly change our lives in the fullness of time, but it will more likely do so incrementally — to the point that if AGI (artificial general intelligence) or superintelligence do in fact one day arrive, it might not seem like much of a leap at all.

There’s perhaps no better example of this happening than OpenAI’s latest and long-anticipated model, GPT-5, which was touted with a bang and landed with a shrug. Ahead of launch, OpenAI’s Sam Altman said he’d felt “useless” compared to the model’s intelligence, even drawing parallels with the Manhattan Project. When it arrived, users apparently felt less intimidated. “The degree of overhyping was too significant,” one person wrote. “In the absence of massive gains, all you have is hype,” wrote another.

But it may be a glimpse at our new reality, where the breakneck speed of AI progress is simply steadying, where progress cannot run on hype alone, and where we will neither experience an overnight white-collar job wipeout nor reach an AI abundance society overnight.

Welcome to AI’s “meh” era. Stay calm. We’ve been here before. It’ll all be fine. Probably.

When the internet revolution took hold in the late 1990s, companies were minting millions overnight with little more than a website and a savvy sales pitch. By the year 2000, the economic reality caught up to the hype, leaving trillions of dollars wiped out overnight. Not familiar? Go ask your parents what happened to Pets.com.

It’s easy to see why talk of a bubble has once again reared its head. Even Altman recently (and in an unusually measured moment from AI’s biggest hype man) said he believes the AI market might be in a bubble.

Progress has been such that you probably won’t even notice the improvements from now on.Carl Benedikt Frey

“If you go back to the 1990s when the dotcom bubble burst, there weren’t the profits necessarily to back the investments up, but there were tangible productivity gains,” says Carl Benedikt Frey, an economist at Oxford. If that sounds eerily familiar, a check on AI now could prevent history from repeating itself.

A recent study published by the Massachusetts Institute of Technology further stirred the pot last month, claiming that just 5% of companies it studied have managed to convert the technology into actual revenue — a revelation scary enough to cause a tech stock sell-off, even if the study had a lot of limitations.

Related stories

Other evidence suggests AI is starting to have an impact on businesses that are adopting it. A study from Stanford University researchers that analyzed payroll data concluded that AI was killing off entry-level jobs for people aged 22 to 25, and was especially doing so in fields where AI was more likely to replace, rather than augment, labor. Marc Benioff claims AI agents are replacing thousands of Salesforce’s support roles, while other companies are boasting about AI automating more of their work.

A study of AI’s effects on the demand for foreign translators by Frey and fellow Oxford economist Pedro Llanos-Paredes, published earlier this year, concluded that the technology was having a small but provable impact on these jobs.

“We seem to be seeing reasonable revenue growth for a handful of firms that are pioneering the AI revolution, but we’re not seeing it translate into broader economic growth,” Frey tells me. “What I find concerning is that we’re still not seeing any hint of it in the productivity statistics, and ultimately that’s what matters. It doesn’t really matter how well AI performs on tests or on some benchmark. What matters is translating that into real economic growth.”

For the markets, a more modest uptake may be perfectly OK. Evercore ISI strategists predict AI excitement will buoy US stocks a further 20% by the end of 2026. “AI is ‘bigger’ than the internet,” they wrote in a note published this week. “In three years, its effect has touched all parts of society and industry even as adoption only begins to inflect.”

Last week’s Nvidia earnings were a strong indicator of where we’re at. The company, which sells the valuable chips on which AI is trained and run, has become something of a bellwether for the entire artificial intelligence boom and counts some of the biggest tech giants as key customers. (Bloomberg estimates that Microsoft spends about 47% of its capital expenditures on Nvidia’s chips.) While it beat Wall Street expectations and its own sales records, its stock still dropped, suggesting investors weren’t impressed with the figures they saw. Some analysts warn that the companies buying these services from Nvidia aren’t yet seeing the returns. One UBS analyst characterized Nvidia’s results in a way that may perfectly sum up the new steady-chug-along paradigm of AI: “good enough.”

All to say, AI seems to have reached its iPhone 4 moment.

When Apple’s iPhone 4 arrived in 2010, it was nothing short of a smash hit. Onstage in Cupertino, Steve Jobs boasted that Apple had made the thinnest phone in the world with a laundry list of new must-have features: a squared-off design, high-resolution display, a front-facing camera for FaceTime and selfies, and the debut of Apple’s custom silicon chip, the A4. It flew off the shelves — despite the antenna fiasco — and further cemented Apple as the king of the smartphone. One could argue that the market has been chasing the iPhone 4’s “sandwich glass” design ever since.

A lot of this sort of feeling of disappointment is due to unreasonable levels of hype.David Krueger

Then things changed: With the exception of a couple of moderate leaps since, the iPhone has been on a more incremental trajectory. Evidence suggests artificial intelligence might be plotting a similar course. Frontier labs have been rolling out a steady flow of updates and mini-leaps, rather than waiting years between generations, and as a result, each new rollout has started feeling evolutionary, incremental.

“If you’re not the leading expert in the field, I think the progress has been such that you probably won’t even notice the improvements from now on,” says Frey.

Last year, the big topic was whether AI labs were seeing diminishing returns when simply trying to throw more data and compute power at the models. That may go some way to explain how GPT-5 landed, but it’s not the only factor.

Between the launch of GPT-4 in March 2023 and GPT-5 last month, OpenAI rolled out well over a dozen models, each one focusing on specific jobs or incrementally improving another. It’s perhaps no wonder, then, that GPT-5 didn’t blow our socks off. (In the same conversation in which Altman claimed AI is in a bubble, he also claimed that OpenAI has more advanced models than GPT-5, but can’t deploy them because it doesn’t have the capacity.)

Google’s latest frontier model, Gemini 2.5, is also a bridge model, and the launch of GPT-5 may serve as a healthy warning to temper expectations for Gemini 3, which is expected before the year’s end.

Related stories

“The progress just feels more continuous,” says David Krueger, an assistant professor at the University of Montreal who studies AI safety and risks.

Krueger still thinks we’ll have the occasional “wow” moment, but said he also believes we’ll need more breakthroughs on the technical side to reach any level of artificial intelligence that can go toe-to-toe with a human — and he says he doesn’t believe we will reach AGI from large language models alone.

“I think LLMs and more broadly deep learning are probably a big piece of the puzzle. If I had to bet, the biggest one,” he says. “But I think we’re maybe missing a couple of puzzle pieces.”

Krueger also lays blame at the feet of certain AI figureheads who have created “unreasonable levels of hype,” and who are finally getting a reality check. Altman may be the worst offender, but he’s not the only one.

In March, Anthropic CEO Dario Amodei predicted that AI would be writing 90% of software developers’ code in three to six months. The actual gains appear to be much more modest: During Alphabet’s Q1 2025 earnings call, CEO Sundar Pichai said that more than 30% of code written at Google was being generated by AI.

“I think a lot of this sort of feeling of disappointment is due to unreasonable levels of hype from the companies,” said Krueger. As the rubber hits the road and the breakneck speed of AI potentially slows, expectations for the future of AI — and the possible, maybe, one day, arrival of AGI — are finally being put in check.

You can see how we got here. In a January interview with Bloomberg, Altman predicted AGI would arrive during Trump’s second presidency. Elon Musk once predicted it could be here by the year’s end. According to some of the best minds in AI, AGI is often just a “few years away.” In truth it feels like we’re finally realizing nobody actually knows.

Perhaps nobody has had more of a humbling in AI than Apple, which earlier this year axed an iPhone 16 ad that promised several much-hyped new AI features that, as it turned out, were far from ready. When Tim Cook steps onstage next week, don’t be surprised if he and other executives strike a more measured tone when talking about AI, as they pull back the curtain on the latest lineup of devices.

I hear the new phone will be a little thinner this time.

Hugh Langley is a senior correspondent at Business Insider where he writes about Google, tech, and wealth.

Business Insider’s Discourse stories provide perspectives on the day’s most pressing issues, informed by analysis, reporting, and expertise.