Day: September 4, 2025

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.

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.

“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.

• Kraft Heinz is splitting up 10 years after Warren Buffett spearheaded the food giants’ mega-merger.
• Buffett gurus said that combining them was a “rare” misstep by the legendary investor.
• Kraft Heinz is battling changing consumer tastes, cheaper rivals, and weight-loss drugs, they said.

Warren Buffett’s dealmaking made him a legend, but the breakup of Kraft Heinz marks one of his few missteps, the investor’s close followers say.

Buffett’s Berkshire Hathaway conglomerate partnered with Brazilian private equity firm 3G Capital to acquire Heinz for around $23 billion in 2013. Two years later, they teamed up again to merge Heinz with Kraft in a $40 billion deal.

Buffett, a fast-food aficionado, called it “my kind of transaction,” adding that the combined company could “supply you Heinz ketchup or mustard to go with your Oscar Mayer hot dogs that come from the Kraft side. Add a Coke, and you will be enjoying my favorite meal.”

An ill-fated merger

Buffett, who plans to retire later this year after 55 years in charge, loves to own famous brands with beloved products, as they can raise prices to counter inflation, and loyal customers help them to fend off competition.

Kraft and Heinz fit the bill. But Buffett’s partnership with 3G Capital was a departure from his playbook.

Private equity firms often seek to quickly cut a company’s costs and change its management to raise its value so they can sell it for a profit and use the money to fund their next acquisition.

In contrast, Berkshire is known for offering permanent, hands-off ownership to businesses it buys, and seeks to avoid bloat instead of eliminate it, as Buffett laid out in his 2015 shareholder letter.

After the Kraft Heinz merger, there were layoffs, management overhauls, massive writedowns, and asset sales. Its aggressive cost controls “significantly impaired” its ability to innovate, the Harvard Business Review wrote.

The company also settled a federal accounting probe and had a protracted decline in net revenues amid shifting consumer preferences.

David Kass, a finance professor at the University of Maryland and a longtime Berkshire blogger, told Business Insider that merging Kraft and Heinz was a “rare mistake” for Buffett.

He pointed to Berkshire having to write down the value of its stake by billions of dollars twice: in 2018 and again in the second quarter of this year. Kass added that the investment has yielded a lower return than the benchmark S&P 500 index over the past decade.

On Tuesday, Kraft Heinz said it plans to split into two businesses. One will house Heinz, Philadelphia, and Kraft Mac & Cheese, and concentrate on sauces, spreads, and seasonings. The other will house Oscar Mayer, Kraft Singles, and Lunchables, and focus on North American staples.

Berkshire and Kraft Heinz didn’t respond to requests for comment from Business Insider but Buffett told CNBC that he was “disappointed.” He told the broadcaster that while the merger wasn’t the best move in hindsight, he was skeptical that dividing up the company would solve its problems.

The “Oracle of Omaha” has said before that acquiring Kraft was an error. He told Berkshire’s 2019 shareholder meeting that he “made a mistake” and “paid too much money” for it.

Kraft Heinz stock was down more than 70% from its 2017 peak before the split was announced. It fell another 7% on Tuesday before regaining some ground on Wednesday. The company’s market value has fallen from over $110 billion at its peak to below $33 billion.

Berkshire is Kraft Heinz’s largest shareholder with a 27.5% stake. Buffett told CNBC he didn’t rule out a sale, saying he would do what was best for Berkshire. 3G sold the last of its stake in 2023.

Buffett’s rare mistake

Kass said it was “puzzling” and an “apparent admission of failure” by management to split up Kraft and Heinz and undo the synergies created. He said it wasn’t clear to him how this would generate value for shareholders. The company has said it anticipates $300 million of “dis-synergies” from the split, although it hopes to mitigate some.

John Longo — a finance professor, investment chief, and the author of “Buffett’s Tips” — said acquiring Heinz was a “home run” for Buffett, but merging it with Kraft was a “rare misstep.”

Longo said the merger generated fewer savings than expected. He also said that Kraft Heinz had faced challenges from weight-loss drugs like Ozempic that reduce people’s appetites, and several years of high inflation, which has pushed some consumers to switch to cheaper, generic alternatives.

Alex Morris, the head of TSOH Investment Research and author of “Buffett and Munger Unscripted,” said he agreed with the Berkshire boss that combining Heinz and Kraft “wasn’t a great deal” but “splitting them up won’t fix the key issues.”

Morris said the company’s challenges largely reflected mounting industry pressures on Kraft’s legacy business, as consumer tastes have shifted toward fresher, healthier, and more natural alternatives.

Adam Mead, the author of “The Complete Financial History of Berkshire Hathaway,” told Business insider that he didn’t view the Kraft Heinz deal as a “major blunder.”

“They’ve collected dividends along the way and still own great assets,” Mead said. ‘They’re just not worth as much as originally thought.”

Buffett’s errors are rare, but he has admitted to them in the past.

He said his 1993 purchase of Dexter Shoe, a Maine shoemaker that crumbled under pressure from cheap foreign imports, was his “most gruesome” mistake and a “financial disaster.” He bought the “worthless business” using Berkshire stock that would be worth nearly $19 billion today.

• The next Super Bowl halftime show will take place in February 2026.
• NFL Commissioner Roger Goodell said Taylor Swift could “maybe” perform at next year’s show.
• Swift has never headlined at a Super Bowl halftime show before.

National Football League Commissioner Roger Goodell said on Wednesday that Taylor Swift could “maybe” perform at next year’s Super Bowl.

Goodell was speaking to NBC’s “Today” show host Savannah Guthrie when she asked him if Swift could headline the Super Bowl halftime show in 2026.

“We would always love to have Taylor play. She is a special, special talent, and obviously she would be welcome at any time,” Goodell told Guthrie.

When asked if plans to have Swift appear were in the works, Goodell said he couldn’t tell Guthrie “anything about it.”

“It’s a maybe,” he said, adding that he was waiting for Jay-Z to confirm the performing lineup. The headlining act for the next halftime show has yet to be announced. The 2026 Super Bowl will be its 60th.

Jay-Z’s entertainment label, Roc Nation, has produced the Super Bowl halftime shows since it teamed up with the NFL in 2019.

“It’s in his hands. I’m waiting for the smoke to come out,” Goodell said on Wednesday.

Representatives for Swift and Roc Nation did not respond to requests for comment from Business Insider.

Swift has, according to multiple media reports, turned down the opportunity to perform at the Super Bowl several times.

In 2022, TMZ reported that Swift was offered the stage at the 2023 Super Bowl, but said no because she wanted to focus on rerecording her first six albums. Rihanna ended up headlining the 2023 halftime show.

A year later, The Daily Mail reported that Swift turned down the halftime show to focus on her Eras Tour. The 2024 halftime show was headlined by Usher.

Last month, Swift announced her engagement to Travis Kelce, the tight end of the Kansas City Chiefs. Swift and Kelce have been dating since 2023.

“Your English teacher and your gym teacher are getting married,” Swift wrote in an Instagram post featuring Kelce’s proposal on August 26.

Swift’s next album, “The Life of a Showgirl,” will be released on October 3.