A market adjustment. A warning. An excellent digestion. Call it what you will: AI is experiencing it.
It seems as though two things are occurring simultaneously. Companies are beginning to realize what AI can do to improve their bottom lines, and more significantly, what it cannot accomplish. And the exaggerated and overhyped expectations that AI companies have been creating for the past few years are finally reaching the ground.
In summary, it appears that both the AI boomers and doomers were mistaken. The trajectory of AI is beginning to resemble computers, cellphones, and televisions more than a time machine or space elevator: In the long run, technology will improve and most likely transform our lives, but it will probably do so gradually. In fact, if superintelligence or artificial general intelligence (AGI) does eventually emerge, it may not seem like much of a jump.
Perhaps no greater illustration of this can be found in OpenAI’s most recent and eagerly awaited model, GPT-5, which was launched with great fanfare and met with a shrug. Sam Altman of OpenAI claimed before launch that he had felt “useless” in comparison to the model’s intellect, even equating it with the Manhattan Project. Users reportedly felt less afraid when it came. “The degree of overhyping was too significant,” one commenter said. Another added, “All you have is hype in the absence of massive gains.”
However, it may be a preview of our future reality, where the rapid advancement of AI is just slowing down, where excitement won’t be enough to propel development, and where we won’t suddenly see the abolition of white-collar jobs or arrive at a society dominated by AI.
Welcome to the “meh” era of AI. Keep your cool. We have already been here. Everything will work out. Probably.
Businesses were making millions of dollars overnight with only a website and a clever sales presentation as the internet revolution took hold in the late 1990s. Trillions of dollars were lost overnight as the economic reality caught up to the hype by the year 2000. Not used to it? Ask your parents about the fate of Pets.com.
It’s easy to understand why bubble talk has come up again. Even Altman just stated that he thinks the AI sector may be in a bubble, which is a very cautious statement coming from the biggest hype guy in AI.
Progress has been such that you probably won’t even notice the improvements from now on.
Carl Benedikt Frey
“There were measurable productivity gains in the 1990s when the dotcom bubble burst, even though there weren’t necessarily profits to support the investments,” says Oxford economist Carl Benedikt Frey. A check on AI today might stop history from happening again if that sounds scary.
A recent study from the Massachusetts Institute of Technology stirred things up even more last month when it claimed that only 5% of the companies it looked at had been able to turn the technology into real revenue. Even though the study had many flaws, the revelation was frightening enough to trigger a sell-off in tech stocks.
According to further data, AI is beginning to affect companies who are implementing it. AI was eliminating entry-level positions for persons between the ages of 22 and 25, according to a Stanford University study that examined payroll data. This was particularly true in industries where AI was more likely to replace labor than to supplement it. Marc Benioff asserts that thousands of Salesforce support positions are being replaced by AI agents, while other businesses brag about AI automating more of their tasks.
Earlier this year, Frey and colleague Oxford economist Pedro Llanos-Paredes conducted a research on the influence of AI on demand for foreign translators. They found that the technology was having a minor but demonstrable effect on these positions.
For a small number of companies leading the AI revolution, we appear to be witnessing respectable revenue growth, but Frey informs that this does not transfer into wider economic expansion. The fact that we’re still not seeing any indication of it in the productivity figures worries me because, in the end, that’s what counts. The performance of AI in testing or on some benchmark is essentially irrelevant. The important thing is to convert that into actual economic expansion.
A more slow adoption would be ideal for the markets. By the end of 2026, AI enthusiasm is expected to boost US equities by an additional 20%, according to Evercore ISI strategists. They stated in a memo that was released this week, AI is ‘bigger’ than the internet and even though adoption is just getting started, its impact has already permeated every aspect of industry and society in just three years.
The results of Nvidia’s earnings last week provided a clear indication of our current situation. Some of the largest tech firms are among the company’s major clients, and it has become something of a bellwether for the broader artificial intelligence boom by selling the expensive chips that AI is taught and operated on. (According to Bloomberg, Nvidia’s processors account for around 47% of Microsoft’s capital spending.) Its stock fell despite exceeding both its own sales records and Wall Street estimates, indicating that investors were unimpressed by the numbers they saw. The businesses who purchase these services from Nvidia are not yet reaping the benefits, according to some experts. In a statement that may well capture the current steady-chug-along paradigm of AI, a UBS analyst described Nvidia’s results as “good enough.”
Overall, it appears that AI has reached its iPhone 4 moment.
The iPhone 4 was a huge success when it was released in 2010. Steve Jobs bragged onstage in Cupertino that Apple had created the world’s thinnest phone with a long list of new essential features, including a high-resolution display, a front-facing camera for FaceTime and selfies, a squared-off design, and the introduction of Apple’s own silicon engine, the A4. Despite the antenna disaster, it was a huge hit and further solidified Apple’s dominance in the smartphone market. Since then, the market has reportedly been vying for the “sandwich glass” design of the iPhone 4.
A lot of this sort of feeling of disappointment is due to unreasonable levels of hype.
David Krueger
After that, things changed. The iPhone has been developing more gradually ever since, with the exception of a few little advancements. There is evidence to imply that artificial intelligence may be planning a similar strategy. As a result of Frontier Labs’ consistent release of upgrades and mini-leaps instead of waiting years between generations, every new rollout has begun to seem incremental and evolutionary.
According to Frey, the advancements have been so significant that you’re unlikely to detect them going forward if you’re not the leading expert in the sector.
Last year, the major question was whether AI laboratories were experiencing diminishing results by just throwing more data and computing power at the models. That may help explain how GPT-5 landed, but it isn’t the only cause.
Between the debut of GPT-4 in March 2023 and GPT-5 last month, OpenAI released more than a dozen models, each concentrating on a certain profession or incrementally enhancing another. It’s perhaps not surprising that GPT-5 didn’t blow us away. (In the same conversation in which Altman asserted AI is in a bubble, he also stated that OpenAI has more sophisticated models than GPT-5 but is unable to deploy them due to capacity constraints.)
Gemini 2.5, Google’s most recent frontier model, is likewise a bridge model. The release of GPT-5 would be a good way to manage expectations for Gemini 3, which is anticipated before the year ends.
David Krueger, an assistant professor at the University of Montreal who focuses on the safety and hazards of AI, says the advancement just feels more constant.
Krueger still believes that there will be “wow” moments from time to time, but he also thinks that more technical advancements will be necessary before we can achieve artificial intelligence (AI) at a level that can compete with humans. He also says that he doesn’t think large language models alone will deliver AGI.
“LLMs and deep learning in general, in my opinion, are likely a significant component of the riddle. “The biggest one, if I had to wager,” he says. “But I think we’re maybe missing a couple of puzzle pieces.”
Krueger also says that certain AI leaders are to fault for creating “unreasonable levels of hype,” and they’re now being held accountable. Even if Altman is the worst offenders, there are others.
AI will write 90% of software engineers’ code in three to six months, according to a March prediction made by Dario Amodei, CEO of Anthropic. The real benefits seem to be far less significant: CEO Sundar Pichai stated that about 30% of the code created at Google is produced by AI.
“I believe that a lot of this kind of disappointment is caused by the companies’ exaggerated levels of hype,” Krueger remarked. Expectations for the future of artificial intelligence (AI) and the potential, perhaps eventual, advent of artificial general intelligence (AGI) are now being restrained as the technology matures and its potentially explosive speed decreases.
Here’s how we got here. Altman said in a January interview with Bloomberg that AGI will be developed during Trump’s second term in office. Elon Musk had said it may arrive by the end of the year. AGI, according to some of the top AI experts, is frequently just “a few years away.” In actuality, it seems like we’re now recognizing that no one truly knows.
Perhaps no one has been humbled by AI more than Apple, which earlier this year removed an iPhone 16 advertisement that promised a number of highly anticipated new AI features that turned out to be far from ready. Don’t be shocked if Tim Cook and other executives discuss AI in a more controlled manner when they unveil the newest range of gadgets next week.
I’ve heard that this next phone will be a bit thinner.
