Arguably one of the dumbest things to ever happen in tech stocks in 2024 was the sudden drop in shares of Nvidia Corp., moments after it posted its fiscal second-quarter earnings in August. CEO Jensen Huang, who has otherwise performed wonders this year, reported a small — and resolved — hiatus in new chip production, but investors panicked anyway.

They soon came together. The world, as it turned out, kept turning. But what he highlighted were the deep concerns that lie just beneath the thin surface of AI optimism. The market is alert for signs of a peak and will overreact whenever it thinks it has found them. This does not bode well for 2025, when cooler heads must prevail, as the progress of AI development looks almost certain to slow, possibly to a crawl.

For the past few weeks, the mainstays in the field of artificial intelligence have been choosing their words carefully. THE Google CEO Sundar Pichaispeaking at a New York Times event, said he felt the “low-hanging fruit” of artificial intelligence had now been picked. Elaborating on the matter, he told Semafor: “As we move to this next level, you need more insightful discoveries».

THE Sam Altmanthe co-founder and CEO of OpenAI, talked about how he still believed his company would reach “general artificial intelligence” but that it would be “much less important” than some observers previously thought. Superintelligence would be the big twist, he said, but it’s further away.

In the background, several reports have suggested that OpenAI is struggling to “exorcise” the big leaps in AI capabilities that were expected. The Microsoft-backed company’s long-awaited Sora video production model can still be considered an extremely expensive gimmick. Recent model releases, with “thinking” capabilities and other bells and whistles, were a clear rehash but no less expensive than the previous models. ChatGPT enthusiasm is waning.

Apple, meanwhile, has yet to release data on the iPhone “supercycle” that some hoped would be fueled by the integration of Apple Intelligence. In fact, the company’s trial AI integration was embarrassing. His AI-generated summaries range from the comical to the seriously inaccurate. With past innovations, the company told itself it would be the best if not the first. He didn’t make it with AI either.

The first indication of its impact Apple Intelligence in the company’s results will be seen in the sales of the holiday quarter of 2024. Those will likely be records, as always, but consumers are still largely making purchase decisions based on better cameras and long-lasting batteries rather than artificial intelligence. And it’s not like there’s a definitive solution on the horizon — Cook said in June that he would “never” claim that Apple’s AI was 100% illusion-free.

This addresses a major concern as we head into the new year, that collecting more and more data on larger models will only go so far when it comes to creating “smart” capabilities, and we’ve only just gotten to that point. Even if more data were the answer, these companies that have been indiscriminately pulling material from any source they can find are beginning to struggle to acquire enough new information to feed the machine. “We only have one Internet,” lamented top engineer Ilya Sutskever, who left OpenAI to start his own AI company. Those engaged in the production of new information now, awkwardly, demand to be paid.

Some commentators say it’s clear that artificial intelligence – or genetic artificial intelligence, to be more specific – has been overhyped by those with a strong financial interest in encouraging excitement about its prospects. Others say the jury is still very much out on what AI will become. Another view is that current capabilities, when logically developed, already represent a change. I don’t know who is right. In my defense, the companies that have spent tens of billions of dollars trying to get up to speed and seem ready to spend even more before the most profitable use cases haven’t been spotted yet either. Meanwhile, efforts to balance the books by charging much more to use AI or by introducing ads could be welcomed.

None of this is an argument for selling AI stocks. It’s an argument to keep them, even if the next year is far less exciting than the previous two. Jefferies analyst Brent Thill published a recent report comparing the trajectory of AI software revenue to a slow takeoff on the runway — not the rocket ship launch enjoyed by companies that sell semiconductors or access to computing power or two.

More substantial AI software revenue may not come until 2026, Thill wrote. This will be a disappointment to many. If Nvidia’s August shakeup is anything to go by, investors will be easily spooked by any hint of bad news. Wall Street’s New Year’s resolution should be to rein in its expectations for artificial intelligence.

Dave Lee is Bloomberg Opinion’s US technology columnist. He was previously a correspondent for the Financial Times and BBC News.