By Asa Fitch
The advance of cutting-edge AI is showing signs of slowing. For many companies looking to harness the technology, that wouldn't be a terrible thing.
Excitement about AI reached feverish levels at the end of 2022 with the release of OpenAI's ChatGPT and has stayed red-hot since. A regular cadence of more impressive large language models from startups and big tech companies alike has kept the party going, lifting stocks -- including shares of AI-chip giant Nvidia -- to lofty heights.
Nearly three years later, it looks increasingly as though those models are plateauing.
This summer, Meta Platforms delayed the rollout of the next iteration of its flagship AI model, called Llama 4 Behemoth, because engineers were struggling to significantly improve it.
The latest model from OpenAI, called GPT-5, was delayed, and when it did come out it didn't measure up to the hype. Normally ebullient Chief Executive Sam Altman has sounded more like an AI realist recently, saying at a media dinner that he thought investors had become overexcited about the technology.
But if the leading AI tools indeed are losing steam, that wouldn't be a huge problem for many of the companies trying to integrate AI into how they work. It might even be welcome.
Generative AI is already powerful and useful in business -- for summarizing large texts, for example, and helping employees code or write emails. Other more mundane forms of AI that predated the language-model explosion have also become increasingly useful, for tasks such as processing invoices or giving recommendations on how to manage a fleet of vehicles.
Yet most businesses have barely scratched the surface with AI in its current form, let alone what it could become if it gets a lot better.
While some have been quick to deploy AI, many more have been slow. Corporate tech leaders worry about sensitive data being leaked through chatbot conversations. They are wary of trusting AI with critical decisions that affect finances, employees and customers.
The tendency of even the best AI models to occasionally hallucinate wrong answers widens the trust gap.
A recent MIT study found that companies were already mostly comfortable with off-the-shelf generative AI tools from OpenAI and Microsoft. But when it came to building custom AI software to streamline their operations -- the kind of things ostensibly most likely to produce real business returns -- the failure rate for pilot projects was 95%.
Corporate AI users "were overwhelmingly skeptical of custom or vendor-pitched AI tools, describing them as brittle, overengineered, or misaligned with actual workflows," the study's authors said.
AI won't stand still even if the purveyors of its most advanced models run into a wall: People will look for novel ways to improve it, and they will likely work eventually. Somewhat paradoxically, the mere perception that AI is slowing might give companies more confidence to invest time and money in it, seeing it as less of a moving target.
And the corporate world clearly does need more time to figure it out. The work of adapting large-language models to be useful for everyday tasks remains in its infancy.
"If I want to increase the pace at which I innovate, if I want to reduce the safety stock of inventory I have, if I want to improve the way I connect with millions of consumers, you have to change a lot more than just, 'Here's a tool that a few of your workers can use,'" said Michael Chui, a senior fellow at the AI arm of McKinsey & Co. "All of that change is really hard."
That difficulty -- a thorny management challenge as much as a technical one -- means corporate AI adoption will be a multidecade effort, Chui says.
That shouldn't be entirely surprising. The internet transformed how people live and do business, but it took much longer than its enthusiastic early boosters would have predicted in the 1990s. For instance, it took a decade for U.S. home broadband to go from near zero penetration in 2000 to more than 60% of adults' subscribing to it, according to figures from the Pew Research Center.
The AI boom is different in many ways, but it could follow a similar trajectory: a burst of enthusiasm followed by a leveling-off as it bleeds into society and business, with the true scope of the benefits only clear years later.
In the shorter term, the sense that AI's rise might be less meteoric than previously thought has sent tech stocks on a bumpy ride. Nvidia, Microsoft, Amazon.com, Meta and other AI leaders sold off last week before Fed Chair Jerome Powell's comments pointing to an interest-rate cut sparked a Friday rally.
Ironically, though, that it is getting tougher to squeeze out better performance from AI models might result in an extension of the boom for some, especially "pick and shovel" makers such as Nvidia. Altman, Meta Chief Executive Mark Zuckerberg and other big AI spenders likely will put yet more money behind their attempts to overcome recent challenges.
Altman recently suggested the cure to OpenAI's recent stumbles was to spend trillions more dollars on AI chips. And even the process of adapting models to real-world business tasks will require more incremental computing power.
Eventually, there is reason to suspect that the big winners from today's AI boom won't prosper quite as much, should the pace of AI innovation moderate.
But it isn't just Big Tech that stands to gain from AI: The payday for all the companies starting to leverage it will come -- it might just take longer.
Write to Asa Fitch at asa.fitch@wsj.com
(END) Dow Jones Newswires
August 24, 2025 06:30 ET (10:30 GMT)
Copyright (c) 2025 Dow Jones & Company, Inc.
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.