Something in that cheery narrative has changed during the past few weeks, however.

Goldman Sachs called generative AI “overhyped” and “wildly expensive”; VC firm Sequoia Capital said “the AI bubble is reaching a tipping point”; a spate of media headlines such as “The AI Hype Machine is Running On Empty” are zealously pouring cold water on the whole affair.

Why? Well, generative AI chatbots struggle to answer basic questions or hallucinate incorrect information. The most sophisticated generative AI models are constantly hungry for data and computing power. Generative AI startups with little to no revenue have to constantly scrounge for massive funding rounds to stay afloat. Fortune 500 companies can’t put generative AI use cases into production because of concerns about accuracy, liability and security. 

And with the S&P 500 suffering its biggest selloff in two years on Monday, there’s a growing sense that the Generative AI bubble has begun to deflate.

Gartner’s hype cycle now says generative AI has passed the “Peak of Inflated Expectations” and is headed straight for a looming “Trough of Disillusionment.” If that’s true, what comes next will be painful and disruptive. Investment dollars could dry up. Startups could fail. There could be layoffs.

For many of the startup employees, founders, and investors who put in the work and took the risks necessary for the generative AI sector to take off, the sting of the market correction will be unjust and brutal. But knocking generative AI off of its lofty pedestal is also necessary for the long-term sustainability of the AI landscape, Kjell Carlsson, a former analyst at Forrester Research who is now head of AI strategy at enterprise data platform Domino Data Lab, told me. 

“I’m fairly confident that folks will recognize that Gen AI isn’t all the AI,” he said, referring to the wide variety of other artificial intelligence technologies, including predictive AI and machine learning, that were already delivering real return-on-investment before generative AI came on the scene. “Gen AI is one set of technologies that are part of this broad toolkit of different technologies that take work,” he explained. “There’s no magic button, it’s all about leveraging technologies for the right use cases.” 

Don’t be afraid of the trough

Let’s be clear: generative AI is not going away. These models and tools, from ChatGPT and Microsoft Copilot to Google’s Gemini, Anthropic’s Claude and Meta’s Llama, have already become part of our lives– for productivity, for efficiency, or just for fun. Just as we’ve become accustomed to getting any information we need in seconds by doing a Google search, so too will the ability to obtain easy-to-read summaries of work meetings, to compose memos to colleagues, and to create images and presentations by speaking just a few words.

But let’s also get real: The massive amount of generative AI investment, estimated to be at the tune of $1 trillion, has yet to pay off. Much of that may not be as ridiculous as, say, the dot-com bubble of UrbanFetch and Pets.com (I well remember getting ice cream deliveries and puppet swag), but it’s difficult to argue against the notion that generative AI is getting the reality check it deserves. 

“The irony of this is that I think I was the first of the industry analysts to jump on the Gen AI bandwagon,” said Carlsson. “While it’s been a success by anyone’s measure, the expectations around how quickly that would impact the bottom line of major organizations weren’t based in reality.” 

That’s where the so-called Trough of Disillusionment becomes an important stage for any tech development, Gartner’s global chief of research Chris Howard said in a recent video. The premise is simple: After an initial burst of excitement and enthusiasm by early adopters, new technology makes its way into the hands of mainstream users who find it doesn’t live up their overinflated expectations. A retrenchment follows, during which the technology is refined and expectations are reset.

“It’s not this dark, dangerous place,” Howard explained in the video. “It’s where we figure out how to make something work–or not.” 

For generative AI, the trough will be a phase marked by small incremental progress in applications that deliver real benefits to businesses and to users, and less by proclamations by OpenAI CEO Sam Altman about creating “the most powerful technology humanity has yet invented” with artificial general intelligence (AGI) — though it might make for less sexy headlines.

Even Dan Ives, a Wall Street tech analyst at Wedbush who remains bullish on AI stocks, said this is a key period for tech companies to walk the walk, not just talk the talk, when it comes to generative AI. They need to “show the use cases and monetization to justify the AI Revolution,” he told me in a text.

Ives said that he believes Microsoft, AMD, Nvidia, Palantir and Oracle have shown they can deliver real value. Still, with so many generative AI startups riding on multi-billion-dollar valuations, the sector as a whole still has a lot to prove.

There are no guarantees, but there is a long history of AI technologies that have become mature and gone on to contribute to other, newer AI disciplines, like computer vision– which has become a key part of today’s multimodal generative AI (AI that can generate not just text but images and video, for example). 

So perhaps generative AI, pushed along by other, newer technologies like agentic AI (AI systems designed to act like autonomous agents to pursue complex goals and workflows) can still reach its full potential.

Now, perhaps, it’s time for the real down-and-dirty work in generative AI to begin. “I think this will be a false AI Winter,” said Steve Jones, an executive VP at tech consultancy Capgemini, in a LinkedIn post,. ”One where hopefully the hype dies, and we can concentrate on getting work done.”