Several signals indicate that technology comes up against difficulties and has an unexpected slowdown in its progression. Should we fear a new winter of AI? Nothing is less certain.
Are AI’s cogs in the process of seeping? Several recent elements make it possible to worry about this sector hitherto marked by an exceptional dynamism in a gloomy economic context. The first alerts came from Meta. During the summer, the company of Mark Zuckerberg froze the hires in its IA teams, after an intense period of recruitment during which it spent without counting to obtain the best talents. The Californian giant has also pushed the launch of its LLM LLAMA 4 Behemoth, because its engineers, however among the best in the world, struggled to improve it significantly compared to previous dirts. Folling to progress, Meta has set up a partnership with the young Midjourney shoot to try to revive.
95% of AI pilot projects in business fail
Meta is not the only one to have experienced this kind of difficulty: GPT-5, the latest Openai model, disappointed and caused a return of a stick for the company of Sam Altman. The father of Chatgpt, usually rather known for his optimistic predictions, recently expressed doubts about the stratospheric valuations of AI companies, establishing a parallel with the Internet bubble.
Eric Schmidt, the former boss of Google, has co -signed a forum in the New York Times in which he worries about the obsession of Silicon Valley for general artificial intelligence, which pushes an esoteric race disconnected from the real economy, where Chinese companies have a more pragmatic approach, which mobilizes the AI to solve concrete problems.
Because the disturbing signals are not only on the side of research: elements also show that professionals still struggle to absorb recent AI progress and derive the profits. A largely relayed study of MIT claims that 95% of AI pilot projects observed in business have made it possible to generate any tangible profit.
Professionals would be “vast majority skeptical towards AI tools, whether tailor-made or acquired directly from the seller, describing them as fragile, too complex and not very suitable for real professional needs”, affirm the authors of the study.
Companies continue to invest massively
Taken together, these difficulties have something to worry, to the point that some experts see a spectacular crash of the ia on the stock market, in order to deflate the bubble that has been created around technology. The current AI difficulties, although very real, should not be overestimated. On the one hand, Nvidia has just made an excellent second quarter, beating the expectations of Wall Street. However, sales of its processors, necessary for the training and operation of AI algorithms, are a good barometer for the general health of the sector.
In addition, companies continue to invest without counting in AI. The Swiss Bank UBS estimates that $ 375 billion will be spent on IA infrastructure this year, a figure that should increase to 500 billion next year, according to UBS. Proof that companies indeed expect concrete economic benefits from this technology.
Software and IT equipment spending on AI has counted for a quarter of American economic growth in the last quarter: this must still be added staggering expenses in data centers, which now support growth than shopping in the United States. The AI thus helped to give a serious boost to the American economy: in the second quarter, it recorded a higher growth rate than expected (3.3 %), carried in particular by this technology. “Investments related to AI mask certain weaknesses elsewhere in the economy. The good news is that there are few signs that this support is in the process of fade,” said Ryan Sweet, economist at Oxford Economics.
“There can of course be bubbles on certain companies, which today display an incredible valuation. But the basic trend of the market remains in adoption. We are carrying out a survey based on conversations with business leaders each year. The latest shows that 75% are in an acceleration logic of investments in AI,” said Alex Bauer, Director General of IBM Consulting.
Why a slowdown in AI would not be just a bad thing
The signals as to a possible AI tray are therefore mixed. But even assuming that progress in technology is starting to slow down, it would not necessarily be negative for the economy. Indeed, all cycles of innovation consist of essor periods followed by equally spectacular crash. In his book Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages, the historian Carlota Perez shows that, during each great technological revolution, from the industrial revolution to that of the NICT, a strongly disruptive first period is always accompanied by overinvestment and a financial bubble. This bubble also has its advantage, since it offers the funding necessary to build the infrastructure which will then allow the development of technology.
The burst of the bubble then allows you to clean up the market and give technology time to infuse in society after a strong growth period, while giving the authorities latitude to adopt new adequate regulations. Thus, at the end of the 19th century, governments adopted antimonopole laws to limit the power of new industrial conglomerates, and implemented the welfare state to protect the precarious workers. Likewise, a slowdown in AI growth, which has progressed at a frantic pace since the launch of Chatgpt, would allow companies, which quickly feel exceeded in the face of the speed of technological changes, to plan the adoption of it more serenely and to extract the noun marrow.
