The recent stock market correction of American technological values coincides with a degradation of the story on artificial intelligence, which recalls the Deepseek episode of the beginning of the year. The hiccups around GPT-5, a MIT report and tense valuations have awakened the spectrum of the speculative bubble, which has never been very far in recent months.
What you will find in this article: a dive into the growing doubts that surround the AI, between a shocking ratio of the MIT, the failed launch of GPT-5 and the febrile financial markets which fear a speculative bubble. You will also see how the promises of Openai and giants in the sector are collided with fragile economic models and a much less triumphant than announced business.
The title of Fortune slap “MIT report: 95% of generative AI pilot projects fail in businesses”. The dissemination of the report has not been unrelated to the access of weakness of American technological actions since the beginning of the week. Yesterday, the counterperformance of the large AI -style displacements pulled down a s&p 500 index, 70% of the components, closed in the green down. The case of Palantant is symptomatic: -15% in five sessions, after the record affected at 190 USD on August 12. We can also quote Coreweave among the new markers of the hype of the sector.
A shock report
The report The GenAI Divide: State of AI in Business 2025published by the Nanda MIT initiative, draws up a severe observation on the use of generative artificial intelligence in companies. Despite the massive craze and significant investments, only about 5% of experiments lead to a significant acceleration of income. Most stagnate and do not bring a measurable impact on the accounts. Successful cases mainly concern agile startups, often very young, which focus on a specific problem, perform quickly and build effective partnerships.
The MIT highlights an integration flaw more than a technological problem. The models themselves are efficient, but their deployment often fails because they do not adapt to internal processes and they remain too generic for business use. In addition, the allocation of resources is often poorly oriented: more than half of generative AI budgets leave in sales and marketing tools, while the best return on investment is observed in the automation of back-office tasks, the abolition of outsourcing and rationalization of operational costs. In the end, the main tangible impact is visible on employment, with a gradual reduction in positions in customer support and administrative functions, mainly via the non-replacement of departures.
For a technology whose adoption is relatively recent, nothing very surprising, especially since important developments such as agents are only little taken into account. But the balance sheet is to be balanced with the colossal investments already engaged and with the promises which have been made by certain influential bosses of the sector.
Show me the money !
However, for a few days, some investors have been afraid of living a second “Deepseek moment” on the stock market. At the start of the year, the irresistible bullish stock market for Western IA companies had been shaken by the emergence of open Chinese competition, cheap and performance deemed relatively close to well -established models. Characteristics that called into question the hegemony of the actors in place and the prospect of ever -increasing infrastructure spending.
Palabuting is visible during the Deepseek episode. We have seen a new dropout for a few days.
Over the weeks, the pitfall was bypassed because everyone ended up repersed that the investment cycle remained extraordinary. Remented two nagging questions: how useful all this is and how far will it go?
It is this double question that makes its return in force at the end of August. First of all because the crazy launch of GPT-5 by Openai aroused new questions about general artificial intelligence (AG). Then because the economic model of companies in the sector is anything but stable. Let’s examine these two points one after the other.
Acted and bubble
Gary Marcus, a recurring criticism of current AI approaches, underlines in an interview with the great continent that the stagnation observed with GPT-5 demonstrates that Optai has lost its technical advance and that Altman’s initial declarations on AGE appear now hollow (the CEO of Openai itself agreed last week that the concept of Agre could lose its relevance). Marcus adds that this highlights a structural limit of the great models of language (LLM): their inability to generalize out of their training data, which resembles a systemic deficiency.
As for the question of the economic model, it is always at the heart of the debates on technological ruptures. Altman, again, compared in The Verge the current period to the Internet speculative bubble of the Millennium Turning, 25 years ago. But he thinks that AI is a major change whose positive economic effects will prevail over the negatives. “People will lose a lot of money but others will earn a lot” he sums up, while recalling in the same interview that he hopes to raise thousands of billions of dollars to build data centers capable of supporting the development of Openai.
Silence, it digs
For the moment, the flagship company of the sector, if we except Nvidia (for equipment) and Palantir (for deployment), is a financial chasm: $ 3.7 billion in turnover and $ 5 billion in losses in 2024 for OPENAI. This year, turnover could rise to 12/13 billion dollars and losses around 8 billion. Analysts find it difficult to project themselves. JPMorgan Chase has advanced the date of 2029 so that the company is positive in terms of liquidity generation, if income growth is there and if the investments are tenable.
The PER 2026 of a few US companies exhibited at AI
Chatgpt would currently have 700 million users, 10% of whom would be paid, estimates Wired (Openai did not want to confirm). Widely insufficient to compensate for the necessary infrastructure expenses, even by adding the additional income generated by OPENAI (professional solutions, agents, etc.). But convincing enough to attract colossal capital from investors who are looking for the next Amazon, Alphabet or Meta.
This source should not dry up in the short term. It remains to be determined if Openai and its peers will finish the current upheaval of the right or bad side. We remember that at the end of the 1990s, telecom operators passed to be the natural winners of the Internet revolution. The rest showed that the value had migrated elsewhere.
