This scenario is almost a classic. Carlota Perez, a researcher specializing in technological revolutions, has shown that every major tech breakthrough comes with an investment bubble that accelerates innovation but always ends up bursting. AI is unlikely to escape this rule: huge amounts of money pour in rapidly, often irrationally, until reality reasserts itself. Even Sam Altman, the CEO of OpenAI, admits that investors are “overexcited” about AI and that we are indeed in a bubble phase. Quite ironic, considering that with his regular “over-promises,” he is probably one of the main architects of this frenzy.
95% failure rate? A diagnosis that needs nuance
Let’s go back to that famous 95% AI project failure figure that fueled the panic. Should we really be alarmed when this number comes from a sample of only 52 large companies and when a project is deemed a “success” only if it yields a return on investment within six months?
Even the report’s authors admit that AI itself is less to blame than companies’ ability to transform. AI isn’t magic: like any technological project, it produces nothing if the organization isn’t structured to adopt it.
And in the flood of AI innovations we’re living through, is it really surprising that large AI projects are stalling? By the time a custom tool is developed and rolled out, a new service comes out and makes it obsolete…
Useful AI should not be confined to the IT department
That large-scale projects are struggling is one thing, but it certainly doesn’t mean we should throw in the towel — or worse, restrict the use of AI in companies by reserving it exclusively for the IT department. Unfortunately, we see some IT teams slamming the brakes whenever employees want to use “consumer” AI tools (like ChatGPT), even when the security risks are minimal. Do employees simply give up? Of course not: they turn to their personal accounts.
In fact, the MIT report highlights the success of this AI “shadow IT” — the unofficial use of digital tools by employees. The researchers found that in more than 90% of the companies studied, employees regularly use consumer AI tools for work, even though only 40% had purchased an official solution. AI spreads from the ground up because it is easy to access and immediately meets concrete needs. Performance-wise, employees often get more value from these “unofficial” practices than from the heavy, complex projects deployed by the company.
Naturally, giving employees a Microsoft Copilot account won’t magically transform the company. Training teams on general-purpose tools will have a limited effect if deeper issues remain unaddressed. The main barrier to AI effectiveness today is not technological — it is organizational.
Back to reality: All the more reason to invest smartly
If the AI bubble does burst, at least it will cool things down and bring us back to earth. Gone will be the tech companies’ wild promises, replaced by the need to design seriously, fine-tune carefully, integrate patiently with existing tools, and provide continuous oversight. In short, to reconsider AI for what it truly is: a formidable tool capable of delivering enormous gains, but one that requires method and perseverance. That’s good news. As with any bubble, the burst clears away illusions and brings forward the projects that truly create value.
In any case, this is certainly not a reason to stop investing in AI — quite the opposite. We must continue investing, but intelligently. That means learning from recent failures: no more high-tech cathedrals disconnected from real-world needs. Instead, we should invest in concrete, human-scale use cases with a clear improvement goal. We should bet on rapid iteration and continuous adaptation, without fearing a modest start. AI is not a passing fad: even after the speculative bubble bursts, it will remain a driver of innovation and performance. In summary, the goal is not to believe in AI miracles, but to invest in it like any strategic asset: progressively, with measurable use cases, and without engaging in escalation.
Alexandre Villeneuve and Edouard Fillias, authors of The AI Survival Manual (2025, Ellipses) and founders of the communication agency JIN.
