Artificial Intelligence: The AI Bubble Won't Burst, It Will Just Deflate

No crash, just a slow leak. Several economists converge on the same diagnosis this summer: the expected correction in the valuations of artificial intelligence will not take the form of a brutal collapse like the internet bubble of 2000. Instead, it will resemble a gradual deflation, spread over time, patiently separating the winners from the losers. Key points of this article: * Economists have converged on the idea that the correction of AI valuations will be a gradual deflation rather than a brutal collapse, thus separating the winners from the losers. * Despite a significant impact on U.S. GDP, daily adoption of AI remains low, raising doubts about the long-term viability of these massive investments. Paulo Carvao, an economist closely following AI funding, defended this thesis on August 18 at the NCPERS forum on public pensions at the University of Chicago, in an article published on Forbes. His argument boils down to a simple idea: technology is now too deeply integrated into the U.S. economy for a pure and simple collapse to be conceivable. This does not mean that nothing will break. Carvao points to billions of dollars in off-balance-sheet commitments, a monetization that struggles to keep pace with capital expenditures, and negative cash flows at some industry leaders. In this scenario, the deflation would not be painless. It would simply be slower than a classic crash, with a gradual sorting between companies capable of generating real revenues and those that rely solely on promises. The macroeconomic weight of AI is staggering. AI-related spending contributed 1.1 percentage points to U.S. GDP, even surpassing household consumption as the main driver of the economy in recent times. Harvard economist Jason Furman goes even further: according to his calculations, investment in AI-related infrastructure would account for up to 92% of U.S. GDP growth in the first half of 2025. The problem is that this mountain of investment has not yet translated into massive on-the-ground usage. Only 13% of American workers use AI daily in their jobs, according to a Gallup survey reported by TechTarget. This figure should be viewed in light of the sums poured into data centers, chips, and models. Another concerning signal: AI adoption is reportedly declining among companies with more than 250 employees, according to data from the U.S. Census Bureau. Some companies are already looking at their AI bills and questioning whether pushing their teams to use it systematically was really the best idea. Several structural flaws are undermining the edifice. The market concentration around a handful of players poses a systemic risk in itself. If one of the big names stumbles, the shockwave mechanically affects the entire ecosystem. Agreements made in private markets remain largely opaque, complicating any serious assessment of the real risk faced by investors. Added to this is growing pressure for the most highly valued AI companies to go public, in public markets that are significantly less forgiving than venture capital. A failed IPO would be enough to change the narrative of the entire sector. The physical limits of the sector, with demand for electricity at the forefront, are also beginning to be felt concretely on the ground. Not to mention institutional concerns about job cuts and upcoming regulation, two fronts that could weigh heavily on market sentiment in the coming months. The crypto ecosystem is closely following this issue, and for good reason. Arthur Hayes, co-founder of BitMEX, sold part of his cryptocurrencies, blaming AI for draining liquidity that could otherwise have flowed into Bitcoin and digital assets. According to him, nearly $1.5 trillion in debt has already financed the expansion of AI since the end of 2022, a bill that will need to be repaid someday if the revenues generated by AI do not keep up. A slow deflation rather than a crash would, however, change the game for crypto markets. A brutal purge would likely have dragged all risky assets down with it, including Bitcoin. A gradual tapering over several quarters would leave more room for a progressive rotation of capital, rather than a widespread panic overnight. Still, none of this is set in stone. The scenario of slow deflation assumes that hyperscalers continue to honor their spending commitments without major refinancing accidents, at a time when their bond debt has already pushed U.S. 30-year rates above 5% for the first time since 2007.