The expiration in 2031 of a measure introduced in 2017 and extended by the Big Beautiful Bill could precipitate the bursting of the artificial intelligence bubble… if there is a bubble.
Remember the DeepSeek tornado, which at the start of the year sowed doubt about the business model of American AI giants. Faced with the Chinese David capable of designing a large language model with (relatively) few resources, was it really reasonable for the American Goliaths to continue to invest crazy sums in the construction of giant data centers aimed at giving them ever more computing power?
Looking at how the rest of the year has gone for the giants in question, the answer is undoubtedly yes. Microsoft, Google, Meta and Amazon combine for more than $300 billion in capital investment spending (capex) in 2025, most of it dedicated to AI infrastructure. xAI, for its part, has undertaken the construction of a giant gargantuan data center bringing together a million graphics processors, a simply unprecedented figure, at a cost estimated between 150 and 200 billion dollars.
A private sector recovery program
These actors do not intend to stop there. “It seems reasonable to estimate that American companies will spend more than a trillion dollars on AI infrastructure over the next five years,” estimates Ian Graham, senior analyst at Te Ahumairangi, an investment fund. AI has become a key driver of the American economy: over the first six months of the year, capital investment spending in AI contributed more to growth across the Atlantic than consumption!
In a note published in July, investor Paul Kedrosky showed that spending on infrastructure dedicated to AI had already exceeded that devoted to telecom and internet infrastructure during the dot.com bubble and continued to grow. He goes so far as to assert that one of the explanations for the current strength of the American economy, despite Donald Trump’s tariffs, is that spending on IT infrastructure is so significant that it would act as a kind of recovery program that would be implemented by the private sector rather than by the State. The markets took note: this summer, Microsoft became the second company in the world, after Nvidia, to reach the stratospheric valuation of $4,000 billion.
The origin of Big Tech’s lavish spending
The causes of these stratospheric investments are naturally multiple. There is of course the arms race phenomenon which pushes each AI leader to invest ever more in computing power for fear of seeing their cutting-edge models eclipsed by those of the competition, with the risk of disappearing in an American technological market marked by the winner-takes-all rule.
There is the bull market that surrounds the AI giants (the “Magnificent Seven” currently account for 37% of the S&P 500, an unprecedented share) and the investor confidence that allows companies to spend lavishly (at $500 billion, OpenAI is the most valuable private company in history).
But the spending frenzy of Big Tech is also due to a law that often goes under the radar and yet plays a determining role. In 2017, shortly after his first takeover of power, Donald Trump passed the Tax Cuts and Jobs Act (TCJA). In addition to various tax cuts on individuals and businesses, it introduces a rule that allows American businesses to immediately deduct, in the year of purchase, all eligible investment expenses related to their activities.
Instead of spreading the tax deduction over several years via depreciation, they obtain an immediate tax reduction equivalent to 100% of these expenses. This measure therefore promotes greater cash flow in the short term, thus encouraging massive investments in technological infrastructure. Indeed, if it does not specifically target investments in AI, the expenses incurred in the construction of data centers being particularly costly, this law has played an accelerator role to encourage tech companies to equip themselves with data centers and semiconductors to increase their computing power.
Towards a bursting of the AI bubble in 2031?
Supposed to expire gradually from 2023, this measure was extended until 2031 by the Big Beautiful Bill, the budget law passed by the Trump administration earlier in the year. Until then, the effective tax cost of large investments is therefore still lowered for American companies, stimulating the rapid development of AI infrastructure.
“Tax reform through President Trump’s One Big Beautiful Bill has played a key role in accelerating capital investment spending on AI,” said Richard Clode, portfolio manager at investor Janus Henderson. “For large tech companies spending tens of billions on R&D and AI/cloud infrastructure, this is a major benefit from a tax and cash flow perspective.”
But this also implies that this effect will disappear after the rule expires, which could slow down the dynamics of the AI market… and lead to the bursting of the famous bubble, which many observers and AI professionals fear.
This is notably what James Thorne, of the Wellington-Altus fund, asserts in a recent message published on his
In other words, tech giants have an incentive to continue spending lavishly until January 1, 2031, after which the incentive disappears, and we enter a great unknown. It is of course possible that the future American president will extend the deadline again to avoid the risk of an economic downturn, but the United States is facing major budgetary difficulties, facing which Donald Trump’s successor could be forced to tighten the screws to redress the deficit, at the risk of bursting the AI bubble.
All this assumes, of course, that we are indeed facing the formation of a bubble, which some experts dispute. “The dot-com bubble of the 1990s reveals what happens when enthusiasm exceeds real utility. In contrast, the deployment of the cloud in the 2010s shows how things can go well with measured deployment and clear paths to monetization, recalls Ian Graham at Te Ahumairangi. The path AI takes depends on two factors: how quickly concrete use cases emerge in the real world, and how effectively companies can turn that use into sustainable revenue.
