Dario Amodei, Sam Altman and Elon Musk call for moderating, not slowing, the development of AI

"Slow down" and not "slow down": Dario Amodei (Anthropic), Sam Altman (OpenAI) and Elon Musk (SpaceX) have not asked to stop artificial intelligence (AI), but rather to moderate the speed at which it develops. For this reason, some of the large management and investment firms yesterday called for calm and not removing all the eggs from the technological basket. Basically, because the technological race between countries to achieve technological "supremacy" is already underway - and none will back down, US President Donald Trump already made it clear yesterday -, and because the most recent forecasts point to stratospheric figures: global investment in AI will reach one trillion dollars in 2026, up to 1.6 trillion dollars in 2027 and 4.3 trillion dollars in 2031, according to analysts at Citigroup, one of the largest financial services companies in the world. "With these levels of demand, we consider a widespread 'sudden stop'" in the development of data center infrastructure unlikely, they argue. Now, Citi sees a greater risk of "a more significant slowdown in spending" in technologies "if burdensome regulations are imposed on the sector."

But, in terms of investment, "continuous learning in AI will transform the demand for long-term memory." For example, they welcome investment in HBM memory chips (High Bandwidth Memory), of which they note a supply deficit despite being essential for training AI models (they expect demand to skyrocket by 62% next year). Although they will not be the only beneficiaries: the expansion of the centers will continue and will require multiplying their production of semiconductors, networks, land, buildings or electrical infrastructure.

All this takes time. And that is the question that worries the markets: how much it will delay benefits if the pace is slowed, as OpenAI and Anthropic are now asking. The British manager Schroders analyzed the situation yesterday based on this question: there is a "relevant" gap between the investment and the economic return of AI data centers. Data center construction can take 18 to 24 months (depending on regulation) and deeper enterprise applications require an additional year or two of implementation and organizational changes before their impact is reflected in the bottom line, so a dollar invested in AI infrastructure in 2025 may not translate into customer value until 2028 or later.

For patients, Schroders estimates that the deployment of AI infrastructures will generate $5.4 trillion annually in cumulative investment (about €4.66 trillion at current exchange rates)… by the middle of the next decade. The volume is equivalent to 20% of current global corporate operating profit and 4.5% of global GDP, according to the same report, released yesterday.