Even though it didn't address the "end of humanity" rethink and related slowdown calls, the bank's exercise did outline what was at stake from any spending stall on AI-related capital expenditure. It showed that capex is set to account for roughly one-third of all U.S. GDP growth this year and next, with that effect likely amplified by the boost to household wealth from sustained investment-driven equity gains.
Morgan Stanley pointed out that demand for computing power still far exceeds supply, meaning any slowdown in the AI buildout may "strengthen the position of the 'merchants of compute' – hyperscalers – by increasing the scarcity value of their installed base and enhancing their pricing power."
But "go-slows" on AI would only significantly affect the long-term direction of the AI capex and infrastructure boom if there were suggestions that AI could either be reversed or even stalled.
The political imperative to press ahead, made clear by both Washington and Beijing on Monday, shows that whatever the private sector does, governments will want to drive this capability to its nth degree — precisely because of, not despite, the risks now flagged.
One country's slowdown is another's catch-up. And agreed guardrails, rules and even regulations are only effective in the global arms race if they are agreed internationally. U.S. President Donald Trump's administration has deliberately and publicly undermined multilateralism in all its forms since he returned to power last year, replacing it with rivalry and a world of bilateral deals in which America holds the upper hand.
The prospect of a multilateral agreement on hard-and-fast rules limiting the capacity of AI seems far-fetched in that environment, and Trump made that clear.
"There is a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China," Trump blasted out on his social media site.
An editorial in China's state-backed Global Times, meanwhile, claimed the AI warnings were a "Cold War playbook" targeted at China.
Trump is scheduled to meet with Chinese President Xi Jinping next week in Washington. Reuters previously reported that the two countries were planning a separate meeting in mid-September to discuss the risks of AI, led by lower-level government officials.
But even if the U.S. and China did agree some AI protocols, other groups of countries or rivals would not necessarily be bound by them.
If arms-race analogies were to apply, you only have to look at the 23 years between the first nuclear weapon being made and the first non-proliferation treaty being signed in 1968. Since then, the number of nuclear-armed states has risen from the five permanent members of the UN Security Council, all recognised under the treaty, to nine.
AI may not be all about weapons of mass destruction, and it clearly has numerous positives, from accelerating the search for disease cures to even clean nuclear energy. But if mutually assured economic and financial destruction is a credo, governments will fiercely resist falling behind.
Investors weigh IPO delays and AI capex
There are two schools of thought. The first dismisses serial scare stories about the power of AI as merely hype to build excitement around upcoming initial public offerings (IPOs). The other is that AI development is now too big to fail strategically, and governments have no option but to press ahead if economic, financial and even military security is threatened.
It can be a bit of both, of course.
Materially, Altman said the long-awaited OpenAI IPO would now likely be delayed to next year while this gets sorted out. Although there had been several reports in recent months that such a delay was coming for other reasons, the news was enough to knock stakeholder Softbank shares 9984.T by about 13% on Monday as it must wait longer to realize its investment. Anthropic, meanwhile, still appears to be going ahead with its public listing as soon as next month.
But the bigger issue for investors is whether tech-industry calls for a slowdown are credible or likely, despite similar AI protocols having been suggested or signed before. The question is whether they will directly affect the frenzied investment, trade and construction building the AI infrastructure of gigantic data centers and computing capacity.
Goldman Sachs expects cumulative global AI spending through the end of the decade to reach as much as 5% of global gross domestic product. With much of that funded by debt and equity, the timing of the eventual payoff may be a concern, especially as estimates of business adoption and earnings impact remain relatively modest.
Looking solely at what rising domestic political opposition to U.S. data center building might mean, Morgan Stanley's strategists estimated that almost $200 billion of projects were cancelled or delayed last year and through the first quarter of 2026.
AI slowdown warnings collide with geopolitical rivalry
The opinions expressed here are those of the author, a columnist for Reuters.
By Mike Dolan
LONDON, Sept 15 (Reuters) - If cutting-edge AI is truly powerful enough to end humanity, it may already be too powerful to slow down. With the technology now locked in a strategic arms race tied to national security, warnings about its risks may only speed up the dash for hegemony — and bring far greater government control with them.
Perhaps greater government control, in ownership if not regulation or restraint, is partly what's unnerving financial markets, which have priced the spectacular investment boom to see out the decade at least.
A series of apocalyptic warnings about AI's threat to human life reached a crescendo over the weekend. They came with indications that the most sophisticated models could be used to develop biological weapons, hack finance and even seize the entire internet.
The warnings began last week when a resigning employee of AI lab Anthropic warned that the new technology "could kill us all" by the end of the decade. By Sunday, there appeared to be a consensus among Anthropic boss Dario Amodei, OpenAI's Sam Altman, Elon Musk at SpaceX SPCX.O, Apple's AAPL.O Tim Cook and others about the need to slow development down, increase human intervention and checks, and ensure clear limitations on the extent of AI agents' autonomy. Microsoft MSFT.O published its own draft code.
QQQ•
