AI’s trillion-dollar gamble
THE risks around artificial intelligence (AI) are no longer limited to whether the technology can live up to its promise. As spending on AI infrastructure keeps climbing, investors are increasingly having to grapple with what happens to markets and the wider economy if the expected returns fail to keep pace.
That question becomes more important as AI-related investment accounts for about half of US gross domestic product (GDP) growth, according to Bloomberg Economics estimates.
Nearly US$33 trillion in market value has also been added to the S&P 500 since OpenAI released ChatGPT publicly in late 2022, with much of that gain tied to companies whose futures are increasingly connected to AI.
The scale of the investment means there is a lot riding on the technology continuing to develop rapidly and companies continuing to spend heavily on it.
“People may not fully grasp just how wound up the market and the economy is in all of this,” Jim Morrow, chief executive officer of Boston investment firm Callodine Capital Management, tells Bloomberg.
“There are just so many things to unravel if it starts.”
The debate takes on a new dimension as some of the people running America’s leading AI development labs begin discussing whether the development of increasingly powerful models needs to slow down.
Anthropic PBC chief executive officer (CEO) Dario Amodei writes in a Sept 12 blog post that AI companies need more time to strengthen oversight and install safeguards following security breaches. He argues that the threats could increase exponentially as AI becomes capable of improving itself more rapidly.
OpenAI CEO Sam Altman and Elon Musk, who is developing AI through SpaceX, quickly express support for the idea.
Whether AI companies can actually coordinate or even choose to slow development remains unclear, however.
The commercial incentives are enormous, while the amount of capital already committed to the technology continues to rise.
Fresh uncertainty
The discussion also comes as SpaceX completes the largest initial public offering (IPO) in history and Anthropic prepares for a potentially huge listing of its own. OpenAI, meanwhile, is not going public this year as it focuses on safety, Altman tells Fortune in an interview on Sept 12, according to Bloomberg.
US President Donald Trump criticises Amodei’s proposal, saying on social media that “the only one that is happy about it is China”.
China rejects the plan as well. Foreign Ministry spokesman Guo Jiakun says that “fearmongering, confrontation and vicious competition will only disrupt the process of global AI governance and serve the interests of no one”, according to Bloomberg.
For investors, the debate adds another layer of uncertainty to an AI trade that is already showing signs of strain.
The Philadelphia Stock Exchange Semiconductor Index, better known as SOX, has fallen 19% from its June 22 peak after doubling at the start of the year, Bloomberg points out.
Chipmakers are among the biggest beneficiaries of the AI spending boom because they supply the semiconductors needed to train and operate increasingly sophisticated models.
“If we see AI development slow, that means capital expenditure (capex) is likely to slow,” Anthony Saglimbene, chief market strategist at Ameriprise, tells Bloomberg. “Any slowdown would reset the profit expectations for the entire ecosystem. Given how concentrated the market is to AI, that would be a severe headwind.”
The spending involved is already moving beyond the balance sheets of the companies that started the AI race.
Funding pressure
For years, Alphabet Inc, Amazon.com Inc, Microsoft Corp and Meta Platforms Inc are able to finance their AI investments using excess cash generated by their core businesses. But their ambitions are becoming so large that the same funding model is coming under pressure.
The four companies are expected to spend more than US$1 trillion on capex in 2027 alone. At the same time, their free cashflow is expected to disappear, pushing them towards debt and equity markets for additional financing.
There are also short- and long-term commitments covering leases and energy worth almost US$2.4 trillion, most of which is related to AI, according to Bloomberg.
That makes the direction of interest rates particularly important. Ten-year US Treasury yields have recently moved above 5% for the first time since 2007.
Higher borrowing costs make it more expensive to finance the data centres, chips, power systems and other infrastructure needed to support AI.
Bond yields for high-grade technology companies have risen by about one percentage point this year. That translates into roughly another US$1bil of annual interest payments for every US$100bil borrowed.
The big technology companies still have healthy balance sheets and plenty of borrowing capacity. But the economics of AI spending become less comfortable as the cost of capital rises.
Torsten Slok, chief economist at Apollo Global Management, tells Bloomberg that the next six months are likely to provide a test of whether AI returns justify the spending.
If the numbers do not add up, he writes in an Aug 29 note that the Nasdaq 100 could fall as much as 50%.
The reason markets are so sensitive to this question is the sheer contribution AI is already making to economic growth.
Bloomberg Economics estimates that AI-related investment accounts for about half of the roughly 2% growth in US GDP over the past year, based on data from the Bureau of Economic Analysis.
Venture capital funding in the United States also exceeds US$400bil in the first half of the year, with almost all of it going to AI-related start-ups, according to PitchBook and the National Venture Capital Association.
That rush of private capital is being followed by a push towards public markets as companies seek more funding.
SpaceX, which owns AI company xAI, raises US$86.2bil in a June IPO. Anthropic is aiming to match or exceed that amount in an offering expected later this year, Bloomberg News reports.
For investors, those listings offer a rare opportunity to look more closely at the financial performance of companies sitting at the centre of the AI boom.
“The IPOs are ‘a very important window’ into the financials of the companies at the heart of the AI boom,” Saglimbene tells Bloomberg, adding that investors need that information to assess the outlook for AI spending.
“OpenAI delaying was the warning shot,” he says. “If Anthropic delays, all bets are off.”
Concentration risks
The concentration of stock market gains makes the stakes even clearer.
Since ChatGPT launches on Nov 30, 2022, almost three-quarters of the S&P 500’s 93% gain comes from just 20 companies, most of which are part of the AI trade, according to Bloomberg.
Nvidia Corp stands out above the rest. Its shares rise more than 1,300% over the period, accounting for about 16% of the S&P 500’s gain.
The chipmaker dominates the market for semiconductors used to train and run AI models.
Its revenue is estimated at US$410bil for the fiscal year ending in January, with net income of US$239bil. Four years earlier, revenue is about US$27bil and net income is US$4.4bil.
The AI boom also spreads into businesses supplying the infrastructure needed to power the technology.
Vertiv Holdings Co, which makes cooling and power systems, sees its shares climb more than 1,700% since ChatGPT’s debut, making it the fourth-best performer in the S&P 500.
Its revenue more than doubles during the same period.
The growth is not merely showing up in share prices. It is also appearing in corporate profits. More than half of the S&P 500’s US$320bil in profit growth since the end of 2022 comes from just 19 companies tied to the AI boom, according to data compiled by Bloomberg Intelligence.
That concentration is prompting concerns about what Bloomberg describes as an “earnings bubble” if those enlarged profits prove difficult to sustain.
The comparison with the dot-com boom of the late 1990s is therefore difficult to avoid.
Back then, hundreds of billions of dollars pour into fibre-optic cable and other infrastructure to support the expected explosion in internet traffic.
A much earlier example comes from the railway boom about 150 years ago, when heavy investment in railroad construction fuels a speculative frenzy.
The common thread is that transformative technologies eventually live up to at least some of their promises, but the path there can involve spectacular market busts.
“This has the potential to play out the same way,” Michael Mullaney, director of global market research at Boston Partners, tells Bloomberg.
“Will there be winners? Absolutely. It’s hard to say who is going to wind up on the other side of this thing and coining money to justify all their expenses. There will be, but it’s not going to be a boatload of companies, it’s going to be a handful of companies.”
Selective measures
There are already signs that investors are becoming more selective.
Nvidia’s valuation has fallen even as its revenue growth remains the fastest among the Magnificent Seven technology companies.
Its shares trade at 16 times expected profits over the next 12 months, compared with an average of 33 times over the past four years, according to Bloomberg.
The Nasdaq 100 also struggles to regain its June 2 record after falling 11% through July 29.
One reason is that AI is changing the financial model of some of the biggest technology firms.
These businesses become market favourites partly because their capital-light operations allow them to generate huge profits. They can then reinvest the money or return it to shareholders through share buybacks.
That model is becoming less straightforward as AI infrastructure demands increasingly large amounts of capital.
In 2022, capex by Alphabet, Amazon, Meta and Microsoft totals about US$150bil. The figure is now about five times higher.
By 2027, the four companies are expected to record combined negative free cashflow of about US$50bil, based on the average of analyst estimates compiled by Bloomberg. In 2024, they generate combined positive free cashflow of about US$230bil.
The change is already affecting shareholder returns.
Most of the companies stop buying back shares, while some are turning to equity markets to finance data-centre expansion. Alphabet announces plans in June to raise a record US$80bil through equity sales before increasing the amount to US$85bil.
For investors, the question becomes whether the enormous infrastructure build-out eventually produces enough revenue to support the capital being deployed.
“I am completely sceptical on being able to build out this capacity, and even if we do build it out, are the companies going to get the revenue that they need to justify the expense?” Mullaney tells Bloomberg. “I just can’t make the math work.”
Alphabet provides another example of how quickly investor expectations can shift.
The company is earlier hailed as an AI leader following the success of its Gemini chatbot and its internally developed AI data-centre chips. But its shares fall 13% from a May peak amid concerns that it is falling behind OpenAI and Anthropic, particularly in coding services.
Still, access to capital is not an immediate problem for the largest AI spenders. Microsoft, Amazon, Alphabet and Meta retain strong balance sheets and significant room to borrow.
The cost of doing so, however, is increasing.
Higher rates, supply-chain bottlenecks, political opposition and the challenge of monetising AI-related expenses are increasingly becoming part of the investment equation. Mullaney says these issues are more likely to matter in 2027 or 2028.
That does not mean the spending suddenly disappears.
AI-related investment continues to underpin a large part of the economy, while investors remain divided over whether the concerns surrounding the technology are getting ahead of the fundamentals.
“The fears around AI are tremendous, but we’re not seeing them come through at this point,” Bob Edwards, chief investment officer at Edwards Asset Management, tells Bloomberg.
“The businesses remain fabulous and we won’t be shaken out of good positions because of the question of what will happen.”
For now, that leaves the AI investment cycle facing several moving parts at once: the pace at which new models develop, the amount companies spend on infrastructure, the cost of borrowing, the ability to monetise those investments and the concentration of market gains and earnings among a relatively small group of companies.
The scale is already much larger than a single technology-sector trade. With AI investment accounting for a substantial share of US economic growth and AI-linked companies contributing heavily to the S&P 500’s gains and profit growth, changes in spending are increasingly reflected across markets.
The next phase therefore depends not only on how powerful AI becomes, but also on how the financial numbers surrounding that technology develop as spending moves from corporate cash piles towards debt and equity markets.
As the AI build-out continues, investors are increasingly watching the gap between what companies are spending today and the revenues and profits that are expected to come from that spending in the years ahead.