Gone are the days of capital-light and cash-flow-rich technology businesses such as internet search, e-commerce and social media. The game now is artificial intelligence, and you had better have hard assets and deep pockets to play, the kind of money few companies have laying around.
Nvidia Corp. and the hyperscalers, a group that includes Microsoft Corp., Amazon.com Inc., Alphabet Inc., Meta Platforms Inc. and Oracle Corp., have already committed trillions collectively to AI-related projects, with as much as $4.2 trillion through debt and contractual obligations. A lot of people have started to ask whether these investments will pay off, but that assumes the companies can afford such huge sums to begin with — and to keep feeding AI’s seemingly insatiable capital demands.
See more: Hidden Debt: Is Our Hyperscaler Thesis Wrong?
It’s far from certain given the amounts involved. We don’t talk about that enough, presumably because these companies are flush with profits and navigating the maze of AI spending is thorny. It doesn’t help that many of their AI-related commitments don’t show up in financial statements but are buried unhelpfully in jargony footnotes. Getting a handle on this spending is important because even the most formidable companies can overcommit, and it looks as if some of the hyperscalers already have about as much leverage as they can handle.
That isn’t necessarily a problem for the stock market in the long run. The gains from AI will be broadly distributed as they were with the internet, lifting the market over time regardless of individual winners and losers. But given the AI giants’ prominence in broad market indexes and investors’ portfolios, and the boost AI spending is giving the US economy, everyone will feel it if they stumble.
These companies are in a tough spot. They can’t afford to sit out the AI arms race because they’ll be left behind. They can’t all win either, which means, at best, some of them will be poorer for the effort.
The unlucky ones may run into more serious financial trouble if AI displaces their core business and profits shrink relative to mounting debts, inviting the indignity of selling AI-related assets to competitors at potentially steep discounts. And it may not take long. Google Zero, the idea that traffic to websites from internet searches could disappear as people rely on large language models for answers, seems to be well underway.
The sprint to dominance also risks overinvestment that results in more supply than near-term demand for AI. We’ve seen this before, and it didn’t end well. As the internet got going in the late 1990s, telecommunications companies poured hundreds of billions into fiber-optic cable and wireless networks in a race to capture surging internet usage. When demand for bandwidth temporarily dried up in the early 2000s, telecom companies struggled to service their debt. Several went out of business, notably WorldCom and Global Crossing.
The infrastructure they built was bought by later entrants at pennies on the dollar and helped power the internet, just as today’s investment spree will fuel AI. It’s not hard to imagine a similar, transitory decline in demand that dries up AI-related revenue.
Much of that revenue comes from frontier AI labs such as Anthropic PBC and OpenAI paying for computing power and other costs associated with building LLMs. These are quickly becoming commoditized and threatened by cheaper, open-weight Chinese competitors. If LLMs get squeezed, AI investment may no longer be monetizable until the next generation of applications, presumably robotics, becomes widely adopted.
The AI giants have a lot of cushion; they will collectively generate $1 trillion in operating earnings this year. It may not be enough. They reported collective liabilities of $1.6 trillion as of June, nearly double the number two years ago. I counted an additional $2.6 trillion of potential off-balance-sheet liabilities that include leases, operating agreements and guarantees, ballooning their total collective obligations.
It’s not an iron clad number. Some of those commitments will be renegotiated, deferred or cancelled, as commonly happens in business. Also, not all the companies are leveraged to the same degree.
One way to size them up is by comparing their total liabilities to annual operating income, which gives investors a sense of how long it would take to cover their current commitments. Nvidia could do it in about a year and a half. Microsoft, Amazon and Alphabet would take closer to four to six years. Meta and Oracle would require a decade or more, mainly because their operating income is a fraction of the other four.
The bond market has its eye on the two outliers. Meta and Oracle’s intermediate-term bonds trade at a premium to Treasuries that is nearly a percentage point higher than the Bloomberg corporate benchmark with a comparable rating and maturity. That’s well higher than the spread for the bonds of the other four relative to the benchmark, signaling that unlike Meta and Oracle, they may have more room for investment.
How much more depends on the degree of leverage investors are willing to tolerate. Meta and Oracle have probably reached their limit. Assuming a liabilities-to-operating earnings ratio of 10 times, which roughly matches Meta and Oracle, Microsoft and Alphabet could conceivably spend another $2 trillion each and Amazon closer to $1 trillion. Nvidia could possibly muster another $3 trillion in spending.
By that count, I expect trillions more in AI-related investment, assuming the companies maintain their current level of profitability. But if they or AI hit a slump, some of them may feel serious financial strain.
Whatever happens, AI will march on. LLMs are just the first and least sophisticated application. Robots are already driving taxis, and soon they will take over domestic work and manual labor, fight wars and perform surgery. Those advances will more than pay for the flood of AI investment in the long run. The only question is which companies will be around to profit from it.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Nir Kaissar
