(Bloomberg Opinion) — The capital spending boom related to generative artificial intelligence is raising many questions about whether it is sustainable. Coming up with definitive answers to that is a fool's errand. Quantifying the size of the increase in capital investment appears to be a more productive endeavor – and one that could provide some clues as to its sustainability.
Even that is easier said than done. After spending the better part of three days creating chart after chart after chart (of which I share a small sample here), I now know that there are many different ways to quantify the AI capex tree that lend themselves to many different stories. My overall feeling after this exercise is that while AI spending has reached the point of great economic significance, it has yet to become the force that technology and telecom were in the late 1990s. Yes, it's the driving force behind much of the current economic growth in the US, and if it were to suddenly end there would be unpleasant consequences, but right now there's nothing particularly alarming. However, at a few major tech companies, the spend is like nothing they've ever tried. Either it pays off or there are some very challenging years ahead.
Here's an overhead view of the latest gross domestic product data released late last month by the U.S. Bureau of Economic Analysis. Private domestic investment in information processing equipment (split here between computers and peripherals and the rest) and software reached 4.4% of GDP in the second quarter, just below the peak of 4.6% in the last quarter of 2000. If the pace of increase over the past two quarters were to continue, that would be over by the end of this year – but that would not be out of step with the long-term trend of rising technology spending.
This includes a lot of spending on things not directly related to AI, while leaving out some AI-related research and development spending, which I left out because it includes R&D in pharmaceuticals and other industries and its share of GDP has been flat over the past five years. In July, Jens Nordvig joined from Exante Data Inc. with a bottom-up estimate, extrapolated from chipmaker Nvidia Corp's revenues, of $387 billion in capital expenditures on AI data centers alone in 2025. That would be about 1.3% of GDP, up from 0.8% in 2024 and 0.3% in 2023. (Nordvig's estimates of AI's impact on GDP growth is greater than that. (due to estimated multiplier effects, and current consensus forecasts of Nvidia's revenues for the rest of the year would put the GDP share of AI data centers closer to 1.5%, but for simplicity I'll stick with the original numbers.)
That 1% increase in GDP since 2023 is much larger than the 0.4% increase in GDP in information technology investment shown in the chart above. It could be that AI spending cannibalizes other technology spending, but the main explanation for the disparity appears to be that the BEA's domestic investment figures exclude spending on imported technology – and that net imports of information technology products are almost half a percent of GDP higher than they were two years ago. In national income and product accounting, net imports are subtracted from GDP, so I can't simply combine the two graphs, but the import figures do bring the spending increase close to Nordvig's estimates.
So does 1.3% of GDP go to AI investments a lot? In an article based in part on Nordvig's work, veteran technology analyst Paul Kedrosky compared his own estimate of 1.2% of GDP for current AI investments with 1% of GDP for telecom investments at the height of that boom in 2020, and 6% of GDP for railroads in the late 1800s. “It is not yet clear whether we are at the peak yet or not,” he wrote, “but… we are above it.”
Maybe above that, but no records broken yet. As the first graph above shows, there was more going on during the technology investment boom of the 1990s than just telecom (which largely falls under 'other information processing equipment' in the graph). Non-telco companies loaded up on computers and software, and the increase in total technology-related capital expenditure far exceeded what we've experienced with AI – which of course could change if AI spending continues to grow.
As for the railroad investment boom of the late 19th century, AI investments still have a long way to go to match the economic impact of that episode in American history. I compiled this chart from economist Melville J. Ulmer's 1960 book Capital in Transportation, Communications, and Public Utilities: Its Formation and Financing and the GDP estimates maintained by MeasurementWorth.
Railroad investment averaged 2.4% of GDP in the 1870s and 1880s, but fluctuated wildly, approaching 6% at the start of the period and falling to 0.3% during the Long Depression that followed the 1893 financial crisis – before a long era of more stable but lower investment relative to GDP began.
Overshoot and correction are part of the investment boom, which historically benefited society, even if not always those who invested. We still use fiber optic cables laid in the late 1990s and early 2000s, and rely on rail infrastructure dating back to 1873. While you can see many potential economic benefits from investing in AI, it's hard to imagine that Nvidia AI chipsets from the year 2025 will be of much use a century or even a decade from now.
Companies pouring money into AI data centers are funneling it into rapidly depreciating assets, and given that a handful of companies are doing most of the work, you can't help but wonder what they're getting into. Technology companies making massive investments in AI occupy the top five of the US capital expenditure rankings. None of the five were in the top 10 a decade ago (Google parent company Alphabet Inc. was at No. 12).
As of the second quarter, capital expenditures at these five companies amounted to $313 billion annually, representing about 1% of US GDP, although not all spending is in the US. That's more than double their 2023 spending.
Another way to measure these expenses is as a percentage of sales. For Google parent Alphabet and Facebook parent Meta Platforms Inc. current ratios are high, but not in line with their expenditure in recent decades. For Microsoft Corp., Oracle Corp. and Amazon.com Inc., where spending can be traced back to the 1990s, the current investment boom is truly something new. Amazon's capital expenditures-to-sales ratio has yet to peak at 17.5% near the end of the dot-com bubble in the fourth quarter of 1999, but it was a small, money-losing company then and is on track to surpass that record soon. Microsoft and Oracle, which were already large, profitable companies in the 1990s, were not spending anywhere near today's rates at the time.
In the 1990s, Oracle and Microsoft sold software and Amazon sold books. Now they offer software as a service and hosting for other SaaS providers. That means they have major infrastructure commitments that they didn't have before, so it's not surprising that their capital expenditures have increased. But the 46.5% of Oracle's revenue going to capital expenditures over the past four quarters is a larger share than even infamous telecom high-flyer (and subsequent precipitous decline) WorldCom Inc. ever hit in the 1990s.
It's a huge bet – and big bets don't always pay off. The wild recent swings in Oracle's stock price, which rose nearly 50% in early September and has fallen 15% since then, are indicative of the stakes.
More from Bloomberg's opinion:
This column reflects the personal opinions of the author and does not necessarily reflect the views of the editorial staff or Bloomberg LP and its owners.
Justin Fox is a Bloomberg Opinion columnist who covers business, economics and other topics related to charts. He is the former editor-in-chief of the Harvard Business Review and author of “The Myth of the Rational Market.”
More stories like this are available at bloomberg.com/opinion
