Wall Street’s $7 Trillion Reality Check: When Tech Dreams Crash into Hard Financial Limits

When someone finally says what everyone has been thinking, a certain kind of silence descends upon the room. That silence sounds a lot like what has been settling over discussions about artificial intelligence and the enormous amounts of money going into it, at least in financial circles. Don’t panic just yet. However, doubt is more subdued and possibly more unsettling.

In January of last year, Governor Tate Reeves stood in an 800,000-square-foot warehouse in Mississippi and announced what he described as the single largest investment in the history of the state: a $20 billion project by Elon Musk’s xAI, which promised nearly two gigawatts of computing power. It was the kind of announcement that was meant to leave you speechless, and it succeeded.

But a more difficult question kept coming up somewhere between the optimism of cutting ribbons and the Excel spreadsheets back on Wall Street. Specifically, who will foot the bill for all of this?

It’s hard to sit with the numbers. Together, Microsoft, Google, Meta, and Amazon spent about $335 billion on capital expenditures last year, and an additional $400 billion is anticipated for the upcoming year. Total earnings from those four titans related to AI? Somewhere between $15 and $20 billion.

When compared to the $1 trillion the company claims it wants to spend before the end of the decade, OpenAI, the company that may have started this whole frenzy, is on track to reach $20 billion in its own revenues. The numbers don’t look good.

It’s possible that this is just the way transformative technologies operate, requiring a significant initial investment before the benefits become apparent. Market historians and economists will swiftly and confidently point you in the direction of the late 1990s internet boom, the electrification of factory floors, and the railways of the 19th century. Prior to appearing profitable, each of those technologies appeared costly. However, each of those technologies eventually caused a crash that eliminated believers who had borrowed excessively, moved too quickly, or priced in a future that came later than anticipated, if at all.

In a recent article in the New York Times, Ryan Cummings, chief of staff at Stanford’s Institute for Economic Policy Research, put it simply: the evidence suggests that we are most likely in a bubble. Because markets have a tendency to penalize certainty, he is cautious about making a definitive statement. However, it is difficult to reject the argument he presents.

He claims that he uses AI for coding, so the productivity benefits are real. However, it is never clear whether these benefits actually occur. Whether they are even close to what the market has already priced in is the question. They aren’t at the moment.

The fact that the initial wave of AI spending was mostly financed by money that was already on the hyperscalers’ balance sheets is what really complicates this situation. Whatever you think of Google, Amazon, and Microsoft, they are incredibly successful companies. Speculation feels different from spending against current profits. However, that narrative is subtly shifting.

Because Oracle doesn’t have the same cash reserves as its bigger competitors, financing that partnership requires using credit, so Oracle’s huge deal with OpenAI marked something of a turning point. debt. At the periphery of the AI buildout, the same mechanism that turned a housing correction into a global crisis in 2008 is starting to emerge.

Co-director of the AI Now Institute Sarah West has been closely observing this change. Her analysis gives the impression that the infrastructure being constructed at the moment—the data centers, the power plants, the copper, water, and labor needed to keep them operating—is being justified by a future that the market as a whole has chosen to regard as certain. That’s a risky stance, according to history.

The investment firm GMO has spent decades researching what they refer to as “two-standard deviation bubbles,” which are instances in which asset prices deviate so much from their long-term real trend that a correction is ultimately mathematically inevitable. More than 300 of these episodes have been cataloged. All of those bubbles have burst and retreated to the pre-existing trend in major developed equity markets.

All of them. The American market in 1929. Once more in 1972. the 2000 dot-com bust. The 2006 housing market. The underlying technology’s quality is irrelevant to the pattern. In fact, railroads changed the world. The internet was as well. Devastating crashes were still produced by both.

The timing of ChatGPT’s arrival in late 2022, when the market was already declining from its peak during the pandemic, is almost poignant. It was abrupt, bright, and capable of making people look up again, much like a flare fired into a darkening sky.

The question that keeps serious investors up at night is whether it prolonged the cycle or just postponed the reckoning. Whether this particular chapter ends quietly, with a gradual bleed of valuations, or loudly, as financial manias usually do when the last optimist runs out of reasons, is still up in the air.

According to Brian Merchant’s article in Wired, there are indicators building up at the periphery, such as debt-financed data centers, revenue gaps, and productivity figures that are consistently referred to as “early,” “promising,” and “just beginning to show up.” When discussing internet advertising in 1999, those same terms were used. It’s difficult to ignore the similarity.

All of this does not imply that AI is useless. That is most likely untrue. The technology is truly amazing—possibly more so than anything from the previous century. However, impressive technology and appropriately priced stock are not the same thing, and Wall Street’s $7 trillion reality check may be more about whether AI operates quickly enough and on a large enough scale to justify the money that has already been spent on the belief that it will.