Wall Street Is Getting Very Nervous About AI
· Jacobin
For an industry that we’re told is central to our collective future, the artificial intelligence (AI) industry increasingly seems to be built on a combination of artificially inflated hopes and bad accounting.The industry’s critics have been raising concerns about its finances for a while: the fact that most of its flagship firms are grossly unprofitable, that they’ve had to jack up their prices to try break even (leading their business customers to pump the brakes on using their products), or that many of the eye-popping investment announcements that are meant to show demand for AI is strong have actually been circular financing — meaning different AI firms cycling money to and from each other.But none of it has seemed to slow the industry’s boom or the relentless hype machine behind it. Until now, anyway. It used to be that you mostly heard concerns about the industry’s underlying economics in newspapers, blogs, social media posts, and podcasts, and that those concerns went completely unreflected in the markets. But these concerns seem now to have finally bled over into investors’ thinking.Money WalksThe price of bonds issued for the borrowing that undergirds the build-out of data centers has sharply dropped in recent months, as has the bond price for “neocloud companies,” middleman firms that provide the computing power for AI firms who may not have their own data centers. The largest of these neoclouds, CoreWeave, recently saw its bonds enter distressed territory.What does that mean? In short, that investors are suddenly abandoning the AI-related debt they’d happily financed until just a few months ago when it seemed like a great investment, because they’re starting to suspect these firms are not actually going to end up being able to pay it off. Nearly half of the bonds raised this year to finance data center projects by some of the biggest tech firms are now rated “junk.”At the same time, the market for credit default swaps (CDSs) for AI is booming, with the value traded of CDSs linked to major AI firms having more than quadrupled between July and September. Made famous by their role in the 2008 financial crisis, CDSs are basically insurance against future debt defaults, another sign of investor doubt that AI industry firms are actually good for the money they’ve borrowed to fund their build-out.To put it plainly: the people who move money around for a living are starting to have serious doubts about how sustainable the AI boom actually is, and we know they are not because they’re just saying so, but because they’re making financial decisions as if they are.What’s changed to bring this shift about? A number of things, including the fact that, some voices in the financial world now are saying so.The AI industry’s current unprofitability was acceptable as long as it seemed plausible that, at some unspecified future date, this technology would actually start making money. But a spate of recent news and analysis has underscored the fantastical math behind AI firms’ accounting.According to one highly cited Brookings Institution paper, the colossal amounts of money already being plowed into the AI build-out mean that US consumers and businessses will need to spend enough on the technology by 2032 to generate $3.5 trillion worth of revenue, or just shy of 9 percent of the country’s GDP, to make good on it. There is quite literally no precedent for this in American history — the railroad boom involved capital spending more than an entire percentage point of US GDP smaller than the AI build-out. According to the Wall Street Journal, that would mean the nation spending around the same amount every year on AI as it currently does for food, or seven times the amount it spends on the internet, phones, and other communications combined.Needless to say, this looks unlikely.Last month, the chief economist of asset management firm Apollo pointed out the wide gulf between the huge future operating cash flow that analysts are expecting for the tech sector specifically, and the much smaller future operating cash flow analysts are expecting for companies in other sectors. In other words, the businesses that are in theory meant to be the tech world’s customers are not even seriously expected to have the money on hand that tech firms are boasting they will extract from them.Someone’s math here is off, and it’s reasonable to assume that it’s the AI firms’. Years of overoptimistic revenue projections and consistent money-losing from the sector were dramatically illustrated earlier this month when one prominent firm, Character.ai, filed for bankruptcy. In its filing, the company revealed that it had been losing around $8–9 million per month over the course of this year — which it said was an improvement from previous years — and that its subscription revenue had only ever been a fraction of the amount it was spending on the computing needed to run its AI.Risks AboundThese fears were heightened by some bad news around several high-profile AI firm initial public offerings (IPOs), when a privately held company starts selling its shares to the general public for the first time.First, OpenAI, the company behind ChatGPT, delayed its IPO over internal worries that it might end up a flop, because retail investors — basically, you and I if we got into the stock market — may not actually end up as eager to buy its shares as they hoped. Among the reasons why were general market uncertainty caused by the Iran war, growing skepticism toward AI, a spate of damaging headlines about the technology’s safety issues, and a stagnated user base for ChatGPT. The company has now had to pare back its valuation for its latest round of funding due to a palpable drop in enthusiasm from investors.Meanwhile, Claude’s maker Anthropic has made headlines for a prospectus — the document put out by a firm going public that explains the business and lays out its financial reality for possible investors — that has seemed to prove true every worry just laid out about AI’s business model. The latest version of the document, obtained by the Wall Street Journal this week, showed the firm posted a more than $80 billion loss this year, while its spending on computing and infrastructure is exponentially ballooning, its demand for computing capacity outstrips what is available, and its list of possible risk factors are longer than the company description.One of those risk factors has become a major general concern for AI investors: the growing public hatred of data centers and the push to restrict or even outright stop their continuing construction. Popular distaste for data centers cuts across party and ideological lines and has become a central issue in the upcoming midterm elections, making it more likely a government clampdown on them intensifies.Just in the first half of this year, environmental activists and local residents have blocked more than a hundred data centers worth roughly $198 billion in capital. That included a major project of Oracle’s in New Mexico that pushed the company into declaring force majeure — meaning, tear up a contract because of an unforeseen act of God — after having already spent millions.This is yet another, and an especially big, reason for investors in AI to worry they’re not going to get their money back. For AI companies to have any hope of meeting their already unrealistic revenue goals, data centers have to be built somewhere; but voters across the country are vehemently opposed to having them built in their neighborhoods, because they find them noisy, intrusive, harmful to the environment, and have a habit of driving up their electricity bills. So they’re organizing to have them restricted or outright banned, which is what they have done in more than five hundred counties and municipalities. Something will have to give here.That’s not to mention investor fears that they could be left holding the bag on data center debt. The Brookings study mentioned earlier estimates the AI data center build-out will cost a total of $10.3 trillion by 2032; market research firm Gartner estimates that AI data center owners will have to generate a staggering $8.5 trillion in revenue through 2030 to cover the build-out while making a decent return. These are enormous sums behind something that is being financed largely by massive and rapidly growing levels of borrowing.On top of this is the fact that the chips AI firms rely on, and which they spend billions of dollars on, are becoming obsolete quicker and quicker — much more rapidly than the lifespan many of these companies initially budgeted for. It’s another case of wishful accounting by the industry, and another reason to worry for investors, since borrowing by AI firms is often secured on the back of chips that will lose their value before the lifetime of the loan is up.So it’s no wonder that investors are now souring on AI and hedging their bets in anticipation of a bubble bursting. That could come sooner than we think: Oracle, considered a “bellwether” for the wider AI credit market, has seen its bonds downgraded to just above junk status. One economist believes that if its bonds ever do hit that nadir, it would be a “sign of contagion” for the sector more widely.Beyond AIThe stakes here are high. By some estimates, AI is now a leading pillar of the economy, making a bigger contribution to GDP growth than the dot-com bubble did in 2000 before it burst, according to the Federal Reserve Bank of St Louis. So any AI slowdown would likely affect the wider economy, particularly the construction industry, where it’s driving a blue-collar job boom.But maybe more important are the ripple effects it could have on the financial system. The Bank of England and the International Monetary Fund have both singled out the shoddy economics behind the AI boom as a potential threat to global financial stability.Jason Thomas, the head of investment firm Carlyle Group’s global research and investment strategy, has warned repeatedly that he sees similarities between AI project financing and the subprime mortgage crisis. Lenders right now are not giving out money based on the economics of a project itself, he said, but on the ability of the firm sponsoring the project to swallow the liabilities if need be — similar to what they were doing in the years leading up to the 2008 crash. He warned that, given this, investors have far too much exposure to the industry than they should to breathe easy.What’s notable about this is that Thomas isn’t just an analyst, but has skin in the game: the firm he works for, Carlyle, is heavily invested in the AI sector.Tied up in this risk is the fact that many AI firms are building data centers without adequate insurance, since the projects are simply too expensive, complex, and risky for many insurers to take on, leaving the companies building data centers on the hook for what could be billions of dollars of losses if something derails their construction. Also tied up in it is the $3 trillion private credit industry, or what used to be called “shadow banking”: the private equity firms, asset managers, and other nonbank entities whose systemic risks have long worried financial analysts, since they act like banks but without any of the accompanying regulation, oversight, and government backstops.Do Not ForgetThis doesn’t mean that a 2008-style crisis is necessarily in the cards. There is an ongoing debate about the level of risk AI poses to wider financial stability, and there are some experts who believe AI’s contribution to the US economy is overstated. We won’t know exactly what the end of the AI boom will look like or how badly it will affect the rest of us until such a thing happens.But it does seem implausible at this point to believe it won’t mean at least some significant amount of pain for US workers and households more generally.What we can safely say is that the math behind both AI firms’ business model and their future fiscal projections simply don’t add up, and that we can be sure of this because the people who have actual money on the line — investors, bond traders, insurers — are acting like it. Despite the supposed rationality of markets and the private sector, the AI boom is looking more and more like yet another historical episode where herd thinking and hype is driving the high-risk decisions of private capital.Beyond everything else, it’s a colossal waste of money. The gargantuan sums we’re talking about are comparable to what we should have started spending years ago on the renewable energy transition in order to prevent a climate catastrophe — a catastrophe that the AI build-out, perversely, is making worse.Instead, because this money is concentrated in private hands, these resources are being plowed into an industry that doesn’t actually make money and seems to be built on wishful thinking. We see a similar thing happening in China, where, because these decisions are made similarly undemocratically, vast sums are also thrown into one big bet on AI while the rest of the economy stagnates.If this all does fall apart and the ripple effects of that collapse are as bad as some predict, it’s more than likely that taxpayers will be asked to foot the bill for another bailout. If and when that time comes, remember that there were umpteen warning signs and that it all could have been avoided.This article originally appeared in Jacobin, a democratic socialist magazine publishing long-form essays and analysis on politics, economics, and culture. Subscribe to the print edition for $20 a year.
Visit moryak.biz for more information.