AI Agonistes
Time for the industry to put up or shut up
My new novel, Colossus, arrives at the end of April. Library Journal calls it a “challenging portrait of a thoroughly modern man.” I agree. Please preorder it now. You won’t regret it.
I confess, at the outset, I won’t spend as much time on this essay as I’d like. I am under pressure to finish a nonfiction book by the spring, and that eats up a lot of the time I have for more exploratory and analytical writing. Such is life; I don’t complain, and the project is exciting. What I will offer, in lieu of the many thousands of words I’d prefer to disgorge, are a few thoughts about artificial intelligence.
If you’d like to categorize me, I am an AI “skeptic.” This does not mean I am a luddite or not a futurist. I want to be awed by human advancement and achievement. I want to see diseases eradicated, new vertiginous cities arise, and planets settled, if the latter is not plausible. (Sadly, everything I’ve read about Mars tells me humans will never, ever live there.) I very much want to be excited by tomorrow. By disposition, I am an optimist; I believe the United States is going to survive Donald Trump, and may well thrive into the 2030s, 2040s, and 2050s. I believe humanity is going to exist in five hundred or even a thousand years from now. I believe we have the technology to adapt to climate disaster and, when the dust settles in the deep future, we probably will.
That all being said, I am not impressed by AI. This does not mean I do not take the threat of AI seriously. The tech is quite strong, already, when it comes to automating away lower-level white collar work. I believe many workers, those at the beginning of their careers especially, are going to suffer. Corporations will, one way or another, hire less. I don’t see much upside to this. It strikes me as a version of what happened to heavy industry in the second half of the twentieth century, when globalization and automation crushed cities like Detroit, Cleveland, Buffalo, and Gary, the steel colossus of Indiana. There was upside to this immiseration—just not for anyone who lived there. Consumer products became cheaper, factories boomed in poorer countries, and the American economy, on the balance, grew. The deindustrialization of America was one of the great tragedies of the last century, but I understand why the professional managerial class and a certain kind of pundit might defend the carnage.
With AI, such a bounty does not seem to exist. Instead, there will simply be more unemployment, as companies trim headcounts because they can automate functions away or they anticipate, in the near-term, new developments that will make the human worker obsolete. Were AI promising Edward Bellamy-style utopic socialism—enormous amounts of leisure time, a social safety net to ensure no American ever wants for housing and healthcare—I might be more intrigued by the future ahead. But while I can’t, truly, take all the prognostications of an imminent AGI or “singularity” arrival all that seriously, I do think there are going to be employment challenges that the federal government, sadly, won’t be up to solving. The current government, at least, wouldn’t dream of a UBI program or a jobs guarantee. Americans will be left to fend for themselves, as they usually are. The politics of it all might grow ugly.
I couch all of this only with the understanding that AI, as of now, appears to have nothing approaching a plausible business model. OpenAI, for example, has said it will burn through $115 billion by 2029. Anthropic thinks it will be profitable by 2028, but still loses billions of dollars. Theoretically, there might be enough businesses and individuals who pay hundreds of dollars a month for AI products to make these companies solvent without infinite venture capital, but the real economic riddle is the data center. How do you keep affording to build them? How do you keep getting permission to build them? Amazon alone wants to spend $200 billion this year on data centers, along with satellites and other tech related to AI. Alphabet has said it would spend as much as $185 billion in 2026, and Meta said last week that its capital expenses, in large part to support AI, could hit $135 billion. Investors, finally, seem to be starting to ask the more difficult questions. When, exactly, will this all pay off? What can profitability possibly look like with spending that’s this titanic, unlike any we’ve witnessed in history? Data centers are still opening, but it’s no longer a guarantee that a tech behemoth can roll up into a town, make a few fast promises, and win political support for new construction. More and more states are pushing for moratoriums on data centers. What should worry tech leaders and investors is the nature of this opposition: in a deeply polarized age, it’s strikingly bipartisan. Local Republicans who might care less about environmental concerns than Democrats but they hate how data centers can drive up electricity bills and suck up water. Rural opposition has grown in states like Texas. Unlike the factories of old, they simply do not employ many people. In the abstract, Americans are fine enough with the existence of data centers. But like with many other issues, opinion shifts when one shows up nearby. There’s little evidence to suggest Americans will warm to data center expansion in the coming years unless they show they will not drive up utility costs and can employ a sufficient number of locals. There was plenty of tolerance for coal plants and textile mills when they lifted up enough townspeople into the middle-class.
Is this a bubble that’s going to spectacularly burst? I do not know. AI hopes it can follow the example of Amazon, which lost gobs of money for years—nowhere near as much as the AI giants, however—before turning handsome profits. That is one path forward, if there’s little to undergird it other than blind hope. Another path that is, in my view, more likely is the one OpenAI spoke out loud last year—one that, for a fleeting moment, terrified investors. Sarah Friar, OpenAI’s CFO, said several months ago that she thought the U.S. government should “backstop” her company’s aggressive AI investments. “The backstop, the guarantee, that allows the financing to happen, that can really drop the cost of the financing but also increase the loan-to-value, so the amount of debt that you can take on top of an equity portion,” she said. In short: what could amount to a federal government bailout for the debt AI companies take on to make their investments. Once it became clear this was effectively an admission that OpenAI might never actually be profitable, Friar and Altman scrambled to clarify that, in fact, they were not seeking a bailout. To some extent, the damage was done, if most AI evangelists pretend Friar never made the remarks in the first place. If, indeed, OpenAI cannot ultimately pay for its $1.4 trillion in infrastructure commitments, the entire industry may implode. Altman has lucrative deals with Nvidia, Amazon, and many others. “If we screw up and can’t fix it, we should fail, and other companies will continue on doing good work and servicing customers,” Altman said later. “That’s how capitalism works and the ecosystem and economy would be fine.”
That’s how it usually works. But my sense is, with AI, we may be entering new terrain, where the federal government might eventually view the technology as too vital in its cold war against China. Like the space race, with far less romance, the government might find itself fully in the AI business, to the wariness or even fury of the American taxpayer. As we learned in 2008, socialism is alive and well in America, but it exists mostly for the extraordinarily wealthy. Bailout packages can always be available for the richest corporations in need. The spending on AI is, collectively, unfathomably profligate, and some of the cash appears to be just swishing back and forth between Nvidia, the chipmaker, and companies like OpenAI. Meta, Alphabet, Amazon, and Microsoft all have other core businesses to make the cash they lose on AI palatable, but Anthropic and OpenAI do not. If there’s no road to profitability and the U.S. government doesn’t arrive with a backstop, will both these companies crumble? Will Amazon, Alphabet, and Meta be content to keep burning cash on their AI investments? We do not know. The rubber, as they say, might hit the road sooner than these companies think.
There are romantic arguments to be made against AI—how it largely debases the arts and most creative endeavors—and I will be glad to make those in the future. For now, though, I prefer to remain in the realm of the dollar. I find that the philistines who most slavishly boost AI are most reachable this way. Or, at least, they are more likely to respond and engage, and bring their own math to the table. The numbers will add up, they yowl. They could very well understand what I don’t—that, in fact, there are endless streams of money just waiting to be had, that don’t rely on a gullible investor class or a government spooked by China. I am open to being proven wrong. In the meantime, I will wait and see what glories or destruction might be wrought. What makes AI such a challenge is that it has placed the American worker in an obvious bind. Either it is a success and we see the mass unemployment that those like Andrew Yang warn against—I do not dispute this is one possibility—or it fizzles, and we suffer the bursting of a bubble that cracks the economy like the dot com and housing crashes. The pop of an AI bubble, as satisfying as it might be to someone like me, is not something I particularly welcome because it’s going to take a great deal of the stock market with it. Recessions don’t help anyone. And neither do frenzies like these, with numbers that go up and up and up and defy base logic.
We race forward, into the dark. Perhaps that’s how it always was. What we can never approximate is the sheer confusion of history. How hard it might have been, for example, in 1942 to truly imagine what 1952 or 1962 looked like. We could be on the cusp of greatness, disaster, or something far more banal. The middling outcomes excite AI evangelists the least. They want you to believe that AI is a development on the order of electricity, indoor plumbing, and penicillin. They long for this great leap forward—to have the grand feeling of living at a pivot point of history, within the eye of the gyre—because the alternative is a bit more chilling: society will grind on as it is, with most of the frontiers settled. Apple has not innovated since the iPhone. Google still hasn’t blown open as many minds as it did when it introduced their search bar. Meta is coasting off the fumes of Facebook and the two savvy acquisitions it made more than a decade ago, WhatsApp and Instagram. In some sense, AI must “happen” because the oligarchs are, for the most part, bereft of other ideas. Elon Musk can dream of Mars all he wants, but mankind can’t live there. The surface is too inhospitable and will remain that way. We can settle for little moon colonies. Perhaps AI will start to cure a bunch of incurable diseases; that would be welcome, and I would never sniff at a medical breakthrough. We will never be immortal, but living a bit longer, in peace, would be nice. If Claude has any ideas about that, may Claude live long and prosper. If not, I’ve got better things to do with my day.



What a great, rollicking read. Thank you.
It will make software engineering dramatically faster and less expensive. If you are in business of building or buying software, you have learned this in the last few weeks. That’s sufficient for a business model.