The AI Bubble

Top Stories of the Year: #5

The Hype Cycle for new technology, courtesy of Wikipedia.

I’m counting down the top stories of 2025 a little bit early this year, while I run a Thanksgiving Sale.

Rather than make you click on a special sale link, I’ve just lowered the subscription price 10% across the board for the next week, while I finish up my review of the year. Then subscriptions go to the regular price through the end of the year. In the new year, I will probably raise the subscription price a bit to keep up with inflation. So if you want to lock in the lower price now, consider changing from a monthly subscription to an annual subscription.

You can also give a gift subscription.

Or donate to support The Tracinski Letter.

And of course, buy my new book, Dictator From Day One, to get ahead of the Christmas rush.

---

The real top stories of 2025 are the ones I covered in my book, which tells the one big story of the year: Donald Trump’s attempt to concentrate in himself unchecked power over the US government and over all of American society.

I will be counting down the top five other stories, some of which I have not discussed in much depth because I was so busy covering that big story.

But to keep that wider context, each item in this countdown will lead off with a reminder of one of the five prongs of Trump’s assault on the American system, as recounted in my book.

---

The Emperor of Everything

Let’s start with that point about Trump trying to assert control over every part of American society. The fifth prong I cover in my book is Trump’s attempt to dominate “civil society.” Here’s how I put the issue:

“Civil society” is a term political scientists use to describe public spaces that are outside the realm of government. Civil society consists of all the ways individuals voluntarily associate with each another independent of politics. It includes civil associations, private charities, bowling leagues, fan clubs—every form of cooperation that voluntarily draws people together in a common cause or interest.

In Democracy in America, Alexis de Tocqueville described this as a crucial strength of American society.

Americans of all ages, all conditions, all minds constantly unite. Not only do they have commercial and industrial associations in which all take part, but they also have a thousand other kinds: religious, moral, grave, futile, very general and very particular, immense and very small; Americans use associations to give fêtes, to found seminaries, to build inns, to raise churches, to distribute books, to send missionaries to the antipodes; in this manner they create hospitals, prisons, schools. Finally, if it is a question of bringing to light a truth or developing a sentiment with the support of a great example, they associate. Everywhere that, at the head of a new undertaking, you see the government in France and a great lord in England, count on it that you will perceive an association in the United States.

If civil society takes the place of government and “great lords,” then these associations are what we need if we do not want to be dependent on government and great lords in every aspect of our lives. It also means that they are precisely what someone who aspires to be a “great lord” will try to draw under his control.

I discuss how this applies to the legal profession, the media, universities, and cultural institutions like the Kennedy Center for the Performing Arts in Washington, DC, which Trump has taken over and put under the control of his cronies and sycophants.

I have a few follow-ups on that last item. First, there is a vignette about the overtly politicized management of the organization, down to the old practice of disrupting shows by politically disfavored performers.

Yasmin Williams, an Alexandria-based and globally hailed guitarist, made headlines earlier this year for sparring by email with new Kennedy Center president Ric Grenell. She nonetheless honored her pre-Trump-takeover contract to perform at the Millennium Stage on September 18. All was going well until about five minutes before curtain, when theater staffers learned that 50 passes had been set aside for men purportedly intent on heckling, booing, and harassing Williams….

Those seats were reserved for the Log Cabin Republicans by Grenell’s office, multiple Kennedy Center employees said. A member of Grenell’s staff even distributed passes and welcomed the men as they arrived, a current staffer confirmed for Washingtonian….

In a newsletter that has since been widely circulated on social media, [Log Cabin Republicans of DC president Andrew] Manik outlined the group’s plan to disrupt Williams’s performance because she is a “liberal” and “a vocal opponent of President Trump.”

But you really know that the president is attempting to take over civil society when you find it necessary to keep up with American politics by reading Playbill and OperaWire. Both of those publications have been covering a recent story about rampant corruption in the management of the Kennedy Center. Here’s Playbill’s overview.

A series of contracts and receipts obtained by the [Senate Environment and Public Works Committee] showed the Kennedy Center providing venue rentals to allies at a deep discount; spending almost $40,000 on luxury hotel rooms, lavish meals, and entertainment for friends over a period of three months; and providing work contracts to friends that were seemingly unrelated to Kennedy Center business….

The EPW documents also showed deeply discounted venue rentals to conservative organizations, including a $19,820 discount for a NewsNation town hall and a $21,982.60 discount to the American Conservative Union Foundation—the latter contract included a note of “waived costs from OOP” (the Office of the President).

In his letter, Whitehouse also detailed that Grenell provided a former colleague a $15,000-per-month contract to perform “policy research and speechwriting,” noting it’s “unclear how this work relates to your role at the Kennedy Center, since the contract is devoid of any detail, nor have you delivered any substantive speeches or remarks that would justify these payments.” The Center also entered into a $10,833.33 per month contract with Jeff Halperin, husband of conservative pundit Kari Lake, for “social media capture/editing” services.

This is exactly what we would expect when cultural institutions are taken over, not just by a political faction, but by a highly corrupt faction.

Seriously, folks, buy my book, and recommend it to or buy it for others. It explains a lot about what is going on right now.

Now to #5 in my countdown of the top five other stories of 2025.

---

The AI Bubble

This is not so much a story of what has happened as what is (probably) about to happen.

One of the reasons the US economy looks a lot better than you might expect, given the impact of tariffs and tariff uncertainty, is because of a giant boom in artificial intelligence. The stocks that are holding up the market are known as the “Magnificent Seven”: “seven technology stocks that drove a large portion of the market’s returns in 2023 and 2024. The list includes Apple, Microsoft, Amazon, Alphabet (Google), Tesla, Nvidia, and Meta Platforms.”

But it’s not just those companies. It’s a wider expectation about the impact of artificial intelligence on the economy as a whole—an impact that seems unlikely to be fulfilled.

“A big part of the high stock prices is not a reflection of high future growth,” [Kenneth Rogoff] said. Rather, it is a sign that AI is expected to boost productivity and shrink employment. “The firms all think they’re going to shed a lot of labor, and that’s why the profits will be high,” he said….

The swoon-level stock market valuations of companies like Nvidia at $5 trillion seem to be predicated on the assumption that rapid growth will go on and on. Some of the players spending billions of dollars have yet to turn a profit.

Critics warn that a small group of tech companies, including Nvidia, are essentially buying and selling to each other in circular deals that inflate their actual value.

When I say that AI is looking like a bubble, I don’t mean to denigrate the real potential in artificial intelligence. But this is a natural cycle that tends to happen when a new technology leads to a “peak of inflated expectations” that has to be brought down to earth before the technology ends up finding its most productive uses.

In the case of artificial intelligence, this cycle started in the last five to ten years with the development of new forms of AI models that seem to follow “scaling laws.” What “scaling” means in this context is that if you put more resources into the technology—in the case of AI, that usually means more data and more computing power—you get noticeably better results. That is in contrast to a technology that reaches a limit or plateau, where a basic design problem limits its performance, and more resources do not necessarily produce better results.

The markets really love a technology that scales, because it means that for every dollar you put in, you are guaranteed to get a return back. But there are two problems. First, a technology can stop scaling, and you don’t really know if that will happen before or after you throw your next $1 trillion into it. Second, the technological results can keep getting better—but not necessarily in a way that leads to increased profit. Artificial intelligence presents both of these problems.

Right now, the markets are assuming that AI won’t hit any kind of wall or plateau and will just keep on scaling. And they are assuming that, if it does keep on scaling, every fantastic scenario about increased productivity will come true.

In the more enthusiastic Silicon Valley circles, there is a kind of millenarian fervor to this, a belief that soon we will not merely have artificial intelligence but artificial superintelligence, which will lead to an even faster geometric increase in productivity and a cure for cancer and on and on. Late last year, Open AI’s Sam Altman predicted, “it is possible that we will have superintelligence in a few thousand days” which would lead to “the discovery of all of physics.” All of it. Basically, we will create a kind of artificial god who will solve all our problems.

On the technological level, there is room for doubt. If part of the scaling required for AI is to provide it with more data to work from, the problem is that we may already be reaching our limits. One of my writing students a year ago at the Roots of Progress fellowship put up a very interesting piece about the “data wall” that AI now faces, because “current language models are trained on datasets fast approaching ‘all the text, ever.’” The more intriguing observation is that AI models are already being trained on far more data than a human will ever encounter in his lifetime.

Every day, an average person reads a few thousand words and hears perhaps 16 to 40 thousand. So a well-educated 40-year old might have encountered 5×10⁸ words in their lifetime. Recent language models are trained on upwards of 10¹³ words—200,000 times more. It’s not even close.

Imagine a fast reader who did nothing but read 300 words/minute for 80 years, never pausing to eat or sleep. They’d still see 1000 times fewer words than AIs do.

The fact that humans still far outperform AI models based on a fraction of the data indicates that just adding more data and computing power is unlikely to achieve results that are superior to human reasoning. AI may soon hit a big plateau after it wrings all the gains out of the world’s largest data sets.

There is an even bigger question about how to use AI profitably. There are some profitable uses for AI, especially in extracting information from large sets of data that would be difficult and time-consuming to analyze with humans alone. But so far, AI does not seem to be as broadly and intensively profitable as the markets are expecting—and not enough to fuel the vast growth of resources being poured into it.

To give you an idea of the kind of resources we’re talking about, let’s take a look at Google’s targets for the growth of AI.

During an all-hands meeting earlier this month, Google’s AI infrastructure head Amin Vahdat told employees that the company must double its serving capacity every six months to meet demand for artificial intelligence services, reports CNBC. The somments show a rare look at what Google executives are telling its own employees internally. Vahdat, a vice president at Google Cloud, presented slides to its employees showing the company needs to scale “the next 1000x in 4-5 years.”

While a thousandfold increase in compute capacity sounds ambitious by itself, Vahdat noted some key constraints: Google needs to be able to deliver this increase in capability, compute, and storage networking “for essentially the same cost and increasingly, the same power, the same energy level,” he told employees during the meeting. “It won’t be easy but through collaboration and co-design, we’re going to get there.”

It’s unclear how much of this “demand” Google mentioned represents organic user interest in AI capabilities versus the company integrating AI features into existing services like Search, Gmail, and Workspace.

To give you an idea of the underwhelming results from some of these AI ventures, consider Elon Musk’s attempt to create a competitor to Wikipedia—which turns out to just be Wikipedia, copied over directly, but then embellished with footnotes to Stormfront.

On a more comical note, the best sign of a bubble is the recent breathless announcement of a partnership between AI chipmaker Nvidia and, um, L’Oreal. The cosmetics company. One of the signs AI is a bubble is that people start to think you can just slap the AI label onto anything, and investors will get excited about it.

But what really makes this look like a bubble is its dependence on debt rather than on actual, present-day profits.

As for the profits, considering this Axios report.

Wall Street’s biggest fear was validated by a recent MIT study indicating that 95% of organizations studied get zero return on their AI investment.

Investors have put up with record AI spend from tech companies because they expect record returns, eventually. This study calls those returns into question, which could be an existential risk for a market that’s overly tied to the AI narrative.

MIT researchers studied 300 public AI initiatives to try and suss out the “no hype reality” of AI’s impact on business, Aditya Challapally, research contributor to project NANDA at MIT, tells Axios.

95% of organizations found zero return despite enterprise investment of $30 billion to $40 billion into GenAI, the study says….

“My fear is that at some point people wake up and say, alright, AI is great, but maybe all this money is not actually being spent all that wisely,” says Steve Sosnick, chief strategist at Interactive Brokers.

That looks bad when you combine it with the debt that is being used to finance this boom, which will be impossible to pay back if the future profits never materialize. See an overview of that issue.

Adding to worries, critics say, some of the deals that OpenAI has made with chipmakers, cloud computing companies, and others are oddly circular. OpenAI is set to receive billions from tech companies but also sends billions back to the same companies to pay for computing power and other services.

Some financial analysts worry that these deals make the market look stronger than it really is. Ultimately, the health of the market will depend on whether companies like OpenAI can turn a profit before debt overwhelms them….

Nvidia has also made some deals that have raised questions about whether the company is paying itself. It announced that it would invest $100 billion in OpenAI. The start-up receives that money as it buys or leases Nvidia’s chips.

On Tuesday, Nvidia announced a similar $10 billion deal with Anthropic, which will buy $30 billion in A.I. computing backed by Nvidia chips. That money will actually go to buy computing power from Microsoft, which also invested $5 billion in Anthropic.

Goldman Sachs has estimated that Nvidia will make 15 percent of its sales next year from what critics also call circular deals.

Here again you see some of the millenarian frenzy, the idea that we just need to get to the other side of a world-changing event, and then everything will be different.

Many companies justify their spending because they’re not just building a product, they’re creating something that will change the world: artificial general intelligence, or AGI, a machine that can do anything the human brain can do. The rub is that none of them quite know how to do it….

“It’s a bet on A.G.I. or bust,” [University of Virginia economist Anton] Korinek said.

The saying I’ve heard is that companies are going to spend trillions to develop artificial superintelligence in the hope that the superintelligence will be smart enough to figure out how to make money from it. It’s the kind of thing you say half-jokingly and half-seriously.

The absolute worst sign is that some of these companies are already anticipating the need for government support to bail them out when the bubble bursts.

OpenAI is seeking U.S. government support to help finance what could become one of the largest infrastructure buildouts in corporate history — exceeding $1 trillion. Speaking at a Wall Street Journal business conference, Chief Financial Officer Sarah Friar said the company is exploring federal loan guarantees to attract broader funding for its AI computing expansion, describing a potential “ecosystem of banks, private equity, maybe even governmental” participants.

Friar said government backing would significantly lower borrowing costs and broaden OpenAI’s access to credit markets, since federal guarantees would protect lenders from losses if the company defaulted. The proposal is highly unusual for a Silicon Valley technology firm, effectively positioning OpenAI alongside sectors such as energy and infrastructure that traditionally rely on state support.

Maybe I and the other skeptics are wrong. I’ll link you to a case that this isn’t a bubble, though I can’t help noticing that it depends on a lot of reasoning about how This Time Is Different.

The problem is that the mental models used to evaluate AI are outdated. This is not a bubble in any conventional sense. It only resembles one if we force AI into the logic of traditional software.

But remember that bubbles are a short-term phenomenon. They have happened before, and they will happen again, and the main question to ask is what will get built during the height of the bubble. All that money will help push technology forward and build up a big infrastructure, even if some of the companies that build it go bust in the process. That’s one of the things that happened in the dot-com bubble of the 1990s. A bunch of telecom companies crashed, but only after they helped build the Information Superhighway. So look to see what part of the AI infrastructure will endure and eventually be used productively.

Over the long term, I remain skeptical of either the utopian or the apocalyptic scenarios about AI. It will eventually be a productive tool for some important things. It will neither solve all our problem nor end the world.

In the short term, it’s worth preparing for the possibility that in the next year, the sector that is currently propping up the American economy could start dragging it down, which would have some big implications for our politics.

---

Remember to take advantage of our Thanksgiving Sale. Subscribe now at 10% Off.

Give a gift subscription.

Or donate to support The Tracinski Letter.

And last but not least, buy my book.