The AI Stock Market Crash Isn't an Emergency. Here's What Is.

Is the bubble bursting? Has China won? Take a step back. We're worried about the wrong thing. Plus: HillmanTok University is the best thing on the Internet.

The stocks of the big AI companies cratered this morning, and it really screwed up my day.

I had a loose and experimental meditation all written up and ready for you about the importance of anticipating the unanticipated, but I had to shelve those hippie ramblings once news broke that Wall Street is punishing American AI giants like Meta and Microsoft.

Why are those companies suddenly worth hundreds of billions of dollars less? Well, Chinese companies just debuted new reasoning and language models like DeepSeek-V3 and Kimi.ai that rival the capabilities of American ones, at a fraction of the cost and a fraction of the computing power, scrambling all expectations about what it takes to make effective AI systems. But I’m here to try to keep you from falling into the jingoistic tech panic that is about to fuel the political and business landscape for the next few months or longer, and to point you at some stuff that matters more.

The AI industry is faith-based. Typified by the venture capitalist (and recently converted Trump supporter) Marc Andreesen and his techno-utopian manifesto, its adherents profess a belief that through technology like AI “we can advance to a far superior way of living, and of being.” What exactly that superior future consists of, and how we get there through for-profit companies, isn’t clear. Instead, the industry seems to be saying, trust the vibes.

Those vibes are full of contradictions. OpenAI founder Sam Altman’s public remarks on the topic usually circle back on themselves, and then return to themes of faith. He told an audience at MIT last year that “AI is going to eliminate a lot of current jobs, and there will be classes of jobs that totally go away. AI is also going to change the way a lot of current jobs function, and it’s going to create entirely new jobs.” Sounds rough. But that trade-off is worth it because “if we could see what each of us can do 10 or 20 years in the future,” — note the near-religious certainty — “it would astonish us today.”

The Biden administration was intrigued by this sort of proselytizing, but not converted, and after issuing a sweeping executive order in October 2023 that AI should be a national priority but mustn’t harm labor, privacy, or trust in the truth, the president said, in the balanced fashion of an experienced lawmaker, that “to realize the promise of AI and avoid the risk, we need to govern this technology.” But those days of balance are over. President Trump does not strike me as a tech convert, but he seems to see no reason to stand in its way, and he rescinded Biden’s order within hours of taking office, and has spoken, if vaguely, about the need to preserve American tech dominance in the world.

Fear of competition from China has long been a favorite tech talking point in trying to convince lawmakers to go easy on American companies. And in a 15-page “Economic Blueprint” published this month, OpenAI’s policy chief lays out the rules and policies the company would like to see from lawmakers, and cites the twin threat of losing both money and political clout to China:

With an estimated $175 billion sitting in global funds awaiting investment in AI projects, if the US doesn’t attract those funds, they will flow to China-backed projects—strengthening the Chinese Communist Party’s global influence.

This week those talking points have caught fire, as Chinese AI burns the stock price of American companies, and upends expectations about how AI is built, and how much it should cost.

DeepSeek, a lab backed by a Chinese hedge fund called High-Flyer, released its open-source model DeepSeek-V3 the day after Christmas. The claims in its documentation must have made Altman’s head pop off.

Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable.

As CNBC reported, “in a set of third-party benchmark tests, DeepSeek’s model outperformed Meta’s Llama 3.1, OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5.”

DeepSeek founder Lian Wenfeng speaking on Chinese state television at a national symposium.

Analysts estimate that Meta’s model cost the company as much as $640 million. The Chinese lab says their DeepSeek-V3 competitor cost less than $6 million. Boom, there goes Meta’s share price, which sank 3 percent Monday morning. This morning DeepSeek is number one in the app store, displacing ChatGPT. Hey, there goes the share price of OpenAI’s backer, Microsoft, losing 4 percent.

The price difference between American and Chinese models has to do, in part, with the chips. See that “H800 GPU” mentioned in DeepSeek’s documentation? That’s a lower-performance chip made by Nvidia to satisfy export controls designed to keep the best chips out of China. Nvidia has profited hugely off the expectation that AI companies will require their absolutely fastest and fanciest chips (like the H100, which cost Meta as much as $40,000 apiece). Now DeepSeek has shown that you don’t necessarily need the top of the line chips reserved for American companies, and you don’t need as many of them. There goes Nvidia’s share price, which sank 16 percent, the greatest single-day rout in market history. (Taiwan chip behemoth TSMC also fell 14%.)

Chinese rivals to American AI companies are now everywhere. TikTok parent ByteDance released Doubao-1.5-pro, a model it says outperforms OpenAI’s. Another Chinese startup, 01.ai, says it’s better than GPT-4, and cost only $3 million to build. Maybe sinking a trillion dollars, enormous carbon emissions, and oceans of cooling water into training American models wasn’t the way to go.

But does this mean Altman and Andreesen and the rest are right, that in order to achieve a promised superior way of life we must make sure that American companies don’t fall behind their Chinese rivals? Here’s another way to think about it.

First, consider that the stock price of any company is a reflection of faith in that company, and literally nothing more. In “The Market as God,” his astounding 1999 article for The Atlantic, the Harvard theologian Harvey Cox told of reading the business press for the first time and realizing that investors and worshippers are part of the same tradition.

The lexicon of The Wall Street Journal and the business sections of Time and Newsweek turned out to bear a striking resemblance to Genesis, the Epistle to the Romans, and Saint Augustine's City of God. Behind descriptions of market reforms, monetary policy, and the convolutions of the Dow, I gradually made out the pieces of a grand narrative about the inner meaning of human history, why things had gone wrong, and how to put them right. Theologians call these myths of origin, legends of the fall, and doctrines of sin and redemption. But here they were again, and in only thin disguise.

The AI industry wears the same disguise. The promises of a better life (an afterlife?) on the other side of wrenching social and economic upheaval made by people who stand to profit from that upheaval — that’s pretty old-school stuff.

But let’s stand back further from the situation we’re in and consider whether the Chinese AI revolution deserves to be front-page news, or whether other developments deserve that placement. I wrote an entire book about the notion that AI could have enormous benefit for scientific researchers, historians, and agencies seeking to help welfare recipients, but that deploying it for profit will lead us to put aside our creativity, rationality, and agency. To me the ascent in the App Store of an open-source, free-to-use AI assistant designed in China is a matter of interest, but not a national emergency.

What is a national emergency is the freezing of scientific research and scientific communications at the federal level. Trump’s new head of Health and Human Services has issued a directive that no federal staff can be hired, no one can travel, and, without the authorization of a presidential appointee, no one can communicate health or scientific information to the public. That has left experts like this neuroscientist with no choice but to take to social media to report that their National Institutes of Health grant applications and research funding meetings have all been cancelled, meaning the sort of lifesaving NIH-funded work that gave us critical HIV treatments and two COVID vaccines have stopped entirely. This isn’t a pipeline of research that can be easily restarted, and it’s not optional, as the need to develop treatments for threats like avian flu becomes all too real.

NIH has, until now, been actively promoting the use of AI for diagnosing complex afflictions, for drug discovery, and for making the American medical system more responsive and efficient. These are the sorts of humanity-minded, practical explorations of the technology that don’t immediately turn a profit, but clearly offer the possibility of a better and longer life for billions of people on the other side: a true offer of real reward for the faithful. But the emergency being reported today has nothing to do with that. Instead, the language of alarm has been reserved for financial loss. And that says as much about the faithful as it does about the faith.

---

HillmanTok University is What We Needed

I recognize that I can be a relentless bummer on the topic of tech, so I want to leave you with something I consider very positive. I’m depressed about the dead-ended business prospects of creative work in an ecosystem entirely owned by a handful of companies, but the creativity itself is unmatched these days, and TikTok has something amazing going on.

Dr. Leah Barlow, an assistant professor in the liberal studies department of North Carolina AT&T University, posted a TikTok hello and syllabus for students in her LIBS 202 - Intro to African-American Studies course, and it somehow went viral. So viral that not only did she have to post an amazing calm-down-everyone thank-you message (“I had to put on my good blazer”), but the tone she set, in which she spoke to all her new followers as if they were actual students in the class, sparked a whole movement on the platform.

“I don’t know why they sent all of y’all to me, but now that you’re here, I know you need this class to graduate, so you can stay, but we have to set some ground rules.” (Courtesy: @afamstudies on TikTok)

Now Black academics from all over the country are posting their own course introductions on TikTok, and a fictional (but real?) university called HillmanTok has emerged. (It’s presumably named after Hillman College, where the eldest Huxtable went to school on The Real World.) You can take everything from Gardening 101 to Algorithms, Power, and Society to Intro American Sign Language, and each instructor is faithfully using first-day-of-class vernacular to welcome students and set boundaries.

Study groups have formed on Discord (so many that it reportedly had to be paused), there are questions about study guides and homework extensions, and students are posting their class schedule. The language of scholarship and awkward-undergrad vibes are amazing. It’s a whole amazing online thing, and fills me with joy. I try to call bad behavior when I see it, and there’s a lot of it these days, but I try to call good behavior too, and this is the best kind.

See you later this week, and thanks.