Sunday, July 26, 2026

Bets, Debts, and Bedrooms: The Week the AI Economy Showed Its Cracks

Imagine you run a lemonade stand charging $45 a cup. It's the best lemonade in town, and everyone says so. Then one morning, a new stand opens across the street selling lemonade that's almost as good — for 87 cents. Not $8.70. Eighty-seven cents.

That, in one image, is what just happened to the American AI industry. And it was one of three stories on last week's episode of The Week from Prof G Media (2026) that, taken together, tell you a lot about where the economy — and maybe your own future — is heading. Let's take them one at a time.

The bets: China isn't trying to beat American AI. It's trying to make it free.

On July 16th, a Chinese startup called Moonshot AI released a model named Kimi K3 — at 2.8 trillion parameters, the largest "open-weight" model ever built (Fello AI, 2026). Open-weight means anyone can download the model's brain and run it themselves, free; the full weights are scheduled for release on July 27[4] (VentureBeat, 2026). On some benchmarks it outperforms the best models from OpenAI and Anthropic, and it runs at about a third of Anthropic's price (Prof G Media, 2026, 00:00:44). Independent evaluators broadly agree: it ranks near the top of major model indexes while being cheaper, and analyst Nathan Lambert calls it "clearly the strongest open model ever released"[5] (Lambert, 2026).

Here's the number that should make you sit up. The price for a million output tokens — roughly the "words" an AI produces — is $45 for OpenAI's top model and $50 for Anthropic's Claude. For China's DeepSeek? $0.87. That's 99% cheaper (Prof G Media, 2026, 00:03:52).

Scott Galloway calls this "AI dumping" — flooding the market with something so cheap that competitors can't survive, the way China once did with solar panels and steel. And it's working: free Chinese models went from less than a third of global AI traffic in late 2025 to about two-thirds recently (00:01:19). Why can they do it? Cheaper power, cheaper chips, and subsidies from local governments (00:01:47). As co-host Ed Elson put it, the Chinese models get the job done "fast enough, well enough," at prices so low "it would be ridiculous not to turn to them" (00:04:32).

But one guest offered a smarter frame. Charlie O'Neil, who trains AI models for a living, argued the real battle isn't China vs. America — it's open source vs. closed source. For years we were told secret, locked-up models would always stay ahead. Turns out "there's no secret sauce" (00:05:53). And a world where intelligence is open might beat a world where two companies "dictate all the terms of access" (00:06:40).

Here's the irony, and it stings: America — home of the free market, the country that spent decades lecturing the world about competition — is now the one hoping to be protected from competition, while communist China plays the role of the aggressive discount capitalist. The student didn't just learn the lesson. He's teaching it back, at 99% off.

And it's not just products; it's reputation. Pew Research Center (2026) found that in most of the 36 countries surveyed, more people now have a favorable view of China than of the US — including America's nearest neighbors, Canada and Mexico[6]. That's the first time in roughly 20 years of tracking that China has come out ahead[3] (NBC News, 2026). Meanwhile, 84% of Chinese say they're more excited than worried about AI; in America, that number is around 10% (Prof G Media, 2026, 00:07:49). One former Bloomberg reporter explained the gap simply: in China, AI isn't a "machine god" or a Terminator scenario — it's a tool your boss tells you to learn (00:08:54).

Cheaper models plus a population eager to use them. As host George Hahn put it, "For America, that's a difficult combination to compete with" (00:09:52).

The debts: the AI boom is being built on borrowed money

Now flip to the money side. If Chinese AI is nearly free, how do American companies justify spending hundreds of billions on data centers?

Increasingly, they don't spend their own money. They borrow it.

Exhibit A: Oracle. Its stock is down 35% this year, and S&P downgraded the company to BBB-, one notch above junk status, citing an uncertain path to profitability amid heavy AI spending[2] (Trader's Union, 2026). "Junk" is finance-speak for borrowers likely to have trouble paying you back. Oracle borrowed $43 billion in a year to build data centers. Its revenue is $67 billion, but as Ed explains, "revenue doesn't pay debt down, free cash flow does" — and Oracle's free cash flow is negative. It burned about $24 billion (Prof G Media, 2026, 00:12:47).

Translation for the group chat: imagine borrowing $43,000 to build a gaming setup while your part-time job leaves you $24,000 short every year. At some point, the bank stops smiling.

That point may have arrived. Wisconsin's utility regulator upheld a rule requiring Oracle to post a $7 billion letter of security for its $15 billion Port Washington data center — costing the company over $100 million annually[2] (Trader's Union, 2026). A security deposit, essentially — the kind landlords demand from tenants they don't quite trust. And Wisconsin isn't alone: twenty-four states have approved similar tariffs for data centres and other major industrial users, typically requiring minimum contract terms, exit fees and collateral[2]. When utility regulators across the country want their money up front, trust is eroding.

Why does this matter beyond one company? Because of the episode's most quotable warning: "bubbles aren't built with equity, they are built with debt" (00:13:25). When a boom is funded by investors' own money and it pops, investors lose money — painful but contained. When it's funded by debt and it pops, the losses cascade to lenders, banks, and pension funds. That's 2008. Combine the bets and the debts and you see the trap: American labs are borrowing billions to build capacity for products a Chinese competitor gives away nearly free.

The bedrooms: the casino economy is turning young men into monks

The third story feels different, but it's connected — it's about what all this technology is doing to the people who grew up inside it.

Writer Derek Thompson calls our era "the antisocial century," and in his essay The Monks in the Casino (Thompson, 2025) he makes a striking argument: people have a fixed appetite for risk, and young men haven't lost theirs — they've relocated it. Risk used to mean asking someone out, moving cities, starting a band. Now, Thompson says, there's been "almost a clean transference" of that risk impulse — away from the real world and into the bedroom: sports betting, crypto, prediction markets like Kalshi (Prof G Media, 2026, 00:14:00). In the essay itself, he describes young men who have become risk-averse in the physical world and risk-seeking in the digital one — they date less and gamble more, finding intimacy scary and betting exciting[6].

He then flips a famous idea on its head. Sociologist Max Weber argued that Christian self-discipline — saving, restraint — gave birth to capitalism. Today, Thompson says, it's inverted: "it is capitalism that is giving birth to a kind of wretched asceticism." The casino economy is producing young men who take wild financial risks on their phones while living like monks — alone, indoors, socially minimal (00:15:00).

Why is that a problem, if someone likes being alone? Thompson's answer is the best minute of the episode. Friendship, he says, works like a vaccine. You don't get vaccinated for the days you're healthy; you get vaccinated so the worst day doesn't destroy you. Same with people: "life is often tragedy" — losing a job, losing a parent, a mental health crisis — and in those moments, "not having a social group to fall back on, that is the real risk" (00:16:03, 00:16:34). If you haven't invested in relationships, "you are entirely on your own at the very moment that you need to be surrounded by love" (00:17:40).

For a sixteen-year-old, this might be the most practical takeaway of the three: the riskiest bet isn't the parlay on your phone. It's assuming you'll never need anyone.

What ties it all together

Three stories, one thread: misplaced bets. America may be betting on the wrong AI business model (closed and expensive vs. open and free). Companies like Oracle are betting borrowed billions on demand that cheap Chinese models may undercut. And a generation of young men is betting its limited appetite for risk on apps instead of on life. The episode doesn't say the sky is falling — but it does suggest that the smartest move, whether you're a superpower, a corporation, or a teenager, is the same: check where your risk actually is, not where it feels like it is.

References

Fello AI. (2026, July). Kimi K3: Moonshot's 2.8T open-weight model explained. https://felloai.com/kimi-k3/

Lambert, N. (2026, July). Kimi K3: The open-weights escalation. Interconnects. https://www.interconnects.ai/p/kimi-k3-the-open-weights-escalation

NBC News. (2026, July 15). China and Xi are seen more favorably than the U.S. and Trump in many nations, new survey says. https://www.nbcnews.com/world/asia/china-xi-are-seen-favorably-us-trump-many-nations-new-survey-says-rcna587789

Pew Research Center. (2026, July 15). People in many countries now view China more positively than the U.S. https://www.pewresearch.org/global/2026/07/15/people-in-many-countries-now-view-china-more-positively-than-the-u-s/

Prof G Media. (2026, July 24). The week [Audio podcast episode]. Prof G Media.

Thompson, D. (2025, November 11). The monks in the casino. Derek Thompson Substack. https://www.derekthompson.org/p/the-monks-in-the-casino

Trader's Union. (2026, July 21). Oracle faces potential $7bn collateral requirement for Wisconsin data centre. https://tradersunion.com/news/financial-news/show/2732284-oracle-wisconsin-data-centre-collateral/

VentureBeat. (2026, July). China's Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems. https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems

Learn more:

83% vs 39%: Why China Loves AI, America Fears It — and Who's Right

This week in AI

Last Thursday, a Chinese startup you've probably never heard of did something remarkable. Moonshot AI released Kimi K3, the largest open-weight AI model ever built — 2.8 trillion parameters, which is a fancy way of saying it's enormous. On some benchmarks it outperforms the best models from OpenAI and Anthropic, and it runs at about a third of Anthropic's price (Prof G Media, 2026, 00:00:44). Even more remarkable: unlike American frontier models, which are locked behind corporate walls, Kimi K3's weights are free. Anyone can download it, study it, and build on it.

Babel or Jerusalem? What the Pope Gets Right — and Half-Wrong — About AI

Pope Leo XIV just wrote Magnifica Humanitas, a 40,000 words Encyclical Letter about artificial intelligence. He compares Silicon Valley to the Tower of Babel, warns that algorithms are quietly deciding who gets a job or a loan, and even apologizes for the Church's own past sins. It might be the most important document about technology you'll never read — so I read it for you. And while I applaud most of it, as an economic historian I have a bone to pick with the ghost of the Luddites hovering between its lines.

A social encyclical for the age of AI

On 15 May 2026, Pope Leo XIV published Magnifica Humanitas ("The Grandeur of Humanity"), an encyclical letter "on safeguarding the human person in the time of artificial intelligence" (Leo XIV, 2026). The date was no accident. Exactly 135 years earlier, on 15 May 1891, Pope Leo XIII published Rerum Novarum, the document that founded what we now call the Social Doctrine of the Church — the Church's body of teaching on work, the economy, and justice (Leo XIII, 1891). Back then, the "new things" (res novae) were factories, industrial capitalism, and the exploitation of workers. Today, the new pope argues, the res novae are digitalization, robotics, and AI (Leo XIV, 2026, para. 4).

Europe's Firms Are Adopting AI — But at Three Very Different Speeds

The people got there first

Something historically unusual is happening with artificial intelligence. For most general-purpose technologies, businesses led and households followed. Electric dynamos powered factories for roughly four decades before households electrified at scale, and the personal computer entered offices years before it entered living rooms (Comin & Hobijn, 2010; David, 1990). Generative AI has inverted that sequence: by August 2024, nearly 39% of the U.S. working-age population had already used it — a faster initial diffusion than either the PC or the internet at comparable points (Bick et al., 2024) — while firm-level surveys were still reporting adoption under 10% (Bonney et al., 2024). Workers, quite literally, dragged AI into their companies in their pockets.

Inside the Mind of the Machine: is AI lying to us?

6 Counter-Intuitive Takeaways on the Road to AGI

Standing amidst the stones of King’s Parade in Cambridge, one cannot help but feel the weight of intellectual history. It is a city where the "intellectual giants" of the past—from Charles Babbage to Alan Turing—once walked, laying the theoretical foundations for the world we now inhabit. Yet, as I sat in a historic Cambridge lecture hall listening to Demis Hassabis and later spoke with the researchers at Google DeepMind, a startling paradox became clear: we have entered an era where we can build systems of immense intelligence that we do not fully understand. We have graduated from traditional software engineering into the "Black Box" problem, creating machines that mirror human intuition more closely than they do traditional logic.

The Root Node of Reality: Why Demis Hassabis Thinks AI is the Successor to Mathematics

A Homecoming to the Future

There is a profound symmetry in Demis Hassabis (Nobel prize winner in Chemistry 2024) returning to the wooden benches of his favorite Cambridge lecture hall. It was here, as an undergraduate, that he absorbed the theoretical underpinnings of computation that would eventually power DeepMind. Before the lecture, Hassabis was invited to sign the Nobel book—a ritual of scientific passage. As he leafed back through the vellum pages, he found himself staring at the signatures of Francis Crick and Albert Einstein. It was a visceral reminder that he wasn't just visiting an alma mater; he was stepping into a lineage of giants who decoded the fundamental scripts of reality.

Hassabis’s journey to this moment began not with a laboratory, but with a "lump of inanimate plastic"—his first chess computer. As a four-year-old child prodigy, he was less interested in winning than in the mystery of how a machine could be programmed to outthink a human. This "relatable curiosity" evolved into a lifelong quest to understand the "root node" of existence: intelligence itself.

Education and Its Limits

Education and Its Limits: What Daniel Susskind Gets Right — and Where I'd Push Back

Notes from Daniel Susskind's Gresham College lecture, "Education and Its Limits" (16 May 2026)

Daniel Susskind closed his Gresham College lecture series on the future of work with a talk that should land squarely in the inbox of every school leader, curriculum designer, and policy maker currently drafting an "AI strategy" (Susskind, 2026, 00:10:33). His core claim: more education remains our best response to technological disruption, but what we mean by "more education" has to change — and even a reformed version of education won't be enough on its own (Susskind, 2026, 00:16:29).

Below is my summary of the argument, followed by where I think it holds up and where I'd want to see it stress-tested.

AI4TL Artificial Intelligence for Teaching & Learning

Followers