The Long View · · 9 min read

The Bubble Isn't the Danger. The Country It Lands On Is.

An analysis of the situation as of September 2026.

By Luke Cathcart

I started out worried about the AI bubble. I ended up worried about something the bubble can't touch: whether the country underneath it can still act as a country.

Here is how the fear goes, and I suspect it's a common one. A handful of American labs have absorbed more private capital than any companies in history. Anthropic raised $65 billion in May at a $965 billion valuation;1 OpenAI raised $122 billion in March at $852 billion.2 Together, by Crunchbase's count, they took 43 percent of all venture funding on earth in the first half of this year.3 Both are headed for public markets at multiples that make 1999 look modest, and the profits are only now starting to appear: Anthropic has told shareholders it expects a second straight quarter of adjusted operating profit, a figure that excludes stock compensation and training costs and that won't be audited until a prospectus is public,4 while OpenAI has reportedly told investors it doesn't expect to be profitable before 2030.5 The five largest U.S. hyperscalers are on track to spend roughly $700 to $800 billion on capital projects this year,6 and a growing share of it is financed with debt, which 2000 mostly wasn't.7 The Bank of England's governor has said out loud that an AI-driven market correction could spread across the world.8

Meanwhile the American public has soured on the technology it is paying for. Pew finds that 52 percent of Americans now say they're more concerned than excited about AI, up from 37 percent in 2021.9 Three quarters of registered voters tell pollsters they'd oppose a data center near them.10 More than 530 counties and municipalities have restricted or banned construction,11 and local opposition blocked or delayed at least 75 projects worth roughly $130 billion in the first quarter alone.12

Put that together and the story writes itself. The bubble pops, the economy cracks, a divided country turns on itself, and some rival, China is the usual name, takes the top spot while America struggles to stand back up. I believed a version of that story for a while. Then I tried to test each link, and each one broke in the same direction.

Every link pointed inward

Start with the crash. A bubble bursting is a multi-trillion-dollar paper loss, a recession, and a wave of bankruptcies. It is not a collapse. The United States absorbed the dot-com crash, which erased roughly $5 trillion in market value, about half of GDP at the time,13 and the 2008 crisis, which hit the banking system directly, and came out of both still dominant. This one could be worse; the ten largest stocks now make up about 41 percent of the S&P 500, against roughly 27 percent at the 2000 peak,14 and more of it sits in ordinary retirement accounts through index funds. One difference cuts the other way: this boom is building physical things, chips, grid capacity, real estate, that have salvage value a 1999 website never did. But the dot-com era overbuilt physical assets too, and its fiber is exactly what got sold for pennies; collateral limits how far a loss spreads, it doesn't prevent it. And GPUs are worse collateral than fiber, because they depreciate in a few years and the next generation makes them cheap. Taken together, that's an argument for a nastier recession. It's not an argument for state failure, which requires the currency, the courts, or the military to stop working. Nothing in the capex numbers touches those.

Then the rival. When bubbles burst, the assets don't vanish; they get bought at a discount. But a crash doesn't automatically transfer America's AI infrastructure to a foreign power. The strategically important pieces, the data centers, the chips, the model weights, remain subject to American ownership rules, export controls, and national-security review, and the distressed private assets mostly create opportunities for domestic firms and capital, exactly as Google bought dark fiber for pennies after 2001.15 A crash tends to concentrate American AI into fewer, richer American hands. What an outside rival picks up is talent and market share at the margins, and a narrative. Real, but not an inheritance.

Then the plot. If you can't invade America, the reasoning goes, the only way to beat it is to poison it from within over decades. That logic is real; every major adversary of the last century has written some version of it down and tried it, from Soviet active measures to modern troll farms. But the best study of the 2016 Russian campaign found that one percent of users accounted for 70 percent of exposures, that exposure was dwarfed by domestic news and politicians, and that there was no measurable relationship between exposure and changes in attitudes, polarization, or voting.16 Foreign campaigns can amplify divisions that already exist. The evidence suggests they are better understood as accelerants than architects. Swap in any enemy you like and the finding holds.

At every step I reached for a bigger external cause, and at every step the evidence pointed back home.

What China actually shows us

If China matters to this story, it isn't as a villain. It's as a mirror.

Strip away the geopolitics and the most striking difference between the two countries is not who builds the better model. It's how each public feels about living with the result. In the 2026 Ipsos AI Monitor, 83 percent of Chinese respondents agreed that AI products and services have more benefits than drawbacks; 39 percent of Americans did.17 Chinese chatbots are free rather than twenty dollars a month, and the leading ones report user bases in the hundreds of millions.18 Alibaba's open Qwen models have reportedly spawned over 100,000 derivative models on Hugging Face, the largest open-weight ecosystem on the platform.19 In China, the technology is increasingly treated as infrastructure: something you plug in and get on with, like electricity.

That comparison deserves a caveat, and it's an important one. Polling a public whose media is state-directed, where open dissent carries risk, and where the baseline expectation of surveillance and automation was already different, does not measure the same thing as polling Americans. Ipsos's own China sample is online, urban, and more educated than the population.17 Some of the reported optimism is real enthusiasm and some is the absence of a permitted alternative. The gap is real; how much of it is cultural and how much is structural is harder to say, and the argument here doesn't depend on knowing.

In the United States the same technology has become a culture-war object. It's a job thief, a power-bill inflater, a thing that ruins the view from the highway. Some of that anxiety is earned. And Americans have started to believe they're losing: in Pew's June survey, 36 percent said China is more advanced in AI, against 12 percent who said the United States, a three-to-one margin that holds across both parties.20 Whether or not that perception is accurate, and most expert assessments say it isn't,21 the belief itself is doing work. A public that thinks it's already behind is a public that's easier to talk out of building.

So the contrast is this: one society is absorbing the technology and arguing about how to use it, and the other is increasingly arguing about whether to have it at all, while half-convinced it has already lost. The first is in a stronger position to recover from a crash, regardless of who owns the servers. Whatever explains the Chinese side of the contrast, the American side is a choice.

The asset nobody is pricing

Here is what fell out when the story collapsed.

Everyone in this debate is counting the wrong kind of infrastructure. A data center goes on a balance sheet; so does a bridge or a chip fab. But a country's ability to coordinate under stress is infrastructure too, and it's the piece that determines whether an economic shock stays economic or becomes political. America survived its previous crises not because its economy was strong but because enough people trusted the institutions, and each other, to accept a shared response. The New Deal, the postwar recovery, and the rescue after 2008 all required enough people to believe the response was legitimate, even when they hated the party delivering it. That trust is the actual national asset. GPUs and data centers are replaceable. A population that can agree on what happened and what to do about it is not, and it has been depreciating for decades while everyone watched the stock tickers.

That depreciation was not a plot. It was the sum of individually sensible choices in an environment that rewarded division. People moved to places where their neighbors agreed with them. The parties sorted until geography, religion, and education all lined up with a single team, which they did not in 1975.22 Cable news and then social media discovered that anger holds attention better than anything else and built businesses on it. The places where people from different camps used to sit in the same room, local papers, unions, churches, civic clubs, thinned out. No single person chose this. Everyone responded to their incentives like anyone would. The incentives got worse.

This is why every version of the outside-enemy story is a distraction, even though each contains a grain of truth. They point outward. The vulnerability is inward. An adversary does not need to beat America. It only needs America to be unable to act together when something goes wrong, and the country has been doing most of that work on its own.

The danger, stated plainly

The danger is not that the bubble pops. The danger is that America faces its next big shock in a condition where it cannot respond as a country.

Great powers rarely fall from a single blow. They fall from a series of bad responses to blows. Japan did not lose its factories after 1990; it lost two decades of growth and confidence. The plausible way for the United States to lose its lead is not a foreign fire sale or a civil war. It is a country that comes out of a crash politically exhausted, cuts the research and immigration that built its advantage, and lets the data-center backlash harden into permanent bans instead of better terms.

Put more simply: a competitive advantage can be lost through an inability to make decisions, not only through technological inferiority. A country that spends ten years unable to agree on what happened, what matters, or what to do next can fall behind a society more at peace with the technology, however good its models are. That is not a prediction. It is the scenario with historical precedent, which makes it the one to prepare for.

The unglamorous fix

The uncomfortable part is that the defense is boring. It is not a policy or a leader. It is whether people rebuild the middle layer of a society: local institutions, relationships across political lines, information habits that don't run on outrage. Nobody goes viral for joining a school board or having dinner with someone who votes the other way. It is also the only defense that works against every version of the threat at once, whether the shock comes from a market, a foreign government, or a political crisis.

I wanted a villain. Most people do; half the media ecosystem is built to supply one. But a problem with a villain is something you can fight. A problem without one is something you have to work on. The good news buried in all of this is that the thing most worth protecting is not in Washington or Beijing or a data center in Texas. It is in the town you live in, and it is still within reach.

Stop asking whether the crash is coming. Start asking what kind of country it lands on. The first question is out of your hands. The second one isn't.

Sources

  1. Anthropic, "Anthropic raises $65B Series H at $965B valuation," company announcement, May 28, 2026. https://www.anthropic.com/news/series-h — Independently reported by Bloomberg and TechCrunch the same day.
  2. Bloomberg, "OpenAI Valued at $852 Billion After Completing $122 Billion Round," March 31, 2026. https://www.bloomberg.com/news/articles/2026-03-31/openai-valued-at-852-billion-after-completing-122-billion-round — Confirmed by CNBC and OpenAI's own announcement.
  3. Crunchbase News, "Global Startup Investment Hit Record $510B In H1 2026," July 2, 2026. https://news.crunchbase.com/venture/global-startup-exits-ipo-ma-soar-ai-q2-h1-2026/ — $217B of $510B.
  4. Bloomberg, "Anthropic Expects an Operating Profit This Quarter, FT Says," September 13, 2026, citing the Financial Times. https://www.bloomberg.com/news/articles/2026-09-13/anthropic-sees-adjusted-operating-profit-this-quarter-ft-says — The exclusion of stock-based compensation is per the FT report; the note that audited GAAP figures arrive only with a prospectus is from Forbes, August 17, 2026, https://www.forbes.com/sites/jonmarkman/2026/08/17/anthropics-groundbreaking-second-quarter-delivers-115b-in-revenue/. Source strength: reported guidance to investors, not audited financials.
  5. Yahoo Finance, "Anthropic on Track for First Profitable Quarter," May 21, 2026. https://finance.yahoo.com/markets/stocks/articles/anthropic-track-first-profitable-quarter-124217577.html — The 2030 figure is attributed to what OpenAI told investors. Source strength: secondary; the phrase "reportedly" in the text reflects this.
  6. J.P. Morgan, "Financing AI infrastructure and U.S. data centers," August 2026 (estimate of $697B for the five largest U.S. hyperscalers). https://www.jpmorgan.com/insights/banking/capital-markets/financing-ai-infrastructure-data-centers — The higher end of the range reflects Q1 2026 company guidance compiled by AL Capital Advisory, August 2026, https://alcapitaladvisory.com/research/intelligence/ai-infrastructure.html.
  7. Debt financing: Amazon raised more than $100 billion in debt in 2025 and Oracle completed an $18 billion bond sale, per Nexi Fund's July 2026 summary (https://nexi.fund/ai-infrastructure-guide-hyperscaler-capex-2026/); Morgan Stanley and J.P. Morgan project roughly $1.5 trillion in new tech-sector debt over three years, per the same J.P. Morgan piece in 6 and CreditSights, November 2025. Source strength: the individual bond figures are from a secondary summary; the $1.5T projection is attributed to the banks.
  8. RTÉ, "AI could cause economic downturn – Bank of England chief," September 1, 2026, reporting Andrew Bailey's letter to G20 finance ministers in his capacity as chair of the Financial Stability Board. https://www.rte.ie/news/business/2026/0901/1589905-ai-could-cause-economic-downturn-bank-of-england-chief/
  9. Pew Research Center, "Young US adults are increasingly wary of AI, concerned it will take jobs," August 18, 2026 (survey conducted June 22–28, 2026). https://www.pewresearch.org/short-reads/2026/08/18/young-adults-in-the-us-are-increasingly-wary-of-ai-concerned-it-will-take-jobs/
  10. Heatmap News, "Exclusive: 75% of Americans Now Oppose Local Data Center Development," August 20, 2026. Heatmap Pro/Embold Research poll of 2,045 registered voters, August 8–13, 2026, ±2.3 points. https://heatmap.news/daily/data-center-opposition-poll-collapse
  11. Same Heatmap article as 10; the 530+ figure is Heatmap Pro's own tracking data.
  12. Data Center Watch, "Q1 2026: Data Center Watch Report." https://www.datacenterwatch.org/q1-2026 — Data Center Watch is a project of 10a Labs, a private research firm; the figure is their count, not a government total. Reported by NBC News, June 12, 2026, and Newsweek, July 24, 2026.
  13. The roughly $5 trillion figure for Nasdaq losses from March 2000 to October 2002 is widely cited (e.g., in Federal Reserve and academic retrospectives). U.S. GDP in 2000 was approximately $10.25 trillion (Bureau of Economic Analysis). Note: I did not run a fresh search on this; it is a well-established figure.
  14. RBC Wealth Management, "The 'Great Narrowing': S&P 500 concentration," January 23, 2026 (top-10 weighting at a record 40.7% in 2025 versus 18–23% from 1990 to 2015). https://www.rbcwealthmanagement.com/en-us/insights/the-great-narrowing-sp-500-concentration — The ~27% figure for the 2000 peak and the 41% current figure are per Kobeissi Letter data reported by Yahoo Finance, May 22, 2026.
  15. Google's purchase of unused ("dark") fiber after the telecom bust was widely reported in 2005 (e.g., Business 2.0, Wall Street Journal). Note: not freshly searched; well-documented historical fact.
  16. Eady, G., Paskhalis, T., Zilinsky, J., et al., "Exposure to the Russian Internet Research Agency foreign influence campaign on Twitter in the 2016 US election and its relationship to attitudes and voting behavior," Nature Communications 14, 62 (2023). https://www.nature.com/articles/s41467-022-35576-9 — The authors caution that their finding concerns individual-level attitudes on Twitter and does not rule out other effects, such as on faith in electoral integrity. This essay's "accelerants, not architects" framing is consistent with that caution.
  17. Ipsos, "The Ipsos AI Monitor 2026," June 2026 (23,532 adults under 75 across 32 countries, March 20–April 3, 2026). https://www.ipsos.com/sites/default/files/ct/news/documents/2026-06/Ipsos-AI-Monitor-2026.pdf — China 83%, United States 39% agreeing that AI products have more benefits than drawbacks. Ipsos notes that samples in China and several other markets are online and skew more urban, educated, and affluent than the general population.
  18. Reported user figures (e.g., Baidu's Ernie Bot at 300 million users) are compiled by DataGlobeHub, February 2026, https://dataglobehub.com/china-ai-statistics-and-insights/. Source strength: weak. This is a statistics aggregator citing company-reported numbers, which is why I hedge the wording above to "report user bases in the hundreds of millions."
  19. Digital in Asia, "What is China's AI Strategy in 2026?", July 25, 2026. https://digitalinasia.com/china-ai-models-chips-strategy/Source strength: secondary. The 100,000 figure is attributed to Hugging Face data, and I did not verify it on Hugging Face directly, which is why the text says "reportedly."
  20. Pew Research Center, "What Americans think about the global AI race," July 23, 2026 (survey of 3,488 adults, June 22–28, 2026). https://www.pewresearch.org/short-reads/2026/07/23/what-americans-think-about-the-global-ai-race/
  21. A Boston Consulting Group comparison published in June 2026 placed the U.S. ahead on capital, talent, IP quality, data, energy, and compute, while noting China closing the compute gap; Epoch AI's 2026 analysis puts Chinese models roughly seven months behind the U.S. frontier. Both cited in New Space Economy, July 24, 2026, https://newspaceeconomy.ca/2026/07/24/why-do-americans-think-china-leads-the-global-ai-race/. Source strength: secondary summary of primary reports.
  22. The sorting of the parties along geographic, religious, and educational lines since the 1970s is the central finding of a large political-science literature; see Pew Research Center's long-running "Political Polarization in the American Public" series (2014 onward) and Lilliana Mason, Uncivil Agreement (2018). Note: not freshly searched; standard reference.

Who wrote this

Start a conversation