The Mortgaged Empire
When the geopolitical race for artificial intelligence meets its financial bubble
A joint reading of “Empires of AI” (Niall Ferguson, Lisa Klaassen, and Eric Schmidt) and “The Global AI Bust” (Ruchir Sharma), Foreign Affairs, November/December 2026
Two essays in the same
issue of Foreign Affairs describe the same race from angles that are
rarely read together. Ferguson, Klaassen, and Schmidt argue that artificial
intelligence has given Washington and Beijing a new form of empire, and that
the decisive question may not be who reaches artificial general intelligence
first, but who can bankroll the race. Sharma answers, from the standpoint of
finance, that the race is being paid for with a bubble: real in its technology,
fragile in its structure of debt and subsidy. Read together, the two texts suggest
a thesis neither fully states: the AI empire is mortgaged, and the date that
mortgage comes due will be set less by generals or engineers than by the debt
markets.
1. Two diagnoses, one variable
For Ferguson and his coauthors, AI is an imperial technology for a concrete reason: it brought scarcity back to a sector that had ignored it for decades. According to the scaling laws, progress depends on energy, critical minerals, chips, and capital on a massive scale, and once a technology’s frontier depends on scarce inputs, its production becomes a matter of geopolitics. Of all those inputs, capital is the one most likely to constrain both empires. The largest U.S. technology firms have committed more than $600 billion in capital expenditure for 2026, while Beijing, according to a draft plan reported by Bloomberg, intends to spend $295 billion over five years, funded mainly through long-dated government bonds (pp. 71–72).
Sharma looks at the same capital from the inside. Technology spending on plants and equipment now stands at 6.3 percent of U.S. GDP, above the 6.0 percent reached at the peak of the Internet bubble. Of the five hyperscalers, only Microsoft still generates surplus cash flow; four have increased their net debt by issuing bonds, and leases and off-balance-sheet arrangements could add roughly $3 trillion to their debt over the next decade (pp. 85–87). What Ferguson presents as the grand strategic wager of two powers, Sharma presents as a familiar cycle: companies spending more than they can recover, investors borrowing to buy their shares, and prices drifting away from any profits the technology can support.
The point of contact is public debt. Ferguson notes that the United States crossed what he calls the “Ferguson limit” in 2024, paying more in interest (about $881 billion) than on defense (about $874 billion), and that on the current trajectory interest will be double the defense budget by 2036 (p. 72). Sharma identifies that same debtor as the signal that matters: bubbles pop when liquidity tightens, and the players to watch are now less central bankers than the “bond vigilantes,” who punish free-spending governments by selling their bonds (p. 88). The joint reading is uncomfortable: the same debt that narrows the empire’s fiscal room is what can close the tap that feeds the AI expansion.
2. The shared architecture: the stack as empire and as chain of contagion
Ferguson describes AI as a lattice of chokepoints—chips, minerals, models—that the metropoles can switch on or off at will. The text cites a recent example: in June, the U.S. Commerce Department told Anthropic to suspend foreign nationals’ access to its two most capable models (p. 69). There the center decides who gets what; the periphery supplies minerals and cheap labeling work and receives the refined product back at a markup.
Sharma traces the same web, but as a map of economic dependence. TSMC, SK Hynix, and Samsung weigh more in their local indexes than big tech does in the U.S. index; South Korea’s market value rose from $1.5 trillion to more than $5 trillion between early 2025 and this summer; Taiwan’s economy has grown at near double-digit rates; high tech accounts for as much as half of Israel’s growth. Further down the chain, Singapore, Malaysia, Mexico, and Vietnam are receiving data centers and server plants, and suppliers of nickel, lithium, and copper close the chain (pp. 81–82).
What is geopolitical dependence for Ferguson is a transmission channel for Sharma. And both converge on the same node: Taiwan. Ferguson identifies it as the most likely scenario for great-power war, because TSMC sits roughly 90 miles off the Chinese coast and capturing the island would be enough to blunt the supremacy of American models (p. 76). Sharma shows it as one of the most extreme stock market concentrations in the world. A military shock and a financial bust would therefore strike the same point; this coincidence, which neither author stresses, is probably the most important finding of reading them together.
3. The dollar: advantage and exposure
Ferguson argues that the dollar and Wall Street underwrite the American AI empire: owning the reserve currency lets Washington fund its data center build-out with other people’s savings, while China must finance from within, through a state-directed banking system (p. 71). Sharma supplies the other side of that privilege. Foreigners now hold more than $20 trillion in U.S. stocks, up from under $2 trillion after the dot-com bust, with much of the increase coming after ChatGPT’s launch. American households keep more than 30 percent of their wealth in stocks, above the 27 percent of 2000, mostly through passive funds tracking an index dominated by AI (p. 85).
Public support adds to this. According to Sharma, the deficit, above 6 percent of GDP, accounts for more than half of U.S. corporate profits this decade, twice the dot-com level, and the Federal Reserve holds its rate a full point below what academic guidelines would advise (p. 87). Ferguson adds a feedback loop: the capital demands of AI are already pushing interest rates higher and worsening the debt (p. 73). Our reading, which combines the two arguments, is of a circle: the reserve-currency privilege finances the empire and at the same time globalizes its fragility, so the periphery does not merely supply inputs; it also holds the paper that funds the expansion.
4. China: the rival that compresses margins
Ferguson observes that Beijing is following the latecomer path that carried challengers past Britain in the nineteenth century: investing where Washington lacks an edge or cannot match the cost—energy, manufacturing, robotics, and open-weight models at or near the frontier—so that the United States must spend lavishly to defend its lead in a sector China is commoditizing (p. 74). Sharma notes that Chinese models are cheaper and perform nearly as well, and that net exports, mainly of AI-related goods, now account for a third of China’s growth, in a domestic economy weighed down by bad debts and a shrinking labor force (p. 81).
Two consequences follow from combining them. First, competition from cheap models limits the pricing power that Western valuations assume; Sharma points out that two frontier labs charge subscribers less than it costs to train and run their models, yet analysts value each at well over a trillion dollars (p. 84). This is our inference, not a claim by the authors. Second, China is no safe harbor: its economy depends on AI through exports, and Ferguson warns that its own crisis could be set off by overcapacity and the margin-destroying competition known as neijuan, with high youth unemployment and local governments already borrowing solely to roll over old debt (pp. 72–73). A bust would not be an American matter alone.
5. The political ceiling: a bust without a safety net
Ferguson sums up the historical dilemma of empires in what he calls the “imperial trilemma”: a power can do two of three things—defend its dominions, maintain technological primacy, or provide generous welfare to the metropole—but not all three (p. 75). Sharma adds the context of a world without a cushion: global debt has surpassed $350 trillion, more than three times the size of the world economy, and few governments will have the money to stimulate battered economies or bail out businesses (p. 90).
Moreover, it is doubtful the public would allow it. Sharma cites that most Americans say they are more concerned than excited about AI, that seven in ten expect fewer jobs, and that local opposition blocked or delayed data center projects worth some $130 billion in the first quarter alone (p. 90). Ferguson records the same backlash, even in Republican-led states such as Texas and Utah (p. 75). The combination suggests that in 2008 and 2020 the state absorbed the blow, and that in an AI bust it may be unable or unwilling to do so. A bursting bubble would thus work as a stress test for the empire: growing deficits, shrinking room, and military commitments that do not go away, just when Ferguson warns that American overextension in the Middle East may tempt Beijing (pp. 76–77).
6. Where they disagree
The authors do not agree on the essentials of the outcome. Ferguson sees in AI a durable imperial resource whose main threat is overreach, rivalry, and, at the extreme, war; his way out is a form of détente that limits the race through agreements, for example by setting capability thresholds for models with cyber-offensive or biological potential (p. 78). Sharma sees a “good” speculative episode, comparable to the Internet or the railroads: after the crash a more streamlined network of survivors will remain, and the technology will keep advancing (p. 91).
The underlying question is whether the bust is the threat or the purge. Our reconciling hypothesis is that the crash would not end the technology (Sharma), but could change who is able to sustain the imperial order built upon it (Ferguson). The survivors would be those with access to the deepest and cheapest capital, whether dollar-financed or backed by the Chinese state. The bust would act as a selection mechanism among firms and also among empires. There is also an important agreement: neither believes AI is a magic card against fiscal problems. Ferguson warns that productivity gains may arrive too late to avoid the debt trap; Sharma recalls that technologies usually lift productivity decades after they appear, and that current forecasts range from 0.1 to nearly 2 percentage points a year (pp. 73, 91).
7. What to watch
• Long-term bond yields and episodes of “bond vigilantes” in the United States, Europe, and Japan: the liquidity signal that, according to Sharma, precedes busts.
• Circular funding and “other income” at the hyperscalers, along with off-balance-sheet leases: they measure how much of the earnings is real.
• Initial public offerings of the frontier labs, with prices below cost and valuations without profits.
• Semiconductor volatility in Korea, Taiwan, and Japan, and any sign of tension in the Taiwan Strait.
• Refinanced Chinese local debt and signs of neijuan in subsidized sectors.
• Local opposition to data centers and governments’ fiscal response to a potential shock.
8. A view from the periphery: a reading for the Andes
Both texts place the region in the supply chain. Sharma lists copper from Chile, Peru, and the Democratic Republic of the Congo among the inputs that will sustain years of demand from the worldwide AI construction project, and warns that the shock of a bust would reach, at the end of the chain, miners and power producers (pp. 82, 89). Ferguson describes the pattern of extraction in the periphery and refining and resale in the metropole (p. 66). A reading of our own, not the authors’: economies such as Peru’s have high exposure to the AI demand cycle and low participation in its rents, so their best defense lies in following global liquidity indicators more closely than technology announcements. Sharma adds that countries such as India and the Philippines are doubly exposed, left out of the investment wave and at risk of displacement in their back-office services (p. 82).
Conclusion
Ferguson, Klaassen, and Schmidt explain why AI is a contest between empires; Sharma explains why that contest rests on a financial foundation that looks too much like those of earlier bubbles. Together they draw a double risk: a war over one node, Taiwan, and a financial bust with that same node at its epicenter, in a world where states have little room to cushion the fall. The lesson they share is an old one: empires seldom fall for lack of ambition, but because they cannot pay for what they aspire to. What is new is the speed. And AI, unlike the railroad, offers no decades in which to learn how to pay for itself.
Methodological note. This text was prepared from a full reading of both articles. Figures and facts are attributed to the authors and were not independently verified. Readings that combine or extend the two texts (the Taiwan node coincidence, the dollar circle, margin compression, the selection hypothesis among empires, and the Andean reading) are inferences of this analysis and are flagged as such. Page numbers refer to the printed edition.
Sources
Ferguson, N., Klaassen, L., & Schmidt, E. (2026). Empires of AI: The Rise and Fall of a New Global Order. Foreign Affairs, 105(6), 62–78.
Sharma, R. (2026). The Global AI Bust: What Will Survive When the Boom Ends. Foreign Affairs, 105(6), 79–91.
No hay comentarios.:
Publicar un comentario