Dwarkesh Patel: the AI buildout could set off a second Volcker shock

Dwarkesh Patel has sketched the scenario in which the AI buildout, not a central bank, causes the next wave of sovereign defaults. The argument runs through his latest podcast with SemiAnalysis founder Dylan Patel — a full chapter of the episode is titled "Will AI cause a sovereign debt crisis?" — and he then wrote it up on X as a standalone case. The mechanism is boring and that is the point: if compute earns extraordinary returns, capital stops flowing anywhere else.

He starts from the SpaceX–Google arrangement, where he says "the entire capex for the data center can be paid off by leasing it out for roughly 1 year." That deal, disclosed in a SpaceX regulatory filing, has Google paying $920 million a month from October 2026 through June 2029 for roughly 110,000 Nvidia GPUs and associated hardware — about $30 billion over the term. His extrapolation: as models approach AGI, demand for compute to serve them becomes "as large as the demand for white collar labor — aka 10s of trillions of dollars a year." In that world, he asks, "why would anyone invest in anything but datacenters/ semiconductors/ energy/ robotics?"

#The crowding-out arithmetic

Dylan Patel supplied the financing numbers. SemiAnalysis models roughly $11 trillion of capex from 2024 to 2029, of which about $6 trillion is funded from cash flow and $5 trillion has to be raised as credit. Five trillion dollars of new issuance moves the price of money for everyone. His illustration: Meta has recently raised debt at 5–6%, and "I don't see why they wouldn't pay 8%" — a 250 basis point move that reprices banks, telecoms, consumer-goods companies and mortgage borrowers alike.

Dwarkesh's version is blunter, and put in an emerging-market lender's voice: "Hey, why am I going to lend Egypt money at 5% when I could just buy relatively safer investment-grade hyperscaler debt in America at 10%?" The feedback loop is what makes it a crisis rather than a repricing — higher rates make default more likely, lenders demand more yield for that risk, and the higher yield accelerates the default.

#A second Volcker shock

The framing comes from economist Basil Halperin, an assistant professor at UVA and a co-author of the paper Transformative AI, existential risk, and real interest rates. "Higher US interest rates from the Volcker Shock in the 1980s spilled over globally to help cause the LatAm debt crises," Halperin wrote. "As AI keeps pushing up real interest rates — US to globally — it seems totally plausible we get a 'second Volcker shock' of sovereign debt crises." His added worry is asymmetry: developing countries feel globally higher rates because capital markets are integrated, but capture less of the AI-induced growth. His practical suggestion for an "AGI-pilled sovereign" is to switch to issuing long-term debt now and lock in today's rates.

One discrepancy worth flagging: the written post puts Volcker's 1980 tightening at "about 10 pp," while on the podcast Dwarkesh called it more than 5%, "something like 8% real." Both cite "some 40 odd countries, mostly in Latin America" defaulting that decade.

#The barbell, and why the US may be fine

Rising discount rates hit equities valued on distant, stable cash flows hardest. "The market overall may be up because of AI stocks, but almost every other stock will be down," Dwarkesh writes. Dylan's example: "Why would I pay this much for Johnson & Johnson?" — or a railway — when the discount rate is 8–10% instead of 3–5%.

On US fiscal exposure, Dwarkesh notes that corporate income is under 10% of federal revenue while payroll and income taxes are 80%-plus — exactly the base automation erodes. About 20% of tax revenue currently services the debt; a 1 pp rise takes that toward 25% over a five-year rollover, 5 pp takes it north of 40%, and the government's ~$2 trillion of annual new borrowing pushes it north of 60%. He still thinks the US "will be mostly fine," because the data centers sit on American soil and Congress can invent new taxes to capture token revenue. Countries with heavy debt, thin tax bases and short maturities — he names Pakistan and Nigeria — have no equivalent option.

This is a conditional scenario, not a forecast or a reported crisis: every step depends on AI delivering the returns assumed at the top. Dwarkesh says so himself, closing on the upside — "if the economy is growing tens of percent a year, this would also be a world of great abundance" — and reducing the argument to one line: the opportunity cost of capital rises so far that "you're paying a ton of opportunity cost to give people pensions now rather than building a new datacenter."