Gavekal: AI Capex Has Not Yet Cracked, but Kaletsky Sees a Profit Identity That Makes the US Uniquely Exposed When It Does
Gavekal Zoominar with Louis-Vincent Gave and Anatole Kaletsky, October 2026, on AI financing, central banks, France and Ukraine
The AI spending boom is still accelerating, but the cost of funding it is rising
Anatole Kaletsky pushed back on the premise that AI capital spending is weakening. "There has been no weakening in the AI spending boom. In fact, it's still continuing to accelerate," he said, noting that Mark Zuckerberg raised his spending commitments again the day before the call. What has changed is sentiment about how long this can last and who will pay for it. Kaletsky argued that if the build-out has to be financed at much wider spreads and with large equity issuance that dilutes existing holders, it cannot be sustained at current levels. He expects equity investors to keep rewarding upward profit revisions for a few more quarters, until actual investment data weakens.
Louis-Vincent Gave supplied the credit evidence. Hyperscaler debt issuance in 2026 has run at roughly $1.5 trillion, equivalent to about 70% of US Treasury issuance. Government bond yields have fallen around the world in recent weeks, but hyperscaler spreads have widened. Triple-C yields have jumped, and CDS protection on even Microsoft and Google has become more expensive. Gave described the Meta episode as telling. The shares initially rose 5% on the larger spending plan and a dilutive equity raise, then reversed as investors focused on the dilution. He also noted that semiconductor names such as SK Hynix and Samsung have charted "really ugly" patterns since July, while investors rotated into Apple, Microsoft, Meta and Google on the view that they win if AI succeeds and generate cash if spending is cut.
The Kaletsky-Levy profit identity is the core of the US bear case
Kaletsky's most substantive argument was that US corporate profits are structurally tied to the AI cycle in a way that profits elsewhere are not. He invoked the Kalecki-Levy profit identity, which he stressed is accounting rather than theory. Profits equal investment plus dividends plus government deficits minus household savings and the current account deficit. In his view, nearly every term on the right-hand side of that equation depends on the AI boom. AI investment feeds profits directly, buybacks and dividends are already coming down, and deficits cannot grow indefinitely. Record-low household savings are themselves a product of the equity wealth effect.
"Every element of the profitability of the US economy is dependent directly on the AI boom," Kaletsky said. He excluded only Korea and Taiwan from that claim, and concluded that even a slowdown in spending would hit US equities far harder than European, Japanese or Chinese ones. He also dismissed the comparison to the dot-com era as reassurance. By his figures, estimated AI investment from 2025 to 2032 equals 3.6% of US GDP, against 1.1% for the dot-com boom and almost twice the scale of the 19th-century railway boom. He added that Jensen Huang's railroad analogy "won't stop a lot of these companies going bust."
Hyperscalers look more like telecoms than software monopolies
Kaletsky argued that investors are extrapolating from the first 25 to 30 years of the technology sector, when these companies were capital-light and protected by network effects. AI, he said, is the opposite on both counts. Capital intensity is rising, and within a year there are three roughly equal, fiercely competitive models. "Microsoft is going to be like Vodafone," he said, referring to the telecom operators that briefly topped the market-cap rankings in the late 1990s.
Gave added two further doubts. The first is shelf life, since today's data centers will not last like railways, and only the broadband laid 25 years ago is still in service. The second is the Cold War argument for AI spending. If a leading model can be copied by China almost immediately, he said, it is hard to see the strategic win, and if it were kept strictly domestic, current valuations would make no sense. Gave also flagged political risk as a new variable. Data centers, which he said poll worse than cancer, are becoming a bipartisan issue as electricity, water and construction costs rise, and the Democrats are now campaigning against the rollout. Lobbying to overcome this would only raise costs and lower returns.
Kaletsky sees a fresh shot at the AI bull case if funding is constrained, but warns of hidden entanglements
In response to a client question, Kaletsky acknowledged a contrarian scenario. If compute constraints stop further development, the existing leaders could lock in moat-like profits, and he noted that OpenAI and Anthropic are lobbying for regulation partly for this reason. The risk is that much of the financing is already committed through circular arrangements whose legal details are opaque. He compared these to the structures that surprised markets in 2008 and said it is unclear how much of Nvidia's valuation, or of the Magnificent Seven's, is exposed if that financing fails.
Equity market drawdowns may owe more to oil and geopolitics than to AI
Kaletsky attributed the recent sharp equity drawdowns mainly to oil returning close to its Iran war peak, fears of military escalation after the US election and the Israeli election, and the earlier jump in bond yields. If those fears prove overdone, he sees a chance of a reversal and one more rally in tech. He said he turned bullish on bonds about two and a half weeks earlier, on the day yields peaked. He now sees a trading range of roughly 4.5% to 5.25%, with yields already in the lower half, so a correction in bonds is increasingly plausible.
Gave found it odd that oil keeps drifting higher despite signs of a Russia-Ukraine settlement, and said it suggests some participants expect action against Iran. Kaletsky countered that press reports of a planned US-Israeli strike may be designed to pressure Tehran. "If you're preparing a secret operation, why do you advertise it in the New York Times?"
US outperformance has been a war trade, not an AI trade
Both speakers challenged the US exceptionalism narrative. Gave noted that since the inauguration, US equities have lagged every major market except China, US bonds every major bond market except Japan, and the dollar every major currency except the euro. Kaletsky added that the low point of US relative performance was February 27, the day the Iran war began. Until then the US had trailed Japan, emerging markets and Europe by about 20 percentage points over 15 months. Since then it has outperformed all but Korea and Taiwan. His inference is that the US advantage came less from technology than from its relative insulation from an energy shock. If the war ends, he expects US relative performance to fade again.
Gave added that S&P technology and communication services together now make up half of the index, before counting industrial names like Caterpillar and Cummins that have been recast as AI plays. By that measure, he said, concentration has surpassed the late 1990s.
Central banks are unlikely to deliver the tightening priced in
Gave said markets still price three ECB hikes, which he finds hard to believe given France's turmoil and the AfD's lead in German polling alongside its anti-euro stance. He expects at most one Fed hike, in December, and thinks the Fed is probably done. Kaletsky said the market has already trimmed Fed pricing from three hikes to two, and he expects less than that. He pointed to the Federal Reserve Act's third mandate of moderate long-term interest rates, which he reads as an obligation to keep government financing affordable. He expects policymakers to tolerate inflation of 3% to 3.5% rather than 2%. Gave agreed that policymakers want that outcome as a way to manage debt, while warning that it is "like being a little bit pregnant."
Kaletsky added that any liquidity from a policy pivot is unlikely to reinflate AI after a burst. He expects it to favor cyclical and financial assets that investors have shunned, especially outside the US, if policymakers visibly act to prevent a recession.
France faces a choice between pension reform and default
Gave framed France's options starkly. It can raise the retirement age by five to seven years, or it can default, leave the euro and suffer a purchasing-power hit he put at about 50% of GDP. With taxes already near 44% of GDP, he said, higher taxes are not available. He noted that about 60% of French debt is held by foreigners, in contrast to Italy, which makes default politically tempting. Jean-Luc Mélenchon is running on that platform, though Gave called it a disaster that would bankrupt France's three systemic banks.
Gave expects the political center to collapse and Marine Le Pen to win a parliamentary majority, which would end the current three-bloc deadlock. His suggested route is a referendum on a five-year increase in the retirement age, which he believes would pass by two-thirds. He expects the ECB and Brussels to provide support after that. Kaletsky said the ECB would probably step in but would demand at least a symbolic concession in return.
A Ukraine settlement looks closer than at any point in years
Kaletsky said Ukraine appears to have accepted losing around 18% to 19% of its territory, not a third, and that the US has endorsed a settlement near current lines, which would imply sanctions relief for Russia. He said a rational Putin would take the deal. Pressure from Ukrainian strikes on Russian infrastructure helps, and both sides are exhausted. The main obstacle, in his view, is now Europe's rhetorical commitment, because French and German fiscal capacity to keep funding Ukraine is running out. Gave said a settlement would be bullish for Europe, China and India, since China and India have tightened policy this year to cope with the energy shock and could ease once Russian supply looks secure. He cautioned that morale at the front could collapse quickly once a deal is signaled.