Is Kevin Warsh Correct About AI’s Impact On Inflation And Interest Rates?

BCA Research argues AI is not disinflationary and may instead be inflationary, citing U.S. tech capex at 4.9% of GDP in Q1 2026, rising electricity and memory-chip costs, and a 2.6% U.S. saving rate. The report says CPI swaps imply inflation stays above the Fed’s 2% target for at least two more years and that AI could lift the equilibrium real rate rather than lower it. It also flags U.S. equity valuations as vulnerable, with MacroQuant putting the equity z-score at -0.69, near levels that have historically preceded bear markets.

Analysis

The market is underpricing the inflationary second-order effects of the AI buildout. The key mechanism is not abstract productivity; it is a real-resource squeeze: power, grid hardware, memory, and advanced packaging are becoming bottlenecks that transmit capex directly into consumer prices and corporate margins. That means the most exposed winners are the infrastructure enablers of the AI stack, while downstream hardware and discretionary tech hardware face a margin-tax from higher input costs and more elastic end-demand.

The bigger macro implication is that AI may be more reflexive than deflationary in the medium term. Rising capex, asset-price wealth effects, and lower household saving support nominal growth even if unit labor productivity improves, which is a negative setup for duration-sensitive assets if the market has to reprice the terminal rate upward. The cleanest tell is not headline CPI; it is whether inflation expectations and real yields stop easing despite softer goods prices elsewhere. If that happens, long-duration growth can de-rate even as AI capex remains strong.

The contrarian read is that the consensus is probably overconfident in a 1990s-style disinflation analog. The historical comparison only works if supply chains, commodity prices, and labor slack behave similarly, which they do not: today’s bottleneck is physical infrastructure, not excess manufactured capacity. The more plausible risk is a capex bust later, but over the next 6-18 months the skew is toward sticky inflation, firmer real rates, and a narrower set of AI beneficiaries rather than a broad equity beta tailwind.

The equity technical backdrop adds a separate warning: sentiment can stay elevated until it doesn’t, but when macro liquidity tightens, the crowded AI complex is vulnerable to a fast unwind. If breadth remains narrow and rates reprice higher, the first casualties are the most expensive, longest-duration names; the relative winners are cash-generative picks-and-shovels and utilities/grid beneficiaries with explicit AI demand exposure. That argues for expressing the view through relative value, not outright market shorts.

Key Decisions for Investors

• Long XLU / short QQQ for 3-6 months: play the divergence between AI power demand beneficiaries and duration-sensitive mega-cap tech; target 8-12% relative outperformance if real yields grind higher.
• Pair trade long VRT or ETN / short a basket of high-multiple AI software names for 6-9 months: monetize infrastructure bottlenecks while avoiding valuation compression in names with no near-term cash flow support.
• Buy put spreads on SMH or SOXX 2-4 months out: if memory prices and power costs keep rising, the semis complex can face simultaneous cost inflation and multiple compression; structure for defined risk and a 2:1+ payoff.
• Reduce exposure to long-duration growth proxies on any further rally in AI enthusiasm; add only on rate relief, not on capex headlines, because the macro transmission is currently inflationary rather than disinflationary.
• For a tactical hedge, own TLT puts or short-duration Treasury futures into the next CPI/NFP cycle: if the market starts pricing AI as a real-rate uplift story, duration is the cleanest macro short.