AI Has Entered the Fed’s Reaction Function

30-Second Signal

Warsh's Jackson Hole speech puts AI inside the macro framework and says more than half of this year's capex growth likely reflects the build-out; the July Monetary Policy Report independently shows 11% annualized first-quarter business fixed investment with AI infrastructure a major source of strength.

Jefferson's AI framework explains why demand can arrive before productivity, while the San Francisco Fed r-star study is the essential counter-evidence: identifiable AI news does not explain the observed post-2020 rise in r-star.

Vanguard's bond-market work estimates roughly $132B of 2026 year-to-date hyperscaler debt, while the BIS policy analysis says the AI boom is increasingly debt financed and that demand and supply effects arrive at different speeds.

The Treasury buyback announcement is a liquidity-support signal rather than QE; BEA's July PCE release records 3.7% headline inflation, BLS employment data show payrolls down 23,000, and Reuters on Barclays records the post-Warsh shift to two additional 2026 hikes.

Artificial intelligence has moved from an earnings story into the Federal Reserve’s reaction function. That does not mean AI has been proven to cause inflation or that it caused the rise in the U.S. neutral interest rate. The narrower and more defensible thesis is that the scale of the build-out is now large enough to alter investment demand, capital formation, electricity constraints, credit supply, real yields and productivity expectations at the same time. The macro question is therefore about sequence: does the capital-intensive demand shock arrive before the productivity dividend that is supposed to pay for it?

What SIAIntel Sees Now

Warsh’s intervention changed the policy map because AI is now large enough to be discussed alongside inflation breadth, financial conditions and the neutral rate. The key point is not a pre-committed September hike; it is that the investment boom has become a macro input while the productivity payoff is still uncertain.

Timing decides the sign

The same technology can be inflationary first and disinflationary later. Data centers, chips, power and specialist labor must be financed before the resulting capacity lowers unit costs. Jefferson’s two-sided framework makes the transmission clear: demand-first outcomes tighten the policy problem; productivity-first outcomes relax it.

r-star: plausible pressure, unproven causality

Higher expected returns on capital can raise desired investment and put upward pressure on r-star, but saving and supply effects can move the equilibrium rate the other way. The San Francisco Fed result is the discipline point: identifiable AI news does not explain the observed post-2020 rise in r-star. SIAIntel therefore treats AI as a new channel, not a proven cause.

Bond-market transmission

The financing mix is changing quickly. Hyperscaler and ecosystem issuance adds long-duration corporate supply while public borrowing remains large, but there is no mechanical dollar-for-dollar Treasury effect. Foreign demand, private credit and strong borrower cash flow can absorb supply. The correct monitor is the combined movement of real yields, spreads, issuance and capex.

Growth, inflation and labor no longer point one way

July PCE remained well above target while payrolls fell 23,000 and unemployment stayed at 4.1%. Barclays shifted to two additional 2026 hikes after Jackson Hole, but market pricing remains divided. That collision means the Fed is not choosing between a clean boom and a clean recession; it is calibrating policy inside a mixed regime.

Thesis vs Anti-Thesis

Channel	AI Rate Trap	Productivity Escape
AI capex	Raises current demand and capital needs	Expands future productive capacity
Productivity	Arrives with a lag	Arrives faster than expected
r-star	Upward pressure from investment demand	Neutral or lower if saving and supply dominate
Corporate debt	Increases competition for long-duration capital	Strong cash flow and broad demand absorb issuance
Inflation	Build-out bottlenecks keep pressure sticky	Unit costs fall as productivity broadens
Fed	Higher for longer or additional tightening	Hold, then easing as inflation falls
Long yields	Remain structurally elevated	Retreat as inflation and term risk ease
AI equities	Higher WACC compresses multiples	Earnings and productivity outrun the discount-rate shock

The anti-thesis is not cosmetic. A rapid productivity diffusion outside technology could expand capacity, reduce unit labor costs and cool inflation while investment remains strong. Strong hyperscaler cash flow could also absorb debt without a persistent credit premium. The San Francisco Fed finding is a direct reason not to treat “AI causes higher r-star” as settled evidence.

Scenario weights

These are SIAIntel editorial scenario weights, not statistical model probabilities. AI Rate Trap — 50%: sticky inflation, strong capex and financing, gradual productivity and tighter-for-longer policy. Productivity Escape — 35%: productivity broadens fast enough to cut unit costs and permit a hold followed by easing. Capital Break — 15%: yields or spreads rise enough to damage capex and employment, destroying the demand impulse that created the tightening risk.

What would falsify the thesis?

The AI Rate Trap weakens materially if inflation breadth normalizes quickly, AI-related issuance slows without credit stress, long real yields fall, measurable productivity accelerates across industries and the Fed can hold or ease without renewed inflation. It strengthens if capex and financing stay powerful while inflation breadth and long real yields remain elevated.

Dates that matter

The next high-information checkpoints are 4 September for U.S. employment, 11 September for CPI and 15–16 September for the FOMC meeting and updated projections. These releases can change the scenario weights before the policy decision.

Final assessment

Wall Street has spent years asking how much growth AI can create. Monetary policy now has to ask what price of capital the economy must bear while building the infrastructure required to deliver that growth. If investment demand arrives first, the AI Rate Trap becomes relevant; if productivity arrives first, it dissolves. The decisive variable is the time gap between investment and productivity.

SIAIntel Signal

AI is now inside the Fed’s reaction function, but the investable edge comes from tracking the sequence of capital demand, inflation breadth, real yields and productivity rather than declaring a one-way AI rate regime.