Context
This note follows our work on AI productivity evidence, the mid-cycle AI supercycle, and the June macro event cluster. The question here is narrower: does AI look more like a 1996-style productivity disinflation story, or a 1999-style expectations and CapEx cycle?
TL;DR
- AI today looks less like a clean 1996 productivity-disinflation regime and more like a 1999-style expectations and CapEx cycle with a 1996 productivity option embedded inside it.
- Workplace productivity evidence is real. Adoption is rising. But broad macro disinflation from AI has not yet been proven.
- AI data centers, chips, power, cooling, land, labor, and credit demand are already visible. These are current macro inputs for the Fed.
- The base case for the June FOMC is a hold, but not a dovish hold. It is more likely to be a hawkish hold or hawkish neutral outcome.
- The statement matters, but the key signals are the SEP dots, the longer-run neutral-rate dot, Chair Warsh’s comments on AI and r-star, and the July 8 minutes.
1996 vs 1999
The 1996 version of the story is positive supply. Productivity rises, unit labor costs slow, inflation pressure eases, and the Fed can be patient even with strong growth.
The 1999 version is different. The productivity story may be real, but asset prices, corporate investment, and capital-market expectations run ahead of realized diffusion. The Fed then has to watch demand, credit, valuation, and financial conditions, not only future productivity.
AI has both elements. The 1996 option is real. Brynjolfsson, Li, and Raymond’s Generative AI at Work finds a roughly 15% productivity gain for customer-support agents. Fed FEDS Notes shows AI adoption has moved beyond experimentation, and the San Francisco Fed argues policymakers need micro and sector evidence before the macro aggregates fully show the transformation.
But the immediate macro shock looks more like 1999. AI infrastructure spending is pulling on chips, power, data centers, cooling, land, skilled labor, and financing. The April FOMC minutes already treated AI as two-sided: possible productivity disinflation over time, but also AI-related investment that could raise input costs now.
Why This Matters for the June FOMC
The June FOMC is scheduled for June 16-17, with the decision at 2:00 p.m. ET and press conference at 2:30 p.m. ET on June 17. The minutes are scheduled for July 8. The current target range is 3.50-3.75%.
Recent data do not give the Fed an easy easing case:
| Indicator | Latest reading | Policy read |
|---|---|---|
| May CPI | +0.5% MoM, +4.2% YoY | Headline inflation uncomfortable |
| May core CPI | +0.2% MoM, +2.9% YoY | Better monthly core, still above target YoY |
| April PCE | +3.8% YoY | Fed’s preferred inflation gauge still high |
| April core PCE | +3.3% YoY | Core still too high |
| May payrolls | +172k | Labor market not weak |
| May unemployment | 4.3% | Stable |
The market already expects a hold. The real issue is what kind of hold.
Base Case
My base case is a hawkish hold.
| Item | Base case |
|---|---|
| Policy rate | Hold at 3.50-3.75% |
| Statement | Data dependent, inflation still elevated |
| Easing bias | Weakened or removed |
| 2026 dot | Moves toward no cuts in 2026 |
| Longer-run dot | Holds at 3.1% or edges higher |
| Press conference | AI productivity is possible, but not a current policy basis |
| Minutes | More hawkish than the statement |
The March SEP had a 2026 federal funds median of 3.4%, roughly implying one cut from the current midpoint. If the median shifts toward 3.625%, markets should read it as “no 2026 cuts.”
Scenario Map
| Scenario | Probability | What happens | Market read |
|---|---|---|---|
| Base: hawkish hold | 65% | Hold, weaker easing bias, higher 2026 dots | Short rates supported, dollar firm, equity multiples capped |
| Hawkish surprise | 25% | Some hike dots, longer-run dot up, r-star language | Front-end yields rise, risk assets volatile |
| Dovish surprise | 10% | One-cut median stays, energy shock treated as temporary | Yields fall, dollar eases, risk relief |
What to Watch
The key post-FOMC checklist:
| Signal | Dovish | Hawkish |
|---|---|---|
| 2026 median dot | Stays near 3.4% | Moves toward 3.625% |
| Longer-run dot | Stays 3.1% | Moves above 3.1% |
| AI language | Productivity and disinflation | CapEx, input costs, r-star |
| Warsh tone | Patience toward headline inflation | Credibility and elevated inflation |
| July 8 minutes | Productivity gains emphasized | Input costs, asset valuations, policy firming |
Market Implication
This is not a stock-picking note. At the macro level, the message is:
AI can raise long-run earnings potential
but if it also raises r-star and real rates
the same future earnings deserve a lower multiple
That is the tension. AI can be good for growth and still bad for rate-cut hopes.
Final View
AI may eventually become a 1996-style productivity disinflation force. But for this FOMC, the observable facts are closer to 1999: CapEx, power demand, input costs, financial conditions, and expectations are moving before economy-wide productivity disinflation is proven.
AI being good is not the same thing as rates going down. Until productivity shows up in broad unit labor costs and core services inflation, the Fed is more likely to treat AI as a neutral-rate and investment-demand issue than as a reason to ease.
Sources: Fed June calendar, Fed July calendar, April FOMC minutes, March SEP, Governor Cook on AI, BLS CPI, BLS jobs, BEA PCE, QJE Generative AI at Work, Fed FEDS Notes, Kansas City Fed, San Francisco Fed.