Diffusion - The Economic Ledger (B2, B3, B5, B7)#
Contents
Whether the technology shows up in the numbers firms and statisticians keep. The four indicators here share a design principle: each is built to survive the J-curve, reading dispersion, pricing structure, and relative moves rather than the aggregates that history says stay silent for a decade.
B2 - Measured productivity#
| Baseline | No clear AI signal in aggregate TFP |
| Trajectory assumed | J-curve; signal emerges 2029–32, not before |
| Trigger - fast diffusion | Sustained US TFP growth >2%/yr before 2029 |
| Trigger - slow diffusion | No signal by 2033 |
| Revises | 2028–2032, Part V row 2 |
Expect this to say nothing for years. That silence is the base case, not a surprise - every general-purpose technology has shown the same lag, and reading early flatness as failure is the most common forecasting error in this literature. → class 1
How to read without fooling yourself. Prefer multi-year averages over single prints; prefer sector TFP in exposed industries (software publishing, business services, wholesale trade) as leading series that can move before the aggregate; treat quality-adjustment revisions as first-class events, because the BEA's deflators for software and cloud services are exactly where AI output is most likely to be mismeasured. A single strong quarter is not the trigger. Sustained >2%/yr means the J-curve has bent, not that a residual got lucky.
B3 - Firm-level margin dispersion#
| Baseline | No large systematic margin lift at median adopters |
| Trigger - Red Queen confirmed | Adoption rises, prices fall, margins flat |
| Trigger - Red Queen broken | Sustained margin expansion at adopters in competitive industries |
| Revises | Game 3 - the surplus-capture argument |
This is the cleanest available test of the document's most contrarian claim. Game 3 predicts adoption without profit; the naive view predicts both. Watch dispersion, not the mean - if the gains accrue to the scarce complements, the mean stays flat while the top decile pulls away, which looks like nothing happening in aggregate data.
Practical series: within-industry operating-margin interquartile range for competitive tradable sectors; gross margin vs. SG&A for software firms that report AI adoption; and the gap between "AI spend" survey series and earnings. The last gap is diagnostic of over-adoption dissipation - boards spending under competitive fear without a measured return.
B5 - Outcome-priced contracting#
| Baseline | Overwhelmingly seat- and token-priced |
| Trigger | Material revenue share priced on delivered outcomes with liability attached |
| Revises | Capital revenue case; Law; Game 3 |
A sleeper indicator. Vendors will only accept outcome pricing when they believe reliability is real, and buyers will only pay it when they can attribute results. The pricing model is a revealed belief about reliability - more honest than any benchmark, because someone is betting on it.
What counts. A contract that pays on resolved tickets, closed cases, or measured cost-down with a liability or indemnity clause attached - not "success fees" that are marketing language around seat licenses. Token and seat pricing can coexist for a long time; the trigger is a material share of vendor revenue, not the first pilot. This is also the cleanest external-demand signal for the capital circularity problem: outcome revenue cannot be an echo of the capex loop the way cloud credits can.
B7 - The two-economy price split#
| Baseline | Cognitive services soft or deflating; energy, shelter, care, trades firm or rising |
| Trigger - claim confirmed | Sustained divergence: CPI/PCE components for software/info services fall or lag while energy+shelter+care outpace for 8+ quarters |
| Trigger - claim fails | Uniform inflation or deflation across both baskets |
| Revises | Prices, 2028–2032, Assets |
This is the macro expression of Game 3. Prefer relative component moves over headline CPI. Headline can look fine while both halves of the claim are true.
Basket construction. Cognitive basket: software, information services, selected professional services where deflators exist. Physical/care basket: electricity and gas, owners' equivalent rent / rent of primary residence, medical care services, and repair/maintenance where available. Eight quarters of relative divergence is the bar so a single energy spike or software sale season cannot fire the claim. If Uncertainty 7 runs, the split narrows from the late 2030s as the inflating categories get manufactured - so a confirmed split that later closes is not a contradiction; it is a dated half-life.
Joint readings#
| Pattern | Reading |
|---|---|
| B5 fires, B2 silent | Reliability believed by counterparties before statisticians measure it - still the base case early |
| B3 Red Queen confirmed, B7 diverging | Surplus leaving competitive firms into complements and consumers - spine second half holding |
| B2 fires early, B3 margins expand in competitive sectors | Either reorganization accelerated or the Red Queen is wrong - re-score row 2 and Game 3 together |
| B7 uniform deflation | Complements eroding or demand collapsed; check U7 vs. recession before rewriting prices |
| B3 margins expand only in licensed/concentrated sectors | Scope condition of Game 3 holding - not Red Queen broken |
Circularity filter for B5/A2. Outcome revenue can still be partly circular if the "customer" is an equity affiliate. Prefer third-party enterprise contracts with public indemnity language over cloud credits between related parties. → Capital
Related: Game 3 · Prices · Capital · Labor and institutions