One Economy, Two Speeds: What Divergent Sectors Do to Aggregate Statistics

01 The Question of a Divided Economy

Can an economy simultaneously experience an AI investment boom and industrial stagnation? The question sounds like a paradox only if one holds an implicit model of the economy as a single thing — a body with one temperature, one pulse. That model is convenient for headline writers and for some macroeconomic policy frameworks, but it is analytically wrong, and the costs of being wrong about it are not symmetric. If the economy genuinely moves at two speeds — a capital-intensive technology sector expanding rapidly while physical, rate-sensitive sectors flatten — then aggregate statistics will systematically mislead, monetary policy will be calibrated against an average that describes neither sector, and the political economy of the expansion will carry internal tensions that a unified boom would not.

The sectoral frame treats this not as a curiosity but as the central analytical object. Sectoral analysis, as the reference record notes, is a statistical analysis of the size, pricing, competitive, and other economic dimensions of a sector of the economy, an approach that Wynne Godley developed further for macroeconomic analysis of national economies (source: Wikipedia summary — Sectoral analysis). The Godley lineage matters here: it insists that the macroeconomy is built up from sectoral balances and flows, not from a representative agent. When one sector absorbs a disproportionate share of new investment while other sectors contract or idle, the interesting action is in the flows between them — capital flows, labor reallocation, and demand spillovers — not in the average.

The framing question is precise: two-speed economy mechanics (capital flows toward AI while physical sectors face both higher rates and weak demand); the measurement problem, in which aggregate statistics mask the divergence; and the historical record of two-speed episodes, which offers both warnings and reassurances. Each of these is examined in turn below, with the standing discipline that observed facts, reported claims, causal inferences, and scenarios are labeled as such. Where a figure is illustrative rather than measured, it is marked illustrative.

02 The Mechanics of a Two-Speed Economy

The mechanism starts with investment allocation. In macroeconomics, investment consists of additions to the nation's capital stock of buildings, equipment, software, and inventories during a year — spending on productive physical capital such as machinery and construction of buildings, and on changes to inventories (source: Wikipedia summary — Investment (macroeconomics)). This is the channel through which the divergence is produced. Capital is not a homogeneous fluid that automatically spreads across the economy; it is allocated by expected risk-adjusted returns, and when one class of assets — in this cycle, AI compute, data centers, and the upstream supply chains that feed them — offers a compelling expected-return narrative, capital crowds toward it. The observed outcome is a rapid expansion of investment in one narrow set of activities alongside stagnant or declining investment in the rest.

The second mechanic is the interest-rate channel, and it operates asymmetrically across sectors. Higher policy rates raise the cost of capital for activities that must be financed — construction, manufacturing expansion, inventory building, housing — while barely touching activities that are funded by equity cash flows, accumulated reserves, or the particular alchemy of technology investment, in which expected future growth, not current borrowing cost, dominates the hurdle-rate calculation. A single rate environment is therefore experienced as two different environments: a tight one for rate-sensitive physical sectors facing weak demand, and a permissive one for a technology sector whose investors are pricing a transformation. This asymmetry is the engine of the divergence, and it means the divergence can widen without any sector behaving irrationally.

The third mechanic is labor reallocation with friction. Workers are not instantly mobile across sectors: the skills demanded by data-center construction and AI-adjacent services differ from those released by industrial stagnation, and retraining, relocation, and matching all take time and impose costs on the workers bearing them. The aggregate unemployment rate can therefore look healthy while both sides of the economy complain — one side about labor scarcity and wage pressure in specialized occupations, the other about layoffs and hiring freezes. Both complaints are accurate at their own level of aggregation. Only the sectoral view can reconcile them.

03 The Measurement Problem: What Averages Conceal

The measurement issue is the analytically hardest part, and it deserves a PhD-level treatment rather than a shrug. Aggregate statistics — GDP growth, aggregate investment, the unemployment rate, average productivity — are weighted averages, and weighted averages of divergent components inherit a specific failure mode: they describe a composite economy that matches no actual sector. If the AI complex contributes strong growth in investment and hours while being a small share of the total, and the much larger physical sectors contribute stagnation, the aggregate will show modest, healthy-looking growth. That average is simultaneously true at the aggregate level and useless for describing the situation of either sector.

Three specific measurement traps follow. First, composition effects in productivity: as the reference record notes, productivity measures are ratios of aggregate output to aggregate input, and the key difference between measures is how outputs and inputs are aggregated (source: Wikipedia summary — Productivity, in the wave's research base for the AI productivity debate). A rapidly growing high-productivity sector can raise measured aggregate productivity even while the median firm experiences no productivity gain at all — the average rises because the weights shift, not because anything improved for incumbents. Second, price deflator problems: quality-adjusted price indices in the technology sector are notoriously difficult, and mismeasurement there propagates into real investment and real output aggregates. Third, the timing problem: investment in structures and equipment enters the capital stock when it is built, but its measured productivity contribution arrives later, so a construction-heavy boom can initially depress measured productivity ratios — inputs rise before outputs follow.

These measurement problems are not academic. They feed directly into policy, because a central bank calibrating against aggregate indicators will observe a healthy average and may maintain a tighter stance than either sector would individually justify: tight for the physical sectors, which bear the full cost, and arguably still permissive for the AI complex, which is not rate-financed in the first place. That policy configuration — restrictive for the majority of the economy and irrelevant to the part that is booming — is a distinct macroeconomic condition, and the aggregate framework has no vocabulary for it.

04 Transmission: How the Two Speeds Interact

The two speeds are not sealed off from each other, and the transmission between them determines whether the divergence is self-limiting or self-amplifying. Three channels deserve analysis. The demand channel: the booming sector's expansion — construction of data centers, hiring, upstream orders for chips and electrical equipment — is itself demand for the physical economy. Some of the stagnating sectors' weakness is partially offset by supplying the boom: concrete, steel, transformers, electrical trades. The offset is real but narrow, concentrated in specific materials and regions, and it does not rescue sectors whose demand problem is consumer-facing, such as housing or consumer discretionary goods.

The financial channel: the boom absorbs capital. When capital flows preferentially into one complex, the cost and availability of capital for everything else changes at the margin — through issuance windows, through the allocation decisions of lenders and funds whose mandates span both, and through the willingness of boards to fund expansion projects that now look unexciting next to the AI narrative. This is the crowding-out channel, operating not through a mechanical interest-rate identity but through the relative attractiveness of investment opportunities. The third channel is political-economic: a boom concentrated in one sector and one set of regions generates distributional grievances that surface in policy — tax proposals, regulatory attention, and demands that the booming sector subsidize the rest — which inject uncertainty precisely into the sector carrying the investment cycle.

Second-order effects follow. If the boom continues, physical-sector firms may underinvest in capacity for years, which would surface later as a supply problem: when demand eventually normalizes, the capital stock to serve it has thinned. If instead the boom falters, the physical sectors inherit a labor market suddenly flooded with specialized workers whose skills do not match their industries, plus stranded construction and disappointed suppliers. The two-speed economy thus concentrates systemic risk in the narrow sector, a third-order effect that aggregate statistics again will not reveal until it happens.

05 Historical Precedent: Two-Speed Episodes Before This One

Historical precedent suggests two-speed episodes are recurring, not novel. The canonical case is the late-1990s United States: technology and telecom investment boomed while manufacturing, buffeted by a strong dollar and import competition, stagnated — the phrase "two-speed economy" was used then about conditions in the United Kingdom, where financial services and London property pulled away from industrial regions. The relevance of the dot-com comparison for this article's framing is treated in detail in the companion analysis of the railroad-versus-dot-com question; the point here is narrower. In the late-1990s episode, the divergence ended when the booming sector's investment outran the cash flows it could plausibly generate, and the aggregate statistics — which had flattered the period — then overshot downward in the bust, because the weighted average fell with the collapsing weight. The measurement problem operates symmetrically: averages masked the boom's concentration on the way up and masked the bust's concentration on the way down.

What is similar and what is different. Similar: a rate environment that is experienced differently by an equity-funded booming sector and a debt-funded rest of the economy; a weighted-average statistical apparatus describing a composite economy matching no sector; a policy framework debating whether a supply-side transformation justifies a looser stance than aggregate inflation indicators suggest. Different, and this matters: the physical sectors in the earlier episodes were not simultaneously facing the specific combination this cycle presents — a higher-rate regime and weak goods demand together with a boom pulling capital and skilled labor toward itself. The direction of the difference cuts both ways. The 1990s boom was lighter in physical capital; the current one is construction-heavy, which paradoxically links the two speeds more tightly — a bust in the AI complex now transmits faster into the physical economy than a dot-com bust did, because construction and equipment suppliers have counted the boom as demand.

06 Scenarios: Three Paths for a Two-Speed Economy

N43 offers three scenarios, each with triggers, mechanisms, and observable indicators. Probabilities are not assigned beyond qualitative ordering, consistent with the no-unsupported-numbers discipline.

Scenario A — Convergence upward. The boom's demand spillovers broaden: the AI complex's construction and supply-chain needs spread income into the physical sectors, goods demand recovers, and the two speeds synchronize at a higher level. Trigger: sustained evidence that physical-sector orders and hiring are rising alongside the boom. Mechanism: the demand channel dominating the crowding-out channel. Indicators: widening of the set of industries with rising hours worked; a narrowing gap between sectoral investment growth rates; capacity utilization in materials industries rising without a price shock. Consequence: a genuine, broad expansion, and the measurement problem becomes temporarily less severe because the components stop diverging.

Scenario B — Persistent divergence. The two speeds continue at different velocities. This is the baseline continuation: the boom persists on its own logic, the physical sectors persist at stall speed, and aggregates continue to describe a composite that matches nothing. Trigger: nothing dramatic — simply the absence of either a boom-break or a demand recovery. Mechanism: the three channels (demand, financial, political-economic) roughly offsetting. Indicators: continued wide spread between sectoral investment growth; the unemployment rate stable while both sectors complain about labor; recurring policy debate about a single rate stance fitting two economies. Consequence: policy calibrated to an average, distributional grievance accumulating, and the fragility concentrated in the boom continuing to build beneath a calm aggregate surface.

Scenario C — Boom break. The AI investment cycle disappoints — the expected returns fail to materialize on the timeline investors priced — and capital flows reverse. Because the boom has been the marginal source of demand growth, the physical sectors do not escape the shock; they inherit it. Trigger: evidence that the AI complex's cash-flow expectations are being marked down materially. Mechanism: financial channel running in reverse, demand channel collapsing, labor reallocation reversing into a soft market. Indicators: cuts in announced capital expenditure; a widening of high-yield credit spreads in technology-adjacent issuers; a stall in data-center construction employment. Consequence: a synchronized downturn arriving from a narrow sector — the two speeds converging downward, the reverse of Scenario A.

07 Indicators: Watching a Two-Speed Economy Properly

A two-speed economy must be watched with sectoral instruments, not aggregate ones alone. The indicator set, each with its diagnostic purpose. First, the spread between sectoral investment growth rates — the direct measure of divergence; a narrowing spread is evidence for Scenario A, a stable wide spread for B. Second, the spread between sectoral hours worked and hiring — the labor-market twin of the investment spread. Third, construction employment in data-center-building trades and regions versus total construction employment: because the current boom is construction-heavy, this isolates the boom's physical footprint. Fourth, capacity utilization in materials and equipment industries: rising utilization without a broad goods recovery means the boom's demand channel is working; falling utilization means the stagnation is deepening. Fifth, issuance and credit spreads by sector: debt-financed physical sectors facing tight conditions will show it in financing costs before it appears in employment. Sixth, the aggregate-versus-median gap in firm-level outcomes — the gap between the weighted average and the typical firm is the cleanest single measure of the measurement problem itself. Seventh, policy communication: when central-bank language begins explicitly discussing sectoral divergences, that is institutional recognition that the aggregate frame has failed. Eighth, political attention: legislative and regulatory proposals targeting the booming sector signal the political-economic channel activating, which injects its own uncertainty into the boom.

08 The Bottom Line

What we know: The question — whether an economy can simultaneously run an AI investment boom and industrial stagnation — is mechanically answerable: yes, because investment allocation, the interest-rate channel's asymmetric incidence, and labor-market friction all permit sustained divergence, and the sectoral analytical tradition (source: Wikipedia summary — Sectoral analysis) exists precisely because aggregates can mislead.

What we think we know: Aggregate statistics are currently describing a composite economy that matches neither sector, and policy calibrated to that composite will be simultaneously too tight for the physical sectors and largely irrelevant to the boom. Historical two-speed episodes suggest the divergence resolves through the booming sector's own dynamics — convergence upward if spillovers broaden, a synchronized downturn if the boom breaks — rather than through the stagnating sectors' spontaneous recovery.

What we do not know: How wide the current divergence actually is, because the sectoral decomposition of the current investment cycle is uncertain; whether the construction-heavy character of this boom makes it more integrated with — and more dangerous to — the physical economy than earlier two-speed episodes; and how long a policy framework can persist while describing an economy that no longer exists at the sectoral level.

What to watch next: The investment-growth spread and its direction of travel; construction employment inside versus outside the boom's footprint; capacity utilization in the boom's supplier industries; the aggregate-versus-median gap in firm outcomes; and the first sustained appearance of sectoral language in monetary-policy communication — the signal that the institution has noticed what the averages were hiding.