Roughly 45% of recent US growth traces to AI spending, and the top tenth of earners account for nearly half of consumer spending. The aggregate looks healthy; the median experience does not. Multi-speed growth is a positioning issue, not just a political one.
This is the second of four letters published together in October 2026. The first argued that rates are higher for longer. This one asks who is carrying the expansion that is happening anyway. The honest answer is: not everyone, and not evenly. Growth that looks solid in aggregate is leaning on two pillars — AI capital spending and high-income households — while the middle of the distribution is treading water.
US real GDP grew at 2.5% in the first quarter and 2.2% in the second. Carson Group estimates that roughly 45% of growth over the last five quarters came from AI-related investment — about 90 basis points per quarter of a roughly 2.1% total — and that figure leaves out data-center construction itself. On the household side, Moody's Analytics found that the top 10% of earners (those making above about $251,000 in 2024) accounted for 49.2% of all US consumer spending in the second quarter of 2025, up from about 46% in 2023 and 43% in 2020, while households in the 40th–60th income percentiles spent roughly $2.1 trillion, essentially flat. That is the latest version of this series we found; it is a 2025 figure and later data may differ. In 2026 the aggregate wealth numbers point the same way: household net worth reached about $186 trillion in the second quarter, about $26 trillion above its level at the end of 2024.
| Engine | Evidence | Direction |
|---|---|---|
| AI capital spending | ~45% of five-quarter growth (Carson Group); 2026 hyperscaler capex $688B–$740B | Carrying growth |
| Top-decile consumers | 49.2% of consumer spending, Q2 2025 (Moody's) | Rising share |
| Middle households (40th–60th pct.) | ~$2.1T spending, flat vs. 2023–24 | Stalled |
| Household net worth | ~$186T in Q2 2026, +$26T since Q4 2024 | Gains accrue to asset owners; the aggregate does not say who |
| Real wages | Wages +3.0% vs. CPI +3.4% | Falling |
| Sentiment | Michigan index 48.1 (Sept.) | Lowest in four months |
A multi-speed expansion is not just a fairness question; it changes where the risks sit. Both pillars are sensitive to financial conditions. AI capex depends on funding costs and on investors' willingness to keep financing a build-out that has already outrun its early-2026 forecasts (roughly $470 billion then, $688–$740 billion now). Top-decile spending depends on asset prices: it is a wealth effect, and wealth effects run in both directions. Meanwhile the broad base of households has less cushion to absorb a shock — real wages are falling, mortgage rates are 7.28%, and surveys show many middle- and lower-income households leaning on discounts and buy-now-pay-later financing. A Dallas Fed research note published in late 2025 made the related point that rising consumption concentration may add slightly to economic fragility.
The practical consequence is a narrower base than the headline suggests. If asset prices stay firm and AI spending continues, the aggregate numbers can keep looking healthy even as the median experience stays strained. If either pillar weakens, there is less in the middle to take over. That asymmetry — a stable-looking aggregate resting on a small number of engines — is the thing worth positioning for.
Not every indicator fits the K-shape story. First-time buyers made up 30% of existing-home purchases in August, up from 28% a year earlier, and employment — until September — was holding. Spending-share figures are estimates, the headline Moody's number is from 2025, and some analysts argue the “K-shaped economy” narrative is overstated or driven by media framing. The multi-speed description is best read as a distributional tilt, not a clean split of the population into winners and losers.
An aggregate number can be perfectly accurate and still describe an economy that almost nobody experiences as average.
This letter's central claim — that growth is leaning on a narrow base — would be wrong if wages re-accelerate, if energy prices fall enough to lift real incomes, or if AI-driven productivity gains begin to show up broadly in pay. It would also be weakened by a labour market that stays firm. The September payroll print of 29,000 points the other way, but it is a single month. The wealth-effect channel is also harder to measure than it sounds; the $186 trillion aggregate cannot tell us how much is held by households that spend from it.
The expansion is real, but it is built on two engines and a thin base. That does not argue for pessimism; it argues for knowing which engine your portfolio depends on. Favour businesses whose revenue does not require both the AI cycle and rising asset prices to keep working, watch low-income credit for the first signs of strain, and treat any “consumer is fine” headline as a statement about the top of the distribution until the data say otherwise.