The U.S. economy grew 2.2 percent in 2025. The number is accurate. It is also a compression of three economies that have almost nothing in common.
Bank of America's consumer spending data shows something that GDP does not. The economy is not K-shaped anymore — two populations, one rising, one falling. It is E-shaped. Three populations. The top tier's spending grew 2.5 percent year-over-year in January. The middle tier grew roughly one percent. The bottom tier grew 0.3 percent. These are not three points on a continuum. They are three different economies sharing a currency.
The aggregate GDP number compresses all three into a single figure. In the fourth quarter of 2025, the economy grew 1.4 percent annualized. That number does not tell you that the top twenty percent of earners account for fifty-seven percent of all consumption, according to Dallas Fed research — up from fifty-three percent. It does not tell you that Moody's Analytics found the top ten percent now drive 49.2 percent of consumer spending, up from thirty-six percent three decades ago. It does not tell you that the New York Fed found lower-income households' inflation-adjusted spending barely recovered to January 2023 levels by the end of 2024, while college-educated households had boosted theirs by four percent.
The number tells you the economy is growing. It does not tell you who.
The Three Economies
Deloitte's consumer products leader said what the aggregate conceals: "The bottom half of the economy is already in recession to some extent." The BofA data makes the mechanism visible. The bottom tier carries credit card balances month to month — fifty-nine percent of households earning between twenty-five and forty-nine thousand dollars — and relies increasingly on Buy Now, Pay Later financing. The middle tier buys in bulk, trades down on brands, and what CNBC called "spending in a nervous way." The top tier's behavior barely changed.
Personal consumption expenditures represent 67.9 percent of GDP. When one population accounts for half of all consumer spending and that population's behavior diverges from the other two by a factor of eight, the aggregate is not summarizing. It is averaging over a discontinuity.
This is what it looks like when a measurement system crosses from useful compression to confabulation. The number is produced by the same methodology. The methodology was designed for an economy where the spread between top and bottom spending growth was narrow enough that averaging was faithful. When the spread was two or three percentage points, the aggregate captured the signal. When the spread is 2.5 to 0.3 — an eight-to-one ratio — the aggregate captures nothing. It produces a confident, low-variance number that sounds like understanding and functions like a mask.
The Load-Bearing Column
The top tier's spending depends on wealth. The top twenty percent hold seventy-one percent of aggregate wealth, up from sixty-one percent in the 1990s. That wealth is concentrated in equities. Those equities are concentrated in technology. And that technology thesis rests on a single claim: that artificial intelligence will generate returns.
The Magnificent Seven plan to spend roughly $650 billion on AI infrastructure in 2026. Amazon alone guided $200 billion in capital expenditure. Alphabet guided $175 to $185 billion. Meta guided $115 to $135 billion. This is not diversified investment across the economy. It is a concentrated bet by a handful of companies whose combined market capitalization represents approximately thirty percent of the S&P 500.
MIT's 2025 report found that ninety-five percent of organizations are getting zero return from their AI projects. S&P Global's survey found that forty-two percent of companies abandoned most of their AI initiatives in 2025, up from seventeen percent in 2024. Stanford SIEPR researchers have called artificial intelligence "the third bubble of the century" and warned that a drop in AI-related spending could push the economy into recession.
Follow the dependency chain. GDP depends on consumer spending. Consumer spending depends on the top tier. The top tier's wealth depends on equity prices. Equity prices depend on AI returns. AI returns have not materialized for ninety-five percent of projects. The entire structure is a column, not a pyramid — and the column rests on the thinnest base.
The Detection Problem
Standard recession indicators were calibrated for a more homogeneous economy. The Sahm Rule triggers when unemployment rises by half a percentage point from its low. But unemployment rose to 4.4 percent in February — and the BLS simultaneously applied a population control adjustment that makes the number unreliable as a time series. The Conference Board's Leading Economic Index tracks ten components, none of which resolve consumer behavior by income tier. The Atlanta Fed's GDPNow model dropped from 3.0 to 2.1 percent in a single update — a legitimate signal, but one that still treats the economy as a single entity.
The February jobs report showed the economy lost ninety-two thousand positions. Average hourly earnings rose 0.40 percent. Fewer people working, each earning more. That combination is consistent with a single economy adjusting. It is also consistent with three economies — one shedding low-wage jobs, one treading water, one bidding up scarce skilled labor — producing a composite that looks like orderly transition.
The aggregate cannot distinguish between these interpretations. It was not designed to. GDP was invented in 1934 by Simon Kuznets, who warned Congress in his first report that "the welfare of a nation can scarcely be inferred from a measurement of national income." Ninety-two years later, the measurement is more precise and the warning is more relevant. The economy has diverged into three populations with different growth rates, different risk exposures, different behavioral regimes, and different relationships to the AI thesis that supports asset prices. The single number that claims to represent all three is the most consequential confabulation in macroeconomics.
If the AI column holds, the top tier's spending sustains the aggregate, and GDP continues to report growth. If it does not, the top tier contracts, the forty-nine percent of consumer spending it drives contracts with it, and the aggregate reveals what was true all along — that two-thirds of the economy was already stalling while the number said otherwise.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)