Morgan Stanley's 4%-for-1% cushion prices one lost wage dollar at 24 dollars of paper gain

The wealth doing the cushioning is recent and narrow: roughly $21tn added to American household balance sheets between the first quarter of 2024 and the first quarter of 2026, half of it from direct equity holdings that make up about 30% of total wealth ^[5]. Morgan Stanley's rule for what that buys is a top-earning household needing its portfolio up 4% to offset a 1% fall in labor income ^[10]. Applying a household rule to the quintile aggregates the bank published, 1% of the top quintile's $8.3tn of labor income is about $83bn and 4% of its $49.8tn of equity is about $1.99tn ^[9], which prices one lost wage dollar at roughly 24 dollars of paper gain ^[20]. What that number measures is leakage from wealth into spending, not buffer depth, and further down the distribution the leakage runs the other way. The bottom 40% hold $1.7tn of equity against $1.8tn of labor income ^[8], so dollar for dollar a 1% wage cut needs a 1.06% market gain ^[21], and at the same 24-to-1 pass-through it needs something nearer 25% ^[22]. No rally arrives on that schedule.

The wage side is the measured side. It is thinner than its headline. Sania Edlich and Torsten Slok matched Anthropic's usage-log index against Bureau of Labor Statistics wages across 321 occupations from 2015 to 2025, splitting at 2023 ^[11], and only 11 occupations cleared their high-exposure bar ^[12], which is 3.4% of the sample ^[23]. Both the aggregate wage gap and the quartile gradient rest on those eleven ^[13][14]. If the logs proxy Claude's adoption pattern rather than an occupation's exposure to the technology, the ranking moves and the penalty moves with it, which is the test Apollo's own design invites.

The pricing mechanism, meanwhile, is the load-bearing part of the political story, and it is also the unmeasured part. Fortune presents it without a margin series, a price index or a productivity figure ^[26], and neither the Morgan Stanley nor the Apollo work measures siting opposition or electricity prices ^[27]. The line from flat raises to angry county hearings is an inference, resting on no dataset that actually connects the two.

My read, or rather the version worth arguing with: the constituency arithmetic points away from the class story. Brookings' Mark Muro found 62 of the 100 most AI-exposed counties were primarily urban and leaned left in 2024 ^[17], and nearly 70% of workers in high-exposure jobs hold a bachelor's degree against about 10% in low-exposure work ^[3], which describes a population that mostly does not live where cheap land and interconnect queues put the buildings. The counter-thesis is duller and may be correct: bills. If opposition tracks what shows up on a utility statement rather than what did not show up in a raise, then siting compute is a rate case with extra steps, and the class grievance is decoration on it.