Bulls Read Growth. Bears Read a Sugar High. The Receipts Name Who Writes the Bill.

S&P 500 per-share earnings rose 53% year-over-year in the second quarter. Sales climbed nearly 16%. The 2-to-1 ratio of guidance raises to cuts in the index marks a structural turnaround from a year earlier. Strip out the one-time equity-stake gains at Amazon and Alphabet that inflated the index headline, and the underlying gain is still the largest since fall 2021. That is what the print says.

Here is what is in it.

The Wall Street Journal’s Monday report identified four forces driving the surge: rapid growth in AI spending, federal spending, refunds of some past tariff payments, and a buoyant equity market supporting continued consumption. Each operates on a different cycle and hits different sectors. The breadth is what makes the print durable; over 40 firms have now reported $9.6 billion in tariff refunds, the disclosure pattern that put tariffs at the center of the conversation mid-summer, and Apollo Global Management’s mid-August estimate put the refund-driven contribution at more than 4% of third-quarter GDP — roughly 0.2 percentage points on top of the Atlanta Fed’s 4-to-5% growth nowcast.

The bull reading starts at the register. Abercrombie & Fitch captured $120 million in tariff refunds and raised its full-year estimates; the retailer also pared back discounts in the most recent quarter, raising effective prices, and customers kept buying. Chief Financial Officer Robert Ball called the underlying business “above our expectations.” With prices higher and the refund money on the bottom line rather than at the register, the consumer still transacted. Garmin’s $21 million in tariff refunds lifted profit margins on top of strong fitness-product demand; Chief Executive Clifton Pemble called the gross margin “impressive by any historical comparison” even excluding the refund. The breadth of catalysts — corporate demand holding at higher prices — is what makes the 2-to-1 guidance spread structural rather than a one-quarter pop.

The bear reading rests on the same receipts, read at the institutional level. The Journal’s finding was categorical: at most of the firms reporting tariff refunds, the money flowed primarily to the bottom line rather than into lower prices for shoppers. The refund is corporate margin, not consumer relief. Walmart — which received roughly $2.9 billion in tariff refunds and applied some of it to price cuts — is the exception, not the pattern. Best Buy, Target, Gap, McKesson, Charles River Laboratories, Deere, and J.M. Smucker all raised full-year estimates without comparable commitments to pass refunds through. Dollar General is the boundary case: the discounter used part of its refund to lower prices for cash-constrained rural shoppers and posted a 3.5% comparable-sales gain in its fifth consecutive quarter of higher traffic. Chief Executive Todd Vasos told investors that “our core customers continue to be financially constrained.” The fact that the discounter needed tariff refunds to keep its base shopping at all is not a sign of consumer health. It is a sign of consumer fragility, with the refund acting as the prop that distinguishes a quarter from a contraction. Walmart itself posted comparable-sales growth at its slowest pace in more than six years even with refund money at the register.

The print therefore reads differently depending on which tailwind is doing the work in which line item. Abercrombie raised prices and kept the refund. Dollar General passed part of the refund through and read the result as survival. Walmart passed part of its refund through and still showed the slowest underlying demand of the cycle. These are not contradictions. They are the same number read at different points of incidence.

Specifying what “AI capex” and “federal spending” mean here is what makes the fragility legible. The AI capex in question is hyperscaler infrastructure — the data-center and accelerator buildout by Microsoft, Alphabet, Amazon, and Meta, capital spending that the hyperscaler 10-K disclosures document as running at a combined pace well above two hundred billion dollars annually across the four largest hyperscalers through 2025 and into 2026. The federal spending the Journal names is the continuing fiscal impulse — the spending-side provisions of the 2025 reconciliation still rippling through FY2026, plus the mandatory-program growth that exceeds the pre-2017 trajectory. The tariff refunds are the third leg, and they are the only leg with a built-in expiration. Apollo’s chief economist Torsten Slok named the conditional directly: “As long as the AI boom continues and the stock market continues to be elevated, and we continue to have strong consumer income growth, the consumer will continue to be in good shape. But if the AI spending fails to justify the investment, we will be having a different conversation.” Three pillars have to hold. The bear case requires only one to break.

The bill, when the AI capex thesis disappoints — when hyperscaler returns fall short of the embedded capital cost, when the data-center buildout outruns the load it was built to carry — comes due in identifiable places. It is written first in the capex supply chains that hired to meet the buildout: the electricians and HVAC technicians wiring gigawatt-scale campuses, the utility crews upgrading transmission for the load, the construction trades building the shells, and the chip-packaging and substrate workers downstream of TSMC and SK Hynix. It lands next on ratepayers whose utilities have financed the grid upgrades through rate increases approved by state public-service commissions. It is written second in the diversified shareholders whose retirement portfolios sit on the long side of the capex: pension funds, 401(k) holders weighted in equity indices, and target-date funds whose glide paths assumed the AI capex would compound at the rates the print assumes. The “growth” the bulls are reading is, in part, an option premium the equity market has paid for capex that has not yet earned its keep. When the option goes unexercised, the premium is what the holders refund — and that refund, unlike the corporate tariff refunds, is delivered to people who did not choose to receive it.

The number is what the print says it is. The composition is what the Journal’s report says it is. The bill, when it comes, lands on workers and holders who did not choose to receive it.