Pent-Up Drag

Aggregate S&P 500 earnings are unusually strong. The Mag 7 prints are doing the heaviest lifting, but breadth has improved, and the headline number has surprised positively for several quarters running. Most strategists are revising 2026 expectations higher. 

The market is focused on the wrong thing. 

Not because the numbers are wrong — the numbers are right. The composition of those numbers reflects a specific accounting mechanic that flatters today’s print (the headline number) at the expense of future ones. A meaningful share of what the market is treating as run-rate earnings is timing. It looks like profit today because GAAP says it is. It will look like drag later for the same reason. 

The Mechanic, Plainly 

We use the phrase pent-up alpha to describe the gap between price and fundamentals. When price lags fundamentals, alpha is charging — poised to accelerate future performance once released. When price leads fundamentals, alpha is being borrowed — pulling forward future results into today’s number. Pent-up drag is borrowing at the macro level. 

Here’s how it works. Hyperscalers spend cash today on AI infrastructure. Their vendors — across silicon, networking, power, cooling, and the data center build — recognize most of that cash as current-period revenue within a year, at full incremental margin. The hyperscalers themselves capitalize the assets and depreciate them over longer schedules: typically five to six years for equipment, longer for buildings. Reported earnings get the revenue side fast and the cost side slow. The bigger the build, the wider the gap. 

The scale is without modern precedent. Combined hyperscaler capex in 2026 is tracking roughly $800 billion when Oracle is included alongside the four largest. That single-year figure is roughly equal to what all non-tech S&P 500 companies spent combined in 2025. Morgan Stanley projects 2027 hyperscaler capex above $1.1 trillion, with the curve still pointing higher1. Capex-to-revenue ratios across the group are running at levels that would have read as financial distress in any prior cycle. They are being celebrated as conviction. 

The Phases 

This dynamic does not end on a calendar. As long as capex continues, the tailwind persists. New spend keeps outrunning the recognition curve, and each year of buildout adds another tranche of deferred drag. The math is mechanical: when capex grows, the gap widens. 

Equilibrium arrives when capex flattens. Depreciation catches current spending. The free dollar disappears from aggregate EPS. 

The headwind arrives if capex declines. Vendor revenue rolls over. Depreciation lingers for years afterward. Useful-life extensions are a one-time lever, already pulled when Microsoft moved from four to six years in 2022 and peers followed. If AI silicon’s economic life proves shorter than its accounting life, impairments become the pressure valve. 

The bear case is lazy. The bull case is complacent. Bears call the print fake. Bulls call it the new baseline. The truth is in between, and it has a name. 

The Historical Rhyme 

Borrowing has many forms — pull-forward demand, working capital releases, low-base optical growth, one-time tax tailwinds. These mechanics are routine. The scale is not. Most are too small to register at the index level. The capex/depreciation arbitrage is today’s exception, large enough to bend the index. 

The analog is fiber. In 1999-2000, telcos capitalized fiber buildouts and depreciated them over long useful lives while Cisco, Lucent, and Nortel booked the revenue immediately. Both sides looked great. Then capex flattened, depreciation kept compounding, ROIC failed to materialize, and vendors collapsed while telcos took impairments. The mechanism was identical. 

The difference today is that AI workloads might generate enough productive revenue to absorb the eventual depreciation stack. That is the actual debate. It cannot be had honestly without first acknowledging the timing distortion. 

The Application 

I wrote in The Peak Margin Canard² that the bear case typically identifies the wrong mechanism for margin compression. We are naming the right one here. In Capturing Pent-Up Alpha³, I wrote that timing matters more than direction. We are naming the inverse phase here. 

This is not a sell-signal post. It is a sizing-conviction post. The wider the disconnect between reported earnings and sustainable earnings power, the more asymmetric the downside in any long thesis built on headlines. The trade depends on whether AI revenue catches up to the depreciation stack before the depreciation stack catches up to AI revenue. 

The current rosy earnings picture is the surface. Underneath it is the largest single timing distortion in modern S&P 500 reported earnings — measurable, mechanical, and entirely visible to anyone willing to look. 

Borrowing in the name. Pent-up drag in the index. You will hear about this again. 

Footnotes 

¹ Ananya Gairola, “David Sacks Says AI Could Drive 75% Of US GDP Growth As Morgan Stanley Sees Big Tech AI Capex Surging Past $800 Billion In 2026″, Yahoo! Finance, May 5, 2026. 

² Mark Scalzo, “The Peak Margin Canard“, Validex.co, March 20, 2026.  

³ Mark Scalzo, “Capturing Pent-Up Alpha“, Validex.co, March 30, 2026.