Q2 Earnings Season Hits Record Highs, but AI Investment Returns Remain an Open Question
nashnova research
S&P 500 earnings per share rose 52.6% year-on-year in Q2 — the largest gain since the 2021 recovery — yet much of the surge came from paper gains and AI infrastructure orders, leaving the central question unanswered: who will actually pay for AI?
A 52.6% earnings jump — how real is it?
Per FactSet, S&P 500 constituents posted 52.6% year-on-year EPS growth in Q2, the highest since the 2021 economic recovery.
But the headline was inflated by paper gains — unrealized mark-ups on equity stakes in other tech firms, not cash in hand. Alphabet booked roughly $98 billion in net gains from its SpaceX stake; Amazon recorded about $53 billion from holdings in Anthropic and others.
This means → the scorecard looks stellar on the surface, but a large share of those profits cannot be spent — they are accounting entries, not distributable cash.
Strip the paper gains — who is actually driving profits?
FactSet analyst John Butters notes that even excluding Alphabet and Amazon, S&P 500 earnings growth still hit 33.8% — fundamentals remain solid.
Yet profit contributions are heavily concentrated: Nvidia and Micron — AI chip and infrastructure suppliers — plus energy stocks lifted by oil-price rises tied to the Iran conflict.
In plain terms = the real money-makers are either selling AI "picks and shovels" or riding an oil-price tailwind. AI's own end-users are nowhere on the profit-contribution list.
The "spend now, earn later" model — where is the risk?
The current AI infrastructure boom runs on a build-first, monetize-later logic — chipmakers collect revenue today while investment costs amortize slowly over time.
This means → supplier profits are front-loaded and investor returns are back-loaded, making the current profit picture potentially rosier than the long-term reality.
This reflects a structural concern: AI infrastructure investment relies increasingly on debt financing, and if the build-out pace slows, supplier profits will contract fast.
Who pays in the end — what signal is the market waiting for?
The market still lacks firm evidence that AI end-users are willing to pay prices high enough for the companies deploying AI to earn a reasonable return on investment.
Rising interest rates are pushing up financing costs — the more expensive it is to borrow for data-center builds, the longer the payback period stretches.
In plain terms = the logic loop of the AI investment chain has not closed: the shovel-sellers have profited, the shovel-buyers have not struck gold yet, and the interest on the loans used to buy those shovels keeps climbing.
市场有风险，内容仅供研究参考，不构成投资建议。