The market keeps acting like AI only matters if it comes with a data-center bill the size of a small nation’s GDP. Goldman Sachs says the real payoff may be coming later, in the boring places investors keep ignoring: software, labor, workflow, and the little productivity cheats that let companies do more with less. The catch is that the earnings tape still looks early, messy, and not especially generous to anyone trying to front-run the punchline.
Goldman’s latest read on S&P 500 Q2 2026 earnings says the market has mostly chased infrastructure and hyperscalers while underpricing the eventual productivity beneficiaries. That’s the setup. The reality is uglier: only 11% of S&P 500 companies quantified AI productivity gains for a specific use case, and only 2% directly quantified AI’s impact on earnings. Translation: everyone is selling the dream, almost nobody is showing the receipts.
Goldman’s report, led by strategist Ben Snider, points to a market still in the awkward teenage phase of AI adoption. S&P 500 EPS grew 31% year over year in Q2 2026, excluding one-time private-investment income, and AI infrastructure and hyperscaler earnings rose 54%, accounting for roughly half of index earnings growth. That’s the part investors love because it’s visible, scalable, and easy to model. The productivity side is still more rumor than line item, which is why the eventual winners may not be the loudest names in the room.
The cleanest takeaway is that productivity benefits are not showing up in a way that the market can easily score in real time. Goldman says the median S&P 500 company, excluding energy, grew earnings 14% year over year, while companies that quantified AI productivity gains showed no statistically significant earnings-growth difference versus peers, at 17% versus 14% median. In other words, the AI savings story is real enough to keep funding budgets, but not yet strong enough to bend the earnings curve and make everyone look like a genius on television.
Goldman’s framework says companies are funding AI by reallocating existing budgets, with roughly two-thirds doing it that way. The biggest shares come from software at 18% and labor at 11%, which is the kind of corporate accounting that sounds inspirational until you realize it usually means somebody’s headcount or vendor stack is getting trim. The practical read for investors: the AI productivity trade is likely to show up first in software-heavy companies that can squeeze more output from the same workforce.
The catch is that the evidence pack does not verify Goldman’s exact 20-stock list from the MarketWatch headline, so don’t let anyone pretend the map is more detailed than it is. What is verified is the broad direction. Companies best positioned for productivity gains are those with high software intensity, meaningful labor leverage, and enough operational discipline to turn AI into margin expansion instead of a fancy demo. The stock market loves a transformation story until the bill arrives; then it starts asking where the free cash flow went.
The Ramp AI Index gives the most concrete snapshot of adoption momentum in the evidence pack. Median monthly AI spend per employee rose from $5 in January to $12 in July. Top-decile spenders went from $240 to $650 over the same stretch. That is still small beer relative to company budgets, but it shows the spend line is moving from curiosity to habit. If AI were just another conference buzzword, these numbers would look flatter than a dead ETF on a Friday afternoon.
Goldman also estimates AI inference costs at under 0.5% of S&P 500 revenues, which helps explain why management teams can experiment without torching the income statement. That low cost base matters because it means the productivity upside can scale before the expense line becomes a wrecking ball. Investors should care less about who bought the biggest model and more about who can embed AI into workflows without turning the budget into a crime scene.
Only 11% of S&P 500 companies quantified AI productivity gains for a specific use case during earnings calls, and only 2% directly quantified AI’s impact on earnings. That gap matters. Corporate America loves to talk about AI in broad, polished language, but once someone asks what coding or customer support saved last quarter, the answers get slippery fast. Goldman’s read is that the gains are coming, just not cleanly enough to show up in the near-term scoreboard.
Goldman’s strategists said, “Q2 results showed a small and statistically insignificant difference in earnings growth between the companies quantifying AI productivity gains this quarter and other S&P 500 companies.” That’s the market’s favorite kind of disappointing sentence: precise, boring, and difficult to trade with confidence. For now, the AI productivity trade is an expectation market, not a proof market. That can still be profitable, but only if investors keep their ego on a short leash.
Goldman expects the AI productivity earnings impact to become clearer in the coming quarters as enterprise AI spending accelerates and companies move from experimentation to deployment. That’s the pivot point. Right now, the market has mostly paid up for infrastructure, hyperscalers, and the firms building the picks and shovels. The later trade, if it shows up, should reward the companies that convert AI from a science project into an operating advantage that survives contact with quarterly reporting.
The practical screen is simple even if the exact stock list is not verified here: look for businesses with heavy software use, meaningful labor exposure, and visible AI deployment in actual workflows, not just in slide decks and conference chatter. If management can point to coding, customer support, or back-office gains with real operating leverage behind them, that is where the market may eventually stop being skeptical and start being expensive. Until then, investors are mostly buying the idea that the machines will do the grunt work while humans keep the bonus pool intact.
This is not a clean momentum trade and it is definitely not a victory lap. The verified data say AI productivity is moving from theory to spending, but the earnings proof is still thin, and the market has not yet rewarded the companies making the clearest case. For now, the smart money posture is patience with a radar on software-heavy, labor-leveraged names, because when the productivity numbers finally show up, the party will probably start in the least glamorous corner of the market.
