Goldman Sachs pegs Big Tech AI buildout at $7.6T through 2031

The number to hold in your head is Goldman Sachs's estimate that Big Tech will spend $7.6 trillion through 2031 building the data centers to run AI, and this week the Nasdaq Composite fell nearly 5% because investors are no longer willing to take that capex on faith. CBS News laid out the case that is now assembling itself against the buildout, and the interesting thing is that the challenge is coming from the demand side, not from any single missed earnings print.

The most pointed quote in the piece comes from Kate Brennan, associate director at AI Now, who told CBS that "the returns are not coming in, and the claims that are being made, in terms of efficiency or productivity numbers, are not netting out." The demand-side evidence cited in the article does not look better. A May study from Gartner found that businesses replacing workers with AI agents often fail to generate positive returns on investment, and Bank of America research indicated that few consumers currently want to pay for AI services despite rising usage. Pew Research, also cited, found 40% of US adults expect AI to negatively affect society over the next two decades, against 16% who see it positively.

Why this matters if you are not an investor in Nvidia or the hyperscalers: the shape of the AI product market for the next few years, how aggressive the pricing gets, how many free tiers survive, how many startups get funded, is downstream of whether the capex thesis holds. Economist Ed Yardeni put the mechanics plainly in the piece: "The AI ecosystem falls apart if the expected end-user demand for the AI/LLM products does not materialize or if prices fall sharply below expectations." Vanguard's researchers, also quoted by CBS, told investors to expect volatility and warned some firms will thrive while others face obsolescence in an AI-driven economy.

The honest caveat is that a single bad week for the Nasdaq is not a bubble popping, and the CBS piece is a synthesis of analyst notes and survey data rather than a leaked earnings model or a specific default. What the reporting does not give you is the split between the hyperscalers themselves, how much of that $7.6 trillion is Microsoft versus Google versus Amazon versus Meta, how much is debt-funded, or which segments of enterprise buyers are actually renewing seats at the prices being modeled.

If demand does firm up, the story becomes a straightforward payoff for the biggest spenders. If it does not, the ones who benefit are the second-mover CIOs waiting out the pricing and the specialized AI companies with proven ROI in narrow verticals, while the pressure lands on whoever borrowed to build the racks.