The $7 Trillion Reckoning: Why Goldman Sachs Says the 'Buy Everything' AI Trade Is Finished

The artificial intelligence gold rush is entering a new, more dangerous phase. After two years of a rising tide lifting all AI-themed stocks, Goldman Sachs has issued a blunt warning to investors: the era of simple, indiscriminate "buy everything" AI investing is definitively over. The investment logic has pivoted from a race to spend the most capital on infrastructure to a high-stakes demand for proof of returns, and the bills are starting to come due .

At the heart of this shift is a staggering financial reality. Hyperscaler AI capital expenditures are projected to explode by 70% in 2026, reaching approximately $737 to $770 billion . This is not a short-term splurge; baseline models from the bank forecast AI CapEx growing to $1.6 trillion annually by 2031, setting the stage for a cumulative spend of $7 to $8 trillion over that period . The market can no longer afford to treat every dollar spent as a dollar well spent. As Goldman Sachs trader Lee Coppersmith warned, the "previously straightforward trading logic for AI-themed assets is gradually unraveling" amid rising market structure vulnerabilities .

The End of the Blank Check

For the better part of two years, tech giants operated in a "build at all costs" phase, and the market applauded them. That phase is now finished. The key forces driving the shift are both financial and strategic. The massive capex build is not coming from spare cash; it is being funded by a reduction in share buybacks and an increase in corporate leverage . Goldman Sachs strategists point out that this financial engineering is beginning to bite. The bank predicts that return on equity (ROE) for the largest tech firms could drop by an average of seven percentage points next year as these massive investments pressure profitability. Hyperscaler capex is now on pace to exceed 90% of cash flows, a level higher than the share seen during the Dot Com Boom, constraining financial flexibility just as the need to prove the durability of these investments intensifies .

The cost of staying in the AI race is, as the bank notes, "becoming harder to ignore" .

Profits Trapped at the Semiconductor Layer: The Downstream Bet

One of Goldman Sachs's most critical findings is that the immense profit pool from the AI buildout is currently trapped. Jim Covello, head of research at Goldman Sachs, has stated that the "economics of artificial intelligence are more questionable today than two years ago," because enterprise buyers, model companies, and hyperscalers have yet to show a return on their spending .

The bank’s research concludes that, for now, the AI windfall has been largely confined to the semiconductor layer—principally companies like Nvidia. The promised profits for the companies building on top of that hardware are unproven. Consequently, Goldman Sachs believes the investment opportunity is now shifting downstream to "nascent enablers," platform stocks, and productivity beneficiaries. The message is clear: the market will begin to richly reward only those companies that can demonstrate a credible monetization pathway for their AI products, not just the capacity to build them .

The $832 Billion Paradox: Massive Demand, Uncertain Margins

This pivot toward demanding returns is not happening because AI demand is slipping. If anything, the evidence of demand has never been stronger. The combined revenue backlog of cloud giants Google Cloud and AWS has surged to a staggering $832 billion, a clear signal that demand for AI compute far exceeds current supply . This backlog confirms that AI is not a speculative mirage but a genuine structural shift in the global economy .

The tension, which defines the new investment landscape, is precisely this paradox: there is massive, confirmed demand, yet that demand has not translated into proportional profits for the hyperscalers or their enterprise customers. The investment thesis now hinges on resolving this disconnect. The market has shifted from asking "how much money is being burned on compute?" to demanding a realistic answer to "when will this trillions be spent produce a return on investment, and what will the margins look like when it does?" .

The new investment regime no longer lifts all boats. Goldman Sachs notes a "Great Decoupling" in the market, where the correlation between AI-exposed stocks has collapsed as investors aggressively discriminate between companies with a clear, credible return on investment path and those without one . With cumulative spending projected to reach $7–8 trillion through 2031 and profits remaining deeply uncertain, the bank’s message is a reality check for the entire tech industry: look past the headline-grabbing spending figures and focus on the massive bill that is coming due .