Goldman Sachs warns $920B AI capex forecast may be too conservative

Goldman Sachs told clients that rising AI capital expenditures are increasing risks for AI stocks, warning that forecasts of $920 billion in AI-related spending by 2027 may understate the true scale of the buildout.

The $920 billion in AI capital expenditures projected for 2027 may be too conservative, Goldman Sachs warned, as hyperscaler spending accelerates faster than analysts can revise their models.

"The scale of AI infrastructure investment is unprecedented, and the risk is that returns fail to materialize in the timeframe markets are pricing," Goldman Sachs' US portfolio strategy team wrote in a research note published June 11.

The bank's estimate covers cumulative AI-related capex across cloud hyperscalers, enterprise IT, and chipmakers. Goldman's James Covello, global head of equity research, has questioned whether the torrent of capital flooding into AI infrastructure will generate returns commensurate with the spending — a question he first raised in 2024 and says remains unanswered.

The warning comes as the Nasdaq suffered its worst sell-off since October, with AI-linked megacaps shedding hundreds of billions in a single session. If $920 billion understates the true figure, the gap between spending and revenue generation widens further, raising the stakes for Nvidia, the hyperscalers, and the entire AI supply chain.

The $2.1 Trillion Question

Goldman's caution echoes a broader debate dividing Wall Street. Deutsche Bank reaffirmed its S&P 500 year-end target of 8,000 on June 1, citing AI-driven optimism and 14.2% EPS growth. JPMorgan Asset Management's David Kelly, writing the same day, flagged that hyperscaler capital spending is projected to jump 78% in 2026 to $739 billion from $416 billion. Vanguard's 2026 outlook puts cumulative AI capex from the start of 2025 through end-2027 at $2.1 trillion.

The divergence between bulls and bears is widening. JPMorgan CEO Jamie Dimon, speaking at Bernstein's Strategic Decisions Conference on May 27, described the environment as "gung-ho" and warned that confidence levels resemble 1972, 1986, 2000, and 2007 — years that preceded major market dislocations. Bridgewater Associates founder Ray Dalio told Bloomberg Television on June 3 that his proprietary bubble indicators show US equity markets "rising close to — not at — the same level in 2000 and the same level in 1929."

The ROI Test That Keeps Failing

Covello's core argument has not changed since his 2024 research note: at some point, the investments must generate returns. "You make investments in a business so that you can generate returns and make money," he said on Goldman's Exchanges podcast on June 2. "And we've gotten further away from that over the last couple years instead of closer to it."

George Lee, co-head of the Goldman Sachs Global Institute, estimated that $7 trillion to $8 trillion will eventually be spent on AI infrastructure globally. The math only works, he argued, if AI creates substantial net new economic activity — not merely disrupts existing profit pools. Enterprise ROI data remain underwhelming so far.

The skeptics have grown louder. Software engineer Benjamin Horne argued on his Substack that a significant share of reported revenue at frontier labs like OpenAI and Anthropic exists only because they have been subsidizing tokens — selling compute at steep discounts to build market share. "Strip out the massive token subsidies," he wrote, "and a gigantic chunk of 'demand' evaporates the instant it touches reality."

What It Means for Investors

For Nvidia, which has been the primary beneficiary of the AI capex boom, the risk is twofold: a slowdown in hyperscaler spending would compress its data center revenue growth, while any shift toward in-house chips from Amazon, Google, and Microsoft could erode its pricing power. Nvidia shares trade at elevated multiples relative to historical averages, leaving limited room for disappointment.

The small-cap AI supply chain has already begun to show cracks. Sanmina, the contract manufacturer that surged 77% year to date after its ZT Systems acquisition, guided Q3 revenue to $3.2 billion to $3.5 billion — a sequential step-down from the $4 billion reported in Q2. Viavi Solutions, up 421% over 12 months on hyperscale data center test demand, has fallen 14% in the past 30 days. The market is already pricing the next leg, not the last one.

The next stress test arrives in late July with Sanmina's Q3 report, followed by hyperscaler capex disclosures from the four major cloud providers. If those numbers keep bending up, the bull case holds. If they roll over, Goldman's warning will look prescient rather than cautious.

This article is for informational purposes only and does not constitute investment advice.