Concerns Over Trillion-Dollar AI Capex: Half of S&P 500 Earnings Growth Built on Debt, Smart Money Begins to Exit

When Wall Street's largest investment bank uses "return to normal" to describe a debt-driven tech frenzy, its own data quietly tells a different story.

According to a series of research reports released by Goldman Sachs this week, approximately half of the earnings growth of the S&P 500 currently depends on AI capital expenditures, which are increasingly sustained by bond financing rather than internal operating cash flows. Goldman Sachs' own credit strategists expect that hyperscale tech companies will cover about 35% of their capital expenditures with debt in 2027, amounting to approximately $400 billion.

Meanwhile, positioning tracking data from Goldman Sachs derivatives strategist Robert Quinn shows that during last week's 7.1% surge in the Nasdaq, non-dealer investors collectively sold a record $21.6 billion in Nasdaq futures—hedge funds and asset management institutions completed a historic exit amidst market euphoria.

These signals combined paint a picture worthy of investor caution: a market capitalization increase driven by debt financing is continuing against the backdrop of institutional funds quietly exiting, while the macroeconomic foundation beneath it—the labor market—is simultaneously weakening.

Half of Earnings Growth Supported by Borrowing

Goldman Sachs Chief Market Strategist Tony Pasquariello pointed out bluntly in this week's report that the S&P 500's earnings grew by approximately 25% year-over-year in both the first and second quarters, a remarkable rate. However, he quickly shifted tone:

"The bad news is that the likelihood of earnings growth slowing down next year is increasing... AI capital expenditures currently drive about half of the S&P 500's earnings per share growth, but given their scale and reliance on external financing, this growth rate is unsustainable."

Goldman Sachs' calculation path clearly reveals the cyclical logic: hyperscale tech companies borrow to invest in capital expenditures, which translate into revenue for chip manufacturers, data center builders, etc.; this revenue lifts overall earnings figures, and earnings growth in turn supports valuations, providing room for a new round of equity financing. Pasquariello specifically mentioned in the report that Alphabet is issuing another $25 billion bond to the market.

Goldman Sachs portfolio strategist Ben Snider disclosed in a concurrent report that U.S. companies have raised $105 billion through secondary stock offerings year-to-date, the highest level for the same period since 2021, with about 40% related to AI. Goldman Sachs expects hyperscale tech companies' capital expenditures to reach $1.1 trillion in 2027, exceeding operating cash flow by $150 billion, a gap that will be filled by borrowing.

AI Investment Scale Far Exceeds Market Consensus

The actual scale of capital expenditures is larger than the commonly cited figures. Goldman Sachs economists, including Joseph Briggs, pointed out in the "Global Economics Weekly" that the market's widely quoted figure of "approximately $800 billion in capital expenditures by U.S. hyperscale tech companies this year" is actually an underestimate. Goldman Sachs' adjusted calculations show that global AI investment in 2026 will be $1.019 trillion, with the U.S. portion at $581 billion; two independent verification methods yielded global totals of $1.002 trillion and $1.06 trillion, respectively.

Measured from a macro perspective, Goldman Sachs expects AI capital expenditures as a share of U.S. GDP to rise from 1.8% this year to 2.8% in 2028—approaching the volume of U.S. residential investment at the peak of the economic cycle in the mid-2000s.

However, Briggs also added an observation in the report that makes bulls less comfortable: "East Asian trade data shows that the recent growth rate of AI-related exports has slowed." Historical patterns indicate that the supply chain side is often where cycle turning points appear earliest.

Credit Market Sounds the Alarm First

The fervor in the stock market has not received simultaneous endorsement from the credit market. In Goldman Sachs' cross-divisional weekly report, strategist Chris Hussey highlighted a notable divergence: Credit spreads for AI-related issuers have begun to underperform non-AI peers, as credit investors are repricing risk for a "multi-year high capital expenditure cycle."

This divergence holds significant signal value. When equity investors are still chasing gains in the same companies, while credit investors begin to demand higher compensation, historical experience usually suggests that the latter's judgment is more reliable.

"Smart Money" Quietly Exits Amidst the Frenzy

The most thought-provoking single data point in this week's Goldman Sachs report comes from Robert Quinn's futures holdings tracking report, titled "Nasdaq Futures: Net Short."

The data shows that during the week of July 28 to August 4, when the Nasdaq Index surged 7.1%, non-dealer investors collectively sold $21.6 billion in Nasdaq futures, which Quinn called a "nominal scale hitting a historical high." Short positions accounted for 72% of the flow, with hedge funds selling $11.9 billion and asset management companies selling $7.4 billion. As of August 4, the net position of Nasdaq non-dealers turned negative for the first time since May 2025.

In other words, institutional funds used the rally driven by retail investors and momentum capital to complete a historic exit. Sarah Cohen's weekly report on Goldman Sachs' equity brokerage business further confirmed this atmosphere: hedge funds were net buyers of stocks for the second consecutive week, but the total leverage ratio of fundamental long-short hedging portfolios remained at the 2nd percentile over the past year, at approximately 201.6%. This is closer to cautious probing than firmly adding positions—not the appearance one would expect at the top of a sustained bull market.

Labor Market: Undercurrents Beneath the Party

Goldman Sachs' optimistic narrative has historically relied on consumer resilience and a soft landing in employment. Non-farm payrolls decreased by 23,000 in July, with the previous two months' data revised down by a combined 103,000. The average new jobs added over the past three months was only about 20,000, a significant shrinkage from the 111,000 before the report was released; Goldman Sachs estimates that the "potential growth rate" of employment has dropped to about 5,000 per month.

The unemployment rate fell slightly to 4.09%, but according to the report's analysis, this was mainly due to 264,000 people leaving the labor force, rather than an increase in employment. Average hourly earnings grew by only 0.05% month-over-month.

This is less a "soft landing" and more a quiet stall in the labor market beneath the stock market party.

The Other Side of Risk

Synthesizing the core data from Goldman Sachs' reports this week, a panoramic view of structural concerns has become clear: Approximately half of the S&P 500's earnings growth relies on AI capital expenditures; global AI investment exceeding $1 trillion is increasingly sustained by $400 billion in new debt and record equity financing; the credit market is demanding higher premiums from AI issuers; institutional funds are reducing Nasdaq futures holdings at a record pace and turning net short for the first time; hedge fund leverage is at historically low levels; and the labor market supporting all this is adding only about 5,000 net jobs per month.

In response, Goldman Sachs' official interpretation is that equity issuance is "a return to normal, not prosperity," and that earnings deceleration is merely "a change in nature, not a change in direction." This judgment may not be without basis, but when an institution needs to exert considerable effort to reassure clients, while its best clients are reducing holdings at a record pace, and the bill for this feast is being silently borne by the bond market, the question investors need to consider is who will still be holding the bag when the music stops.