Goldman warns three reflexivity loops threaten $725 billion AI spending cycle

Three reflexivity loops — oil at $100, hyperscale capex overshoot, and AI bond repricing — push markets toward a fragile equilibrium, Goldman Sachs warned.

Brent crude surged past $100 a barrel Thursday as Houthi strikes on Saudi tankers in the Bab al-Mandeb Strait compounded a technology-sector rout that erased more than $140 billion from Tesla alone, with Goldman Sachs warning three reflexivity loops now threaten market stability.

"Each of these three cycles — oil and politics, hyperscale capital expenditure, and AI infrastructure debt — is reinforcing the others in ways that make the current equilibrium fragile and dangerous," said Rich Privorotsky, head of 1-Delta trading at Goldman Sachs, in a client report this week.

Alphabet fell 7 percent Thursday after raising its 2026 capital expenditure forecast to as much as $205 billion, driving free cash flow to negative $5.9 billion in the second quarter — the first time the company has posted negative free cash flow since going public in 2004. Tesla dropped 14.5 percent after operating income plunged 57 percent to $398 million, with capital expenditures surging 142 percent to $5.79 billion. The four largest hyperscalers — Alphabet, Amazon, Microsoft, and Meta — are now projected to spend a combined $725 billion on capital expenditures in 2026, up roughly 77 percent from $410 billion in 2025, according to company filings and analyst estimates.

The three loops converge on a single question the Federal Reserve will confront at its July 28-29 meeting: can AI infrastructure investment generate sufficient returns before rising energy costs and sticky inflation force a rate hike that would make those future returns worth less? Markets assigned roughly a one-in-three probability of a rate hike before Thursday's oil spike, a probability that almost certainly moved higher after Brent settled above $100.

The Oil-Politics Loop

Brent crude settled above $100 a barrel Thursday for the first time since May, part of a more than 30 percent monthly rise driven by escalating conflict in the Middle East. Yemen's Houthi rebels claimed strikes on two Saudi oil tankers — the Encelia and the Layla — in the Bab al-Mandeb Strait, a chokepoint that handles about 8 percent of global oil trade, according to US energy data. Saudi authorities confirmed the Encelia was hit and set ablaze.

The attack compounds a broader supply shock: a near-halt to shipping through the Strait of Hormuz has effectively reduced Iranian oil exports, and the Houthi closure of the Bab al-Mandeb now threatens Saudi Arabia's alternative export route. US West Texas Intermediate crude settled near $92 a barrel, up about 6 percent on the session. Average US gas prices climbed to around $4.10 a gallon, up from $3.94 the prior week, according to AAA data.

Privorotsky noted that markets had previously assumed President Trump would intervene to suppress oil prices once they crossed a threshold that threatened his approval ratings. "Each day that passes without policy response at the same price level, the market increasingly has to drive the outcome itself," he wrote. The last time Brent traded above $100 for a sustained period was in 2022 following Russia's invasion of Ukraine, when it peaked near $128 a barrel.

The transmission to the real economy is already visible. American Airlines lowered its 2026 guidance Thursday despite record revenue and solid demand, citing a roughly $1.6 billion increase in fuel costs since the start of July.

The Hyperscale CapEx Loop

Alphabet's second-quarter results crystallized the tension at the heart of the AI investment thesis. Revenue rose 24 percent to $119.8 billion, Google Cloud surged 82 percent to $24.8 billion, and the company's Gemini app reached 950 million monthly active users. Yet the stock fell 7 percent because investors focused on the spending plan: 2026 capital expenditure of $195 billion to $205 billion, up from a prior range of $180 billion to $190 billion, with management adding that 2027 spending would increase "significantly."

The arithmetic is stark. At a $200 billion annual capex midpoint, Alphabet's spending represents roughly 42 percent of its annualized revenue run rate — a ratio more typical of utilities and telecommunications companies than of a software-driven business. Free cash flow swung to negative $5.9 billion from positive $10.4 billion a year earlier.

The competitive dynamic makes the spending cycle self-reinforcing. Any hyperscaler that unilaterally reduces AI infrastructure investment falls behind in model capability while rivals keep building. That is why Alphabet raised guidance even knowing the stock would likely fall: standing pat would have been a competitive retreat. The risk, Privorotsky warned, is that today's capacity expansion becomes tomorrow's compute overcapacity, with depreciation charges beginning to erode earnings.

The AI Debt Loop

The third reflexivity loop is the least visible but potentially the most dangerous. Meta's $27.3 billion "Hyperion" bond issuance, structured through its Beignet SPV, was priced at par and traded as high as 109 before falling to about 95, according to Goldman Sachs data. The repricing signals that credit markets are beginning to question the returns on AI infrastructure investment.

Goldman Sachs has documented the structural dimension: hyperscaler capex is expected to consume roughly 100 percent of combined operating cash flow in 2026, and the resulting depreciation build will push technology companies' debt-and-amortization burden from 7 percent of revenue in 2022 to a projected 12 percent by 2027. The S&P 500's return on equity hit a record 22 percent in the first quarter of 2026; Goldman's analysts project ROE across the largest technology companies will decline by an average of seven percentage points as depreciation charges accumulate.

Privorotsky identified two near-term catalysts that will test the reflexivity thesis. Microsoft's earnings call Wednesday will be scrutinized for its capital expenditure guidance, cloud growth trajectory, and free cash flow trajectory. Separately, Chinese memory chip maker ChangXin Memory Technologies, now the world's fourth-largest DRAM producer, is preparing a roughly $8.6 billion initial public offering on the STAR Market in Shanghai — a new competitor that could pressure the semiconductor supply-demand balance.

"Each variable is both cause and effect," Privorotsky wrote. "It feels a bit like circular referencing on the oil question."

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