Goldman Sachs says global markets are now facing three reflexive loops that are reinforcing one another: oil and politics, hyperscaler capital spending, and AI infrastructure debt. Put together, the bank says, they have left the market resting on a fragile and dangerous equilibrium.
In a client report this week, Rich Privorotsky, Goldman Sachs’ head of 1-Delta trading, wrote that the combined pressure from surging oil prices and higher rates is becoming harder for markets to absorb, with the negative shock in bonds worsening sharply. If there is no meaningful political de-escalation, he said, markets will have to absorb the risk on their own, pushing conditions in a worse direction.
Oil, politics and inflation form the first loop
Privorotsky said the first reflexive chain he is watching most closely is the two-way feedback between oil prices and politics.
Brent crude briefly moved above $100 a barrel this week, testing market assumptions around how President Donald Trump’s policy response would unfold. Before that move, many investors had expected that once oil crossed a certain threshold, lifting retail gasoline prices and weighing on presidential approval, the Trump administration would step in to suppress prices.
That expectation had helped support equity resilience. But Privorotsky said that with each day passing at the same price level and no policy response arriving, the market has to do more of the work itself to force an outcome.
Rates are becoming the hardest variable in that loop to ignore. He also said the pass-through from energy costs into food inflation is about to become more visible.
Warnings are already appearing in the real economy. American Airlines cut its 2026 guidance despite record revenue, higher fares and demand that remained relatively solid, because fuel costs have climbed by about $1.6 billion since early July.
Geopolitical tensions are also rising. According to China Central Television, Trump said at the White House on July 24 local time that the U.S. had two choices on an "exit strategy" for the war with Iran: continue current military action and possibly intensify strikes to gradually destroy Iran’s military capability, or reach a deal through negotiations.
Earlier that same day, The Wall Street Journal Chinese-language affiliate Wallstreetcn said Reuters, citing sources, reported that Pakistan was exploring efforts to revive stalled U.S.-Iran talks.
Another date on the calendar is also drawing attention. Israeli Prime Minister Benjamin Netanyahu is due at the White House next Tuesday for talks with Trump. Privorotsky said that meeting could set up what he called a "TACO moment," meaning a sudden negotiation- or compromise-driven market move.
Hyperscaler spending is the second reflexive loop
The second loop centers on capital spending by large technology companies and a simple question: is the market still willing to treat massive outlays as a no-cost growth signal?
Google became the clearest negative marker of this earnings season. The company raised its 2026 capital expenditure guidance to $195 billion to $205 billion, posted quarterly free cash flow of negative $5.9 billion, and then saw its shares fall 6.9%.
Operationally, cloud growth of 82% would normally have stood out. But investors are no longer prepared to treat spending as a strategic input with no price attached. The report said questions around the frontier product roadmap and return on investment were met with little in the way of direct answers.
The knock-on effect is broader competition. If Google spends more, peers may have to follow, which would put pressure across the wider hyperscale cloud group. Stress is already showing up in hardware.
- STMicroelectronics missed expectations on core profit and issued a somewhat soft third-quarter revenue outlook, with the stock down about 14%.
- Texas Instruments, trading under TXN, held up relatively better, but still closed down 3%.
On the AI race itself, Privorotsky said the gap between leading closed-source models and Chinese open-source models has narrowed sharply. A lead that used to be measured at nine to 12 months has, in some benchmark tests, compressed to just a few weeks.
He said the cost and marginal return profile of pre-training, compared with reinforcement learning and post-training, is producing very different economic models. Competition at the application layer is unusually intense, and the market structure there is unusually flat.
He added that the risks tied to smaller models and higher efficiency are a story for later, but not one that should be ignored.
AI infrastructure debt forms the third loop
The third reflexive loop sits inside the bond and financing structures of the same hyperscale technology companies.
Privorotsky said the bond market may be the clearest risk signal right now. He pointed to Meta’s "Hyperion" financing through Beignet SPV. The $27.3 billion bond deal was priced at par when issued, later traded above 109, and has since fallen back to around 95.
He said hyperscalers as a group still have solid financial positions and are not highly levered on their balance sheets. Even so, valuation repricing is already putting meaningful pressure on equity multiples.
The more serious issue, in his view, is what happens as capital spending accelerates and free cash flow conversion keeps deteriorating. The leveraged entities providing financing for infrastructure build-outs would be hit harder.
His warning is direct: today’s capacity expansion can turn into tomorrow’s compute oversupply, and once that happens, larger depreciation charges will begin to wash through income statements.
Microsoft earnings and CXMT’s listing are the next tests
Looking ahead, Privorotsky named two upcoming events as key tests for the broader reflexivity thesis.
The first is Microsoft’s earnings call on Wednesday. "If the reflexivity thesis is going to play out, this may be the most important call," he said. Investors will be watching how Microsoft balances capital spending, cloud growth and free cash flow to judge whether the hyperscaler narrative can stabilize.
The second is the STAR Market listing of Chinese memory-chip maker ChangXin Memory Technologies. The report said CXMT is already the world’s fourth-largest DRAM producer and is seeking to raise about $8.6 billion.
Privorotsky said, "This is absolutely not an irrelevant new competitor." If the company begins trading near the valuation implied by the over-the-counter perpetual market, he said, the implications for the broader memory-chip sector could be significant.
He closed with a line that summed up his view of the current setup: "It feels a bit like circular reference on the oil problem." In a market ruled by reflexivity, each variable can act as both cause and consequence.