Goldman Sachs has told clients that this month's decline has grown beyond a selloff in artificial intelligence stocks, and has listed three developments that would support a bottom. The note, written by the bank's equity strategists, landed before Friday's session was fully reflected in its data, which means the market may already be past the point it described.
The S&P 500 closed Friday at 6,918.45, down 3.59% for its seventh consecutive loss and 12% below its October 6 record. The Nasdaq Composite fell 5.28% and is now lower for the year.
"What began as an AI unwind expanded this week to a broader growth downgrade," the strategists wrote, according to Fortune. Until this week, the index excluding AI names had held up; it is now 6% below its own high. Goldman's basket of AI stocks is down 32% since October 6, and cyclical shares have trailed defensives by five percentage points over the past ten days. The correction itself has been building for a week.
Condition one: evidence that AI earnings are real
The first thing Goldman wants is news that raises confidence in the profits companies expect from AI. That is the test the market has been running, and the early results are mixed.
Dan Niles, an investor who has been skeptical of the trade, told CNBC that a price war between Anthropic and OpenAI is hurting both. Their combined sequential revenue growth fell from 75% in the second quarter to 48% in the third, while combined losses rose from $7.9 billion to about $8.4 billion.
Valuations still assume a lot. Reuters put Anthropic at roughly 40 times trailing sales, a multiple that one US Bank Asset Management executive said "assumes years of rapid growth." Goldman's own view is that the AI outlook for 2028 and beyond "is highly uncertain."
The IPO that was supposed to settle the question did not. Anthropic raised $93.5 billion after underwriters exercised their option, passing SpaceX for the largest listing on record. The shares rose 17% on their first day, then gave most of it back on Friday to close at $88.91, not far above the $83 offer price. Morgan Stanley, the underwriter responsible for stabilising the stock, closed its support position.
Condition two: a dovish Fed
The second condition is a signal from the Federal Reserve that rates are not going much higher. That looks unlikely before the December 9 meeting. Futures put the odds of a December increase at about 95%, and Goldman's economists expect another in March, taking the federal funds rate to a peak of 4.25% to 4.5%.
Philadelphia Fed president Anna Paulson said on Thursday that the central bank will need to raise rates again because inflation is too high. The 10-year Treasury yield is near 5%. Treasuries rallied on Friday as investors looked for safety, which is not the same thing as a change in policy.
Higher rates weigh on profits that arrive years from now, which is the shape of an AI infrastructure bet. A market that needs cheaper money to justify its valuations is waiting on an institution that has said it is in no hurry.
Condition three: investors actually give up
The third condition is a large drop in investor positioning, which has often marked the end of past drawdowns. Goldman says hedge funds have already cut back, with net leverage at its lowest since April 2025. The most popular long positions were crowded into AI, and that basket has trailed the S&P 500 by 14 percentage points over six months, among the worst stretches in its 20-year history. The bank described the hedge fund community as "bruised and broadening."
The wider picture is less washed out. Goldman's sentiment indicator remains above average, US equity funds and ETFs have taken in nearly $170 billion this quarter, and asset managers' net equity futures positions sit in the 82nd percentile of the past decade. Retail and institutional money has not capitulated; it has rotated.
What the bank expects
Goldman's November 17 outlook, published before the worst of the selling, forecast 8,700 for the S&P 500 by the end of 2027. In the November 20 note the range widens to roughly 5,800 to 8,100, depending on whether investors regain confidence in earnings.
History offers a mild argument for patience. After the 22 corrections since 1980, the index returned an average of 4% over the next three months and rose in 17 of them. E*Trade's research is less comforting: of the 14 times since 1983 that the index first closed 10% or more below a record, five became bear markets.
The VIX closed at 35.89, its highest in 19 months. The case that AI spending had become a bubble has been circulating for weeks; what changed this week is that the people making it now have a chart to point at, and Goldman has published the conditions under which it would stop agreeing with them.