The key to AI trade's bottom: Meta's debt "prisoner's dilemma" — Goldman advises selective dip-buying as Nomura gives the arms race 75% odds

Where does the AI trade find its floor? The market's focus has settled on a game-theoretic structure — the debt "prisoner's dilemma" facing the Metas of the world. Over the past week the Nasdaq logged its worst day in months, US high-yield risk premia rose to a 17-month high, and CCC junk-bond yields climbed to 15.85% — the highest since 2020; amid the wreckage, Goldman's advice is to selectively buy the dip in AI, but with limited conviction^[1][2].

How deep the positioning purge: zeroed baskets and bottomed leverage

Goldman's data capture the violence of the unwind: its AI beneficiaries-versus-risk long-short basket is down about 55% from its June peak, with all year-to-date gains surrendered; the Goldman AI & AI-hedge basket (GSPUARTI) has fallen from 583.45 on June 22 to 261.71 — a ~55% drawdown; fundamental long-short net leverage fell another 3.5 points this week to 47%, the 1st percentile of a three-year range — seven of the past eight sessions were de-levering days, faster than any period in 2025; Thursday's tech-exposure reduction was the largest single-day in two years, entirely driven by long liquidation^[1][2]. Tony Pasquariello's verdict: "this selloff has reached the market's core holdings"^[2]. Yet the divergence is real too: the S&P 500 ex-AI index is up 6.5% in October, on track for its best month in nearly three years, and 337 S&P members rose the day the SOX fell 6.7% — the AI premium is being surgically removed, not indiscriminately sold^[2].

The debt prisoner's dilemma: race to raise vs collective damage

Meta's financing plan has shoved the contradiction between AI capex and credit-market supply onto the table: 2026 capex narrowed to $135–145 billion, Q4 spend of $49–59 billion, negative free cash flow, and hundreds of billions to be raised from this autumn. Barclays notes Meta is financing early partly because others in the AI ecosystem have already launched or plan raises — the market's absorption capacity is shrinking; that exposes the giants' prisoner's dilemma: each has an incentive to lock up scarce funding first, but if they all accelerate issuance, new supply raises funding costs and squeezes everyone's room^[1]. The credit deterioration has texture: the biggest book-skration wave of the year in HY primaries, deeply distressed loans below 60 cents at the highest since March 2020, and loans at 80 cents or below up nearly 90% over twelve months per JPMorgan — yet the broader credit market has not deteriorated in tandem, showing the risk concentrated in the weakest issuers; DoubleLine's Robert Cohen thinks some CCC borrowers may head toward restructurings, but that doesn't mean the credit cycle has turned^[1].

Goldman's dip call and Nomura's counterweight

In the wreckage, Goldman strategist Louis Miller offers three grounds for bottom-fishing: momentum exposure is at its lowest since June 2025; the correction's size has improved risk-reward on some assets; and — the key pillar — hyperscalers' earnings estimates have not been revised down since July, with datacenter demand still strong^[2]. Nomura quantifies the other side from the industry: it puts about 75% odds on the AI race continuing and the semiconductor supercycle extending; 2027 memory supply and prices were already locked through long-term agreements (LTAs) this October, with defaulters potentially owing margins worth 50–100% of annual demand value — TSMC CEO C.C. Wei says AI demand still exceeds supply, and Amazon's Andy Jassy says even at $220 billion of 2026 capex the company can't meet demand, with some 2028 needs pre-booked^[3]. Nomura calls it a game of chicken: whoever cuts investment first risks losing the next round of AI competition — Google's Sundar Pichai even says frontier-model training compute outranks cloud revenue in priority^[3].

What to watch

First, the final size and spread of Meta's bond deal — the dilemma's first equilibrium, about to be revealed. Second, whether credit spreads peak once issuance lands. Third, when the first downward revision hits AI leaders' earnings estimates — once Goldman's core pillar wobbles, the rebound thesis fails. Fourth, the flip side of Nomura's 75% — any early signs of the two "game over" conditions: closed financing channels and power shortages^[1][2][3].