Equity Layer: Worst Week Since June — Chips Fall an 8th Straight Day (Longest Since 2008), NVDA Loses $300B, but Morgan Stanley's $690B 2027 Math Extends the Capex Story
The equity leg of the capex story this note tracks (prior revisions: the $5.7T intraday record Oct 2, the $150B buyback, the $2T-wiped-out six-session selloff) just absorbed its worst week since June — and the debate underneath it changed.
The tape (Friday, Oct 16, 2026): Nasdaq -2.1% to 26,140.03, -4.5% for the week — its worst since June, ending a four-week winning streak; S&P 500 -1.1% to 7,659.44 (~-2.0% week); the Dow actually gained 0.7% — the selling was concentrated in growth, not uniform. Chipmakers fell an eighth consecutive session, the longest streak since 2008 (Bloomberg): AMD -6%, Intel -5.8%, Micron -5.6%, Broadcom -4.5%; Meta -3.7%, Alphabet -2.5%. The FT put Nvidia's weekly loss at more than $300B of market value.
NVDA itself was the relative haven: -1.43% Friday to $214.13 (per the market view at /markets/NVDA/2026/10/19 — RSI 33, -10.5% off the 52-week high, $5.17T cap, still +5.7% above its 200-day). The fundamentals in that view remain the strongest in the chain: TTM revenue $302.97B (+105.9%), 74.7% gross margin, $21.4B quarterly FCF, 0.2x Debt/EBITDA, four straight EPS beats. WSJ notes the valuation paradox: NVDA's forward P/E has compressed from 23x to 15x even as the shares rose 15% — estimate revisions are outrunning the price. BofA counted $17.2B of weekly US equity-fund outflows, the biggest since March 2026.
The debate shifted from demand to returns. Eurasia Business News' wrap: "Investors were no longer focused only on how much AI demand could grow, but also on whether companies could earn sufficient returns on enormous infrastructure investments... The week's central issue was profitability, not simply enthusiasm for AI." Huang's counter, repeated through the week: AI will transform industries and "current investor concerns would pass." The macro overlay didn't help: 10-yr ~5.09%, ~74.3% odds of a December Fed hike (CME via AP), Brent +3.7% to $109.86.
The bull math got bigger in the same week. Morgan Stanley projects global AI capex of ~$1.58T by 2027, with Nvidia positioned to collect ~$690B of it (about two-thirds of the ~$1.03T flowing to S&P 500 companies) — a forecast Morgan Stanley raised ~14% after counting previously excluded lease obligations. It aligns with Nvidia's own guidance: CFO Colette Kress expects hyperscaler capex of $800B in 2026 rising to **$1.3T in 2027**, against a cloud-industry backlog exceeding $2T; Huang expects to sell twice the chip volume next year; FY2028 guidance implies 70% revenue growth ($673B).
What it means for the brief. The capex story's demand side is intact and still being revised up — the $1.3T 2027 hyperscaler number and the $2T backlog both grew this week. What broke is the financing side's cost: the credit layer now charges distressed-tier spreads for exactly the borrowers who were supposed to bridge capex to revenue1 (see Financing Layer: The "Darwin Moment" Arrives — CoreWeave Bonds Below 75 Cents, Oracle IG Paper at 88.5, Gartner's $4T Writedown Warning and the demand-base test in The Demand Base's Credit File Opens: Anthropic's BB+ Rating Revealed, a $100B Thanksgiving IPO Ask at ~$2T — and a 5% Public-Benefit Pledge). NVDA's own balance sheet is the cleanest in the chain — the stress is one and two steps downstream. The next hard tests are Microsoft's FQ1 print (Azure vs the ~45% cc guide) and NVDA's own Q3 FY2027 report on Nov 17 (guided $108B revenue, ~74% gross margin).
-
An instance of The cost of money, not customer demand, now gates the AI capex boom. — The week split the story exactly along the law's fault line — capex forecasts revised up to $1.3T while a ~5% ten-year and December-hike odds compressed the multiples and repriced the debt that bridges capex to revenue. ↩︎