The tape split at Monday’s close: the Nasdaq added 0.4% to about 25,625, the S&P 500 edged up 0.2% to roughly 7,469, and the Dow slipped 0.4% to near 51,954. The fuel for everything green was a single report. The Information said Google is developing “Frozen v2,” a server chip that bakes parts of Gemini’s architecture directly into silicon and could serve six to ten times more tokens per unit of power than its current TPUs. Alphabet closed up about 2% after touching 3% at midday, and the chip complex ran with it — Micron up 5%, Intel 3.7%, Broadcom 2.8% — stacking a second leg onto the morning’s analyst-driven bounce. The iShares Semiconductor ETF finished up 1.7%.
So why did the Dow fall on a chip-rally day? Arithmetic, not sentiment. The Dow is price-weighted, and Caterpillar trades near $870 a share — roughly 10% of the entire 30-stock index. Its 1.2% slide erased about 62 Dow points by late morning, canceling almost exactly the 61 points Alphabet was contributing at its peak. Add Apple’s 2.7% breather after a 5% run last week, and “stocks fall” became Monday’s most-printed headline — describing an index where one machinery stock can outvote the day’s biggest story.
The backdrop didn’t improve; the market just priced it. The U.S. hit Iran for a ninth consecutive night, Brent briefly topped $90 before retreating on mediator diplomacy, and the national average gas price is back at $4 a gallon — which is why refiners Marathon Petroleum and Valero rose again and sit among the S&P 500’s biggest gainers of the past month. The real event risk starts Wednesday: Alphabet and Tesla report, with Intel close behind and more than 300 companies out by Friday. Tesla fell 2.1% into its print. The chip index, even after two green sessions, still sits roughly 19% below its June record.
Our take: Two things were true at the bell: the rally was real, and the index that made the headlines told you nothing. When one $870 stock is a tenth of a 30-stock average, the Dow stops being a market barometer and becomes a Caterpillar tracker — read the S&P and the semis instead. The bigger tell is what got bought. Frozen v2 is a bet that inference cost, not model quality, decides who profits from AI; if serving tokens gets 6–10x cheaper per watt, the capex panic that put chips in a bear market last week is asking the wrong question. But notice what the market paid for Monday: a chip reportedly targeted for 2028, in small volumes, wired for one company’s models. That’s conviction on a prototype timeline — 48 hours before actual earnings start grading the AI trade in public.
What to watch
- Alphabet’s Wednesday print. The capex line remains the number that started the chip selloff — and any comment on custom silicon now doubles as a Frozen v2 confirmation hearing.
- Caterpillar’s share price. Near $870 and a tenth of the Dow, it distorts every “Dow rises/falls” headline you’ll read this earnings season — a stock split would rewire the index overnight.
- Friday’s low on the chip index. Two green days haven’t ended a 19% drawdown. If hyperscaler guidance wobbles and that low breaks, the bear market resumes — this time with evidence.