ThinkPatternGet the app
Perspective
BUSINESS · NOV 18, 2026

The AI Trade Now Sells Insulation, Not Upside

In eight weeks, Wall Street's advice flipped from "be long compute" to buying Exxon, GM, and a trash hauler as AI plays — while the credit side of the same market kept the buildout's financing open.

Goldman Sachs's 2027 outlook, out this week, tells clients where to hide from the trade the bank spent the year selling. The flagship picks are ExxonMobil, General Motors, and Waste Management — an oil major, a carmaker, and a trash hauler — selected not for what AI will do for them but for being "less vulnerable to AI disruption" than software or insurance, and for wage bills at least 30 percent automatable [1]. An AI portfolio whose marquee names are an oil company and a garbage collector is either a hedge or a confession. The turn took eight weeks, compressed. On November 3, the same bank told clients to hold more cash than the benchmark for the next three months — an "overweight" in portfolio terms — after the Nasdaq-100 had slid more than 10 percent from its October record. Its strategist put the reason in one sentence. [2]

AI risk has dominated cyclical risk. — Christian Mueller-glissmann

Six weeks earlier, on September 21, the bank's global co-head of fixed income was arguing the opposite, that AI infrastructure rather than government debt was the decade's asymmetric bet. [3]

I don't think the sustainability issue is really that credible in the long run. — Anshul Sehgal

The same institution went from "buy the buildout" to "hide from it" inside of eight weeks. It is not one bank's mood swing. On October 27, JPMorgan's private bank told its own clients to move from "AI enablers" to "AI adopters," balancing AI exposure with quality credit, Treasuries, and real assets [4]. This is the second descent of the year. In February, U.S. investors were already rotating out of the software giants — Alphabet and Amazon — into physical suppliers such as Caterpillar and Western Digital, which posted double-digit gains while the Magnificent Seven ETF fell 7.3 percent [5]. The banks have a stated reason, and it is worth taking in their own words rather than as settled fact. Goldman's published chain runs like this: AI capital spending growth will slow next year for the first time since the buildout began, squeezed by rising interest expense and a likely peak in hyperscaler spending — the handful of tech giants building most of the data centers that train and run AI models. The result, the bank forecasts, is S&P 500 earnings growth collapsing from 36 percent this year to 11 percent next [6][1]. The wry proof that Goldman believes its own thesis sits inside its own walls: it has spent roughly $6 billion integrating AI into its operations, and the technology now builds models and slide decks for its junior analysts [7]. The other side of the same market kept the financing open. The hyperscalers — Alphabet, Amazon, Meta, Oracle — raised $194 billion in bonds in the first half of the year, on pace for $250 billion, competing with sovereign governments for insurance and pension money; Alphabet sold a 100-year bond, an instrument that will still be paying out when the argument over 2027 capital spending is ancient history [8]. Japan's largest life insurers — Nippon Life, Daiichi Life, Meiji Yasuda — began lending overseas to data centers and power networks in late October, drawn, they say, by higher rates [9]. Spreads on bonds backed by data centers leased to hyperscalers have narrowed from roughly 180 to 146 basis points — hundredths of a percentage point — since late September [10]. JPMorgan projects $4.1 trillion of AI-related debt issuance by 2030 [11]. The same firm's name sits on both sides of this divergence. Even inside credit, the buyout firm Apax reports, the market is sorting, with issuers that hold data moats financing easily while over-leveraged software companies get squeezed [12]. Line the instruments up and one sorting rule appears. The frontier still gets growth pricing: Anthropic's debut this week still drew a record [13]. The middle of the stack is priced for contracts rather than dreams — Micron has locked roughly 35 percent of its revenue through 2030 and Sandisk about half its clients [14], while the resellers who rent out GPU computing power post losses [15] and delay their listings [13]. Sam Altman put it bluntly. [15]

And many of those people think they're going to sell most of the volume to us and Anthropic, and others. — Sam Altman

And the advice layer now sells insulation — which is what Exxon and Waste Management are for in this list. The escape hatch turns out to lead back into the machine. Copper, the archetypal hard asset, has tracked AI data-center stocks since late 2025, and it fell 11 percent into the selloff even as physical demand stayed tight [16]. Goldman itself offered a verdict on the drop. [16]

We see the recent correction as positioning-driven rather than reflecting loosening fundamentals. — Goldman Sachs Private Wealth Management

The marketed refuge — the physical economy — is coupled to the very trade it is sold as an escape from. NextEra Energy, the renewable-power giant, trades below its record even as utilities broadly lag [17]. None of this settles the number the whole flip rests on. Goldman's rotation assumes a 2027 peak in hyperscaler spending; that is the bank's attribution, not a fact. Nvidia's own forecast has the top five hyperscalers' capital spending rising from $800 billion this year to $1.3 trillion in 2027 [18]. Raymond James, more mildly, sees S&P earnings growth easing from 28 to 23 percent rather than collapsing [6]. It matters because AI capital spending has accounted for roughly half of S&P 500 earnings growth [19]. So the market has already reclassified the buildout — into financed infrastructure, priced for contracted certainty rather than growth — while the number under that reclassification is still contested. The advice side sells the exit; the funding side signed a 100-year bond. One is betting on a peak next year, the other on a toll road that outlives the argument.


Sources
  1. 1. Goldman Sachs Advises Shift Toward AI Adopter Stocks
  2. 2. Goldman Sachs Advises Overweight Cash to Hedge Market Volatility
  3. 3. Goldman Sachs Executive Favors AI Infrastructure Over Treasury Yields
  4. 4. JPMorgan Private Bank Urges Shift Toward AI Adopters
  5. 5. U.S. Investors Shift Capital Toward AI Infrastructure Companies
  6. 6. Goldman Sachs Warns of Sharp 2027 Earnings Slowdown
  7. 7. Goldman Sachs Invests $6 Billion to Integrate AI
  8. 8. AI Hyperscalers Raise $194 Billion Amid Market Volatility
  9. 9. Japan's Largest Life Insurers Expand Overseas AI Infrastructure Lending
  10. 10. PIMCO Predicts Tightening AI Infrastructure Bond Spreads
  11. 11. AI Infrastructure Faces Trillion-Dollar Risks and Physical Bottlenecks
  12. 12. Apax Partners Reports AI-Driven Split in Credit Markets
  13. 13. Anthropic Targets November IPO Amid Safety Concerns and Valuation Shifts
  14. 14. Micron and Sandisk Secure Long-Term Contracts Amid AI Chip Surge
  15. 15. AI Leaders Warn of Chip Overcapacity Amid Nvidia Instability
  16. 16. Copper Prices Hit April Lows Amid AI Stock Selloff
  17. 17. NextEra Energy Stock Lags Amid Utility Sector Shift
  18. 18. Nvidia Projects Hyperscaler Spending to Hit $1.3 Trillion by 2027
  19. 19. Goldman Sachs Urges Shift to Defensive Stocks Amid AI Slowdown

Keep reading in the app

The full perspective, free in the app.