AI's Two Trades: Coatue Spots a 405% vs. 2% Spread and Most Portfolios Aren't Hedging It

The October 7, 2026 edition of Coatue's Public Markets Update leans heavily on a single slide: the Goldman Sachs AI Winners basket versus the AI Losers basket, indexed to ChatGPT's public release on November 30, 2022. The winner stack is up roughly 405%; the loser stack is up roughly 2%. The 403-percentage-point gap between two equally AI-labeled equity baskets is the framework for everything else in the deck.

Coatue Management is reusing the chart, not originating it. The same dispersion shows up in a Coatue chart-of-the-day dated September 1, 2026 (\"2026 is a record year for stock dispersion, but strip out the top 10 winners and it looks like any other year\"), and in Philippe Laffont's May 6, 2026 deck where he argues market dispersion is at a level \"seen only twice before: 2008–09 and Covid.\" Coatue's authorship contribution is the framing: that the spread is not a coincidence but the central trade of the cycle.

Goldman's public long/short basket separate AI winners from losers — and the GS trading desk's long/short version of the same theme is up about 60% since launch while the S&P Software Index sits near zero, per public breakdowns of the trade. The winner list is dominated by infrastructure and AI-native software; the loser list by incumbents whose pricing power is being repriced by the same models that benefit the winners.

Three cohorts, three different economic exposures. The first cohort sells scarce inputs whose pricing power moves with hyperscaler demand. The second cohort monetizes the workflow software that AI agents must orbit around. The third cohort sells headcount, the unit AI agents first substitute for. Selling GPUs and selling bodies produces opposite stock-market outcomes once the model collision resolves.

NVIDIA's Q3 FY2026 results, reported November 19, 2025, are the densest single data point behind the winners basket. Total revenue of $57.0 billion was up 22% sequentially and 62% year over year; data-center revenue of $51.2 billion was up 66% year over year. By Q4 FY2026, reported February 25, 2026, revenue reached $68.1 billion (+73% YoY) and data center $62.3 billion. Fiscal 2027 has started with Q1 revenue of $81.6 billion and Q2 of $96.2 billion — the data-center-compounding machine that turns the 405% basket number into reality.

Memory is the second leg. Micron reported Q4 FY2026 (ended August 28, 2026) revenue of $54.2 billion on September 30, 2026, with gross margin holding near 81.6% and operating margin near 80% — a profile no other memory maker has ever sustained. Taiwan Semiconductor reported Q2 2026 revenue of NT$4.44 trillion on May 15, 2026, +36% YoY, with the AI-accelerator mix pushing operating margins toward 60%. Microsoft's FY2026 (fiscal year ending June 30, 2026) revenue reached $331.8 billion with a 40.3% net margin; the AI capex behind those numbers is the very force compressing the loser basket.

Salesforce is the cleanest case study in the loser cohort. The stock was trading near its 52-week low of $146.06 in 2026, having given back nearly 33% of its value year-to-date despite a Q1 FY2027 print on May 27, 2026 that beat expectations — $11.13 billion in revenue, up 13% year-over-year — and management raising FY27 guidance to $45.9–46.2 billion. Growth that would historically support multiple expansion is, in the Coatue/GS framing, the wrong kind of growth: incremental SaaS revenue at a time when AI agents are repricing the per-seat model.

Workday and ServiceNow tell the same story but from different angles. Seeking Alpha data from January 2, 2026 shows ServiceNow down 30% over the prior 12 months and Workday falling on continued AI-displacement narratives. Goldman Sachs's \"Will AI Eat Software?\" note on March 10, 2026 sized the addressable threat at the ~$400 billion global SaaS market facing more than 10,000 venture-backed AI-native entrants — the closest thing the cycle has produced to a parallel of late-1990s dial-up pressure on Yahoo. A 2% basket return over almost four years means an incumbent SaaS investor has earned roughly the coupon rate on a 30-day Treasury bill while losing the S&P 500 by more than 60 percentage points.

Coatue's May 6, 2026 update framed the trade explicitly: AI participants split into \"sellers of the shortage or buyers of the scarce equipment.\" Sellers of the shortage are the 12 names in the winner basket: GPU silicon, HBM memory, advanced packaging, networking optics, power equipment, cooling, the operating-system-level cloud platforms. Buyers of the equipment are the enterprises whose capex appears on Coatue's slide 29 — hyperscaler EBITDA growing at roughly 10% per year from a ~$1 trillion base funds the AI buildout comfortably through 2031.

The chain runs in three steps. First, hyperscaler AI capex (~$518–800 billion in 2026 consensus per Coatue's April 11, 2025 update — and likely to be revised higher again) flows into the revenue lines of NVIDIA, TSMC, Micron and Vertiv. Second, the same capex creates power and grid bottlenecks that pull in Eaton and NextEra. Third, the existence of these models depresses the terminal value of per-seat SaaS contracts — the asset class on which Salesforce, Workday and ServiceNow priced their forward multiples. Coatue's October 7 framing argues only the first two steps of that chain have been fully priced.

The clearest signal that the 405% vs 2% gap is structural rather than cyclical lives inside the Mag 7 itself. From the October 30, 2025 prior market peak through Coatue's May 2026 deck, slide 8 showed Google breaking out roughly +40%, Amazon near +20%, and one Mag 7 name down more than -20% — a 60-point internal spread in seven stocks that share the same megacap label. By September 30, 2026 only three of the seven were among the top-ten S&P 500 contributors despite the group representing about 35% of the index.

This is the dispersion trade compressed into the index's largest constituents. Coatue's own portfolio got clipped by it: the flagship fund fell 8.3% in July 2026 — the steepest single-month drop in more than a year — leaving the year-to-date return at 14.3% despite the dispersion call being correct. Even with the right view, the trade carries convexity risk: when the AI cohort sells off together, dispersion trades do not protect a long-only book. The Coatue July loss is itself the lesson in the deck.

Short term: NVIDIA reports fiscal Q3 FY2027 results in November 2026 (consensus revenue tracking near $54 billion, excluding H20/Russia-related adjustments); guidance commentary on Blackwell Ultra and Rubin platforms will set the ceiling on what 405% can compound into next. Microsoft reports FY27 Q1 at the end of October 2026 — capex commentary is the test of whether AI demand absorbs the next $80+ billion or whether the shortage narrative starts to wobble. Salesforce reports Q2 FY27 in late August with Agentforce bookings and the gross-retention narrative the swing factors on the loser basket.

Long term: Coatue projects AI capex could reach $2.8 trillion by 2028 (per Coatue note on capex funding, slide 29 of the May 2026 deck). If that path holds, the seller-of-shortage basket compounds into a structural revenue base that dwarfs today's valuation — even a generous multiple on NVIDIA's Q2 FY2027 $96 billion revenue annualized at the current 23x EV/Sales argues for the cohort's earnings stream continuing to re-rate upward. The risk is not that AI capex falls but that the loser cohort fails to discount — i.e., a 2% four-year compound on legacy SaaS is still rich if AI agents make seat-based pricing irrational within 18–24 months. The dispersion curve has two slopes; both still have somewhere to go.