TL;DR
The basket's divergence has moved from balance sheets to income statements: the hyperscaler capex boom is now visibly inflating Apple's input costs, the group's headline profits are increasingly unrealized marks on private AI stakes, and Tesla just posted its first upward estimate revision in years. Meanwhile Michael Burry has turned GPU depreciation into the sharpest bear argument against the whole buildout — and Nvidia's own investor deck walked into it.
The capex arms race found its victim: Apple's income statement
The divergence inside the basket is no longer coincidental — it's causal. The June-quarter capex spread was the widest on record: Amazon $54.21B, Alphabet $44.92B, Microsoft $35.80B, Meta $30.12B — against Apple's $2.46B, one twenty-second of Amazon's capex-divergence-hyperscalers-vs-apple
. The hyperscalers' memory buying is now Apple's cost shock.
"Bernstein now sees iPhone gross margin at 40.4%, down from 45.0% a year ago, as memory costs jump."
— Exencial Research (Bernstein note summary), Oct 3, 2026
The per-unit math is brutal: 12GB DRAM at "$136.62, up about 360%," and on the Pro Max "the gross profit per phone falls from $515 to $437" capex-divergence-hyperscalers-vs-apple. Morgan Stanley calculated a $200 iPhone price increase was needed to hold margins; Apple chose $100 — deliberately trading margin for volume, with Tim Cook calling increases "unavoidable." Meanwhile Nvidia collects both the hyperscaler capex and the memory-scarcity windfall while Apple pays the bill.
What to watch: Bernstein's December-quarter company gross margin estimate of 46.2% sits about 1.2 points above the registered AAPL tripwire at 45% — one bad print from firing.
Tesla flipped the estimate trajectory — the margin test is now the story
Tesla just posted its first clean beat in months, and it did the one thing this brief tracks: it moved consensus. Q3 deliveries of 486,532 came in roughly 29,600 units above Visible Alpha's 456,896, and full-year forecasts rose to 1.82 million from 1.65 million in the June consensus tesla-existential-autonomy-capex-pivot
.
"The strong numbers put Tesla on track for full-year deliveries growth following two years of declines. I point to FSD (Full Self-Driving) as being a differentiator that drives consumers to choose Tesla over other autos."
— Morningstar's Seth Goldstein, via Reuters, Oct 2, 2026
But the P&L underneath is still the widest divergence in the basket: a 1.4% operating margin against Microsoft's 45.1%, negative $1.09B quarterly FCF, and a 346x P/E tesla-existential-autonomy-capex-pivot. The volume recovery arrives just as memory-cost inflation and price competition squeeze auto gross margin — the same externality hitting Apple, on a business with a fraction of Apple's pricing power.
What to watch: The October 21 print — whether auto gross margin holds under memory inflation, and whether the Semi ramp raises capex guidance.
Meta is re-rating a business line that doesn't exist on the P&L yet
Meta is now the clearest case of price and analyst sentiment decoupling — upward — from reported fundamentals: up 26% in a month on a Q2 EPS miss, with TTM earnings down 13.4% and operating income compressing since December meta-muse-consumer-ai-agent
. The Muse adoption numbers are doing the work: 2.8 million U.S./Canada downloads in 12 days, versus 1.3 million for ChatGPT on a comparable basis.
"Muse reached 2.8 million downloads in the U.S. and Canada during its first 12 days, according to Apptopia."
— Barchart, Oct 2, 2026
The revision wave is broad — Deutsche Bank to $820, JPMorgan to $920, consensus now 45 of 54 at Strong Buy — but scenario math is the heavy lifter: Jefferies models $10.8B a year at ≥3% paying penetration of a billion users, while BofA says meaningful revenue "probably won't arrive before 2028" meta-muse-consumer-ai-agent. Q2 FCF was just $1.75B against a $130–145B capex guide — Muse revenue is doing all the valuation work.
What to watch: Whether Q3 estimates catch up to the multiple, and whether Meta guides 2027 capex up again to serve Muse demand.
The earnings-quality question is about to get a market price
The basket's headline profits are increasingly not operating profits. Alphabet's $112.19B June-quarter net income included a ~$98B unrealized SpaceX gain; Amazon's $62.65B included ~$53B on Anthropic — and Anthropic is planning what Reuters calls potentially the largest IPO in history, targeting a $2 trillion valuation ai-stake-paper-gains-earnings-quality.
"In just the two years ending in 2025, the company lost over $50 billion, but it has more than $500 billion in spending commitments in the coming years."
— Reuters, Sept 30, 2026
The entanglement cuts both ways: 47% of Anthropic's 2025 revenue ran through big-tech partners who are also its compute suppliers — and could "undermine it by constraining or competing with it directly" ai-stake-paper-gains-earnings-quality. Layer on Burry's depreciation critique — that Nvidia's retained-value chart is a DCF of rental income, not resale prices, and that accelerated depreciation is "an implicit admission of how severely each new product generation devalues its predecessors" gpu-depreciation-debate-burry-vs-nvidia
— and there are now three separate channels (marks, useful lives, FCF squeeze) through which headline AI earnings overstate economic profit.
What to watch: Nvidia's November 20 print, with gross margin guided down to 74% on memory scarcity — the first hard data point in the depreciation fight.
What surprised us
- Burry didn't short Nvidia stock — he bought puts. He covered shorts across the AI complex and "replaced the Nvidia short with September 2027 puts at strikes in the mid $100s" — a dated, leveraged bet that the repricing comes late, not now gpu-depreciation-debate-burry-vs-nvidia.
- Apple chose volume over margin on purpose. Morgan Stanley's math said $200 was needed; Apple took $100 — meaning the memory shock is being absorbed in gross profit rather than passed through, which makes the December-quarter margin tripwire live sooner capex-divergence-hyperscalers-vs-apple.
- Microsoft is the quiet outlier on capital allocation. It's the one hyperscaler still expanding shareholder returns — dividend up 8% to $0.98 — while running a $19.64B FCF surplus against $35.80B of capex capital-allocation-and-free-cash-flow-divergence.
- Tesla's recovery isn't American. EU registrations rose about two-thirds January–August, FSD is approved in eight European countries, and the robotaxi now runs without an in-car safety supervisor in Texas and Florida tesla-existential-autonomy-capex-pivot.
Open threads worth a vote
The Mag 7 Divergence
Since last time
Promoted
- Nothing — no prior minor thread became a core section. The two new core sections (Apple's cost shock, earnings quality/Burry) are entirely fresh topics.
Escalated
- Tesla: last time the story was the Semi factory opening; now it's the estimate trajectory flipping — first clean delivery beat in months and the first upward consensus revision in years. The margin test is now the framing.
- Meta: last time it was the Muse launch and adoption curve; now it's a full decoupling story — price targets and sentiment running far ahead of a P&L where Muse earns nothing yet, with hard download comparisons vs. ChatGPT.
Demoted
- The Tesla Semi ramp: was a full section; now survives only as a clause in the Tesla "what to watch" (whether the ramp raises capex guidance).
- The hyperscaler capex figures (Amazon $54.21B, Alphabet $44.92B): previously evidence in the Alphabet equity-sale story; now inputs to the new Apple cost-shock section.
Disappeared — check these before assuming continuity; none appear anywhere in this briefing:
- The Fed hike and rate-shock framing entirely: the September 16 hike, the 10-year at 5.18%, the funding-cost tier as the key macro variable, and the consensus dispersion targets (46% Nvidia upside, negative Apple upside).
- Alphabet is selling equity: the Berkshire $10B private placement and the negative-FCF/buyback inversion are gone — as is the related open thread.
- The intra-basket Muse collision: Amazon blocking Muse from its marketplace and Muse as a rival search channel for Alphabet — both gone, along with the "agentic AI dents Google Search" thread.
- Previous surprises with no follow-through: Apple ignoring the record $5.7B Taction patent verdict, and the Nvidia China/MIIT RTX Pro 5500 catalyst (that thread is also gone).
- Apple at its 52-week high and Alphabet's 17.27 P/E as the cheap-name counterpoint.
Unchanged
- No section carries over intact. The Tesla P&L stats (1.4% operating margin, −$1.09B quarterly FCF, ~346x P/E) and Meta's rally-on-an-EPS-miss pattern are the only substantive repeats — skim those sentences below and move on.
[NEW] The capex arms race found its victim: Apple's income statement
The divergence is no longer coincidental — it's causal. The June-quarter capex spread was the widest on record: Amazon $54.21B, Alphabet $44.92B, Microsoft $35.80B, Meta $30.12B — against Apple's $2.46B, one twenty-second of Amazon's capex-divergence-hyperscalers-vs-apple. The hyperscalers' memory buying is now Apple's cost shock.
"Bernstein now sees iPhone gross margin at 40.4%, down from 45.0% a year ago, as memory costs jump."
— Exencial Research (Bernstein note summary), Oct 3, 2026
The per-unit math is brutal: 12GB DRAM at "$136.62, up about 360%," and on the Pro Max "the gross profit per phone falls from $515 to $437" capex-divergence-hyperscalers-vs-apple. Morgan Stanley calculated a $200 iPhone price increase was needed to hold margins; Apple chose $100 — deliberately trading margin for volume, with Tim Cook calling increases "unavoidable." Meanwhile Nvidia collects both the hyperscaler capex and the memory-scarcity windfall while Apple pays the bill.
What to watch: Bernstein's December-quarter company gross margin estimate of 46.2% sits about 1.2 points above the registered AAPL tripwire at 45% — one bad print from firing.
[Escalated] Tesla flipped the estimate trajectory — the margin test is now the story
Last time the hook was the Semi plant; the new fact is that Tesla moved consensus. Q3 deliveries of 486,532 came in roughly 29,600 units above Visible Alpha's 456,896, and full-year forecasts rose to 1.82 million from 1.65 million in the June consensus tesla-existential-autonomy-capex-pivot.
"The strong numbers put Tesla on track for full-year deliveries growth following two years of declines. I point to FSD (Full Self-Driving) as being a differentiator that drives consumers to choose Tesla over other autos."
— Morningstar's Seth Goldstein, via Reuters, Oct 2, 2026
The P&L underneath is unchanged from what you read last time — still the widest divergence in the basket: a 1.4% operating margin against Microsoft's 45.1%, negative $1.09B quarterly FCF, and a 346x P/E tesla-existential-autonomy-capex-pivot. The new tension: the volume recovery arrives just as memory-cost inflation and price competition squeeze auto gross margin — the same externality hitting Apple, on a business with a fraction of Apple's pricing power.
What to watch: The October 21 print — whether auto gross margin holds under memory inflation, and whether the Semi ramp raises capex guidance.
[Escalated] Meta is re-rating a business line that doesn't exist on the P&L yet
The Muse story has moved from adoption to decoupling: up 26% in a month on a Q2 EPS miss (essentially the same pattern you saw last time, now larger), with TTM earnings down 13.4% and operating income compressing since December meta-muse-consumer-ai-agent. What's new is the hard comparative: 2.8 million U.S./Canada downloads in 12 days, versus 1.3 million for ChatGPT on a comparable basis.
"Muse reached 2.8 million downloads in the U.S. and Canada during its first 12 days, according to Apptopia."
— Barchart, Oct 2, 2026
The revision wave is broad — Deutsche Bank to $820, JPMorgan to $920, consensus now 45 of 54 at Strong Buy — but scenario math is the heavy lifter: Jefferies models $10.8B a year at ≥3% paying penetration of a billion users, while BofA says meaningful revenue "probably won't arrive before 2028" meta-muse-consumer-ai-agent. Q2 FCF was just $1.75B against a $130–145B capex guide — Muse revenue is doing all the valuation work.
What to watch: Whether Q3 estimates catch up to the multiple, and whether Meta guides 2027 capex up again to serve Muse demand.
[NEW] The earnings-quality question is about to get a market price
This section didn't exist last time, and it changes what the basket's headline numbers mean. The group's profits are increasingly not operating profits: Alphabet's $112.19B June-quarter net income included a ~$98B unrealized SpaceX gain; Amazon's $62.65B included ~$53B on Anthropic — and Anthropic is planning what Reuters calls potentially the largest IPO in history, targeting a $2 trillion valuation ai-stake-paper-gains-earnings-quality.
"In just the two years ending in 2025, the company lost over $50 billion, but it has more than $500 billion in spending commitments in the coming years."
— Reuters, Sept 30, 2026
The entanglement cuts both ways: 47% of Anthropic's 2025 revenue ran through big-tech partners who are also its compute suppliers — and could "undermine it by constraining or competing with it directly" ai-stake-paper-gains-earnings-quality. Layer on Burry's depreciation critique — that Nvidia's retained-value chart is a DCF of rental income, not resale prices, and that accelerated depreciation is "an implicit admission of how severely each new product generation devalues its predecessors" gpu-depreciation-debate-burry-vs-nvidia — and there are now three separate channels (marks, useful lives, FCF squeeze) through which headline AI earnings overstate economic profit.
What to watch: Nvidia's November 20 print, with gross margin guided down to 74% on memory scarcity — the first hard data point in the depreciation fight.
What surprised us
- [NEW] Burry didn't short Nvidia stock — he bought puts. He covered shorts across the AI complex and "replaced the Nvidia short with September 2027 puts at strikes in the mid $100s" — a dated, leveraged bet that the repricing comes late, not now gpu-depreciation-debate-burry-vs-nvidia.
- [NEW] Apple chose volume over margin on purpose. Morgan Stanley's math said $200 was needed; Apple took $100 — meaning the memory shock is being absorbed in gross profit rather than passed through, which makes the December-quarter margin tripwire live sooner capex-divergence-hyperscalers-vs-apple.
- [UPDATED] Microsoft is the quiet outlier on capital allocation. It's the one hyperscaler still expanding shareholder returns — dividend up 8% to $0.98 — while running a $19.64B FCF surplus against $35.80B of capex. This extends last briefing's funding-model split, now with the survivor named capital-allocation-and-free-cash-flow-divergence.
- [UPDATED] Tesla's recovery isn't American. EU registrations rose about two-thirds January–August, FSD is approved in eight European countries, and the robotaxi now runs without an in-car safety supervisor in Texas and Florida — new geographic texture on the pivot story you already know tesla-existential-autonomy-capex-pivot.
Open threads
Closed or absorbed from last time — all three prior threads are gone:
- Nvidia RTX Pro 5500 China approval — disappeared entirely; no update in this briefing.
- Agentic AI dents Google Search — disappeared; the Muse/Alphabet collision is no longer tracked here.
- Alphabet Q3 Berkshire-backed capex raise — disappeared along with the Berkshire equity-sale story itself.
Live threads now: