06-reference

stratechery apple earnings hardware regression

2026-11-03·reference·source: Stratechery·by Ben Thompson
stratecheryappleearningsmemory-shortagememory-cyclesupply-chainsiri-aiservices-businesshardware

Apple Earnings, Memory Problems, The Hardware Company Regression (Stratechery Update, 2026-11-03)

Why this is in the vault

One continuous argument carried across three section headers, not three separate items: Apple's Q4 FY2026 beat and hardware launches are strong enough that the memory-cost crisis is pushing the company back toward being judged as a hardware maker rather than the "services company" Thompson has tracked since 2016 — the latest, and sharpest, data point in the memory-cycle thread this vault has followed across three prior quarters.

The core argument

Apple earnings: Apple beat on hardware demand (iPhone Duo, new MacBook Pro with touch, iPad mini OLED) but guided down on gross margin for the December quarter, and shares fell up to 4% post-call. Ternus's rookie earnings call leaned entirely on the hardware launch slate as evidence the "innovation engine" is fine.

Memory problems: Memory costs are not stabilizing — CFO Kevan Parekh says September-quarter margins fell ~1pp from memory even after June price increases; December-quarter margins will fall another 1.5–2pp despite September iPhone and October Mac/iPad price hikes. The hard number: Apple's manufacturing purchase obligations jumped $33B quarter-over-quarter (June→September) to lock in memory supply — Apple has flipped from dictating terms to suppliers to having to forecast demand and carry the inventory risk itself. COO Sabih Khan confirms Mac/iPad supply constraints for December are now memory-driven, not advanced-node (SoC) driven as they were in Q4.

The hardware company regression: Thompson's throughline — demand for Apple hardware keeps beating the company's own forecasts even as prices rise, which is the best evidence yet that Apple could raise prices further without denting demand (he's wondered this since 2013). The flip side is Siri AI: Ternus fields several questions on its weak agentic capability and falls back to "Apple devices are the best places to use all forms of AI" — an implicit concession that Apple's edge may end up being the hardware, not its own AI layer. Thompson's personal read: he wants Apple's hardware without Apple's software/security constraints getting in the way of his agents, and flags this as a real (if still hypothetical) tension between Apple's integration commitment and its hardware upside.

Mapping against Ray Data Co

Hard data point for the memory-cycle capital-cycle thesis (project_investing_markov_capital_cycle): a $33B single-quarter jump in Apple's manufacturing purchase obligations to secure memory supply is exactly the kind of anchor data this thesis needs — it's capex-commitment evidence of accelerating (not peaking) memory scarcity, extending the same thread tracked in the 02-04, 05-06, and 08-10 notes below. Secondary but notable: the Siri-AI-vs-agentic-capability gap Ternus got pressed on is the same "agents in production" lens RDCO uses for credibility-building (project_credibility_for_phdata_sales) — Apple is the largest live example of a company whose own AI layer lags the agentic capability bar, while outsourcing "all forms of AI" to the device.

⚠️ Sponsorship

"Presented by" slot continues to Stripe — present in both the top banner and a mid-Update plug for Stripe's SaaS-revenue data (Stripe Economics Substack), with a utm_source=stratechery&utm_campaign=paid-web tracked link. This is the fifth consecutive Stratechery-family issue carrying Stripe since the 2026-10-21 sponsorship announcement (10-21, 10-22, both 10-23 issues, now 11-03) — standing default sponsor, no break. Bias implication: none specific to this issue's Apple/memory content — Stripe's business doesn't intersect Apple earnings — but the standing financial relationship is the same disclosed conflict Thompson flagged in the original announcement.

Related