"Apple Price Increases, Apple Intelligence and the E.U." — @BenThompson
Why this is in the vault
Two interlinked structural arguments — memory scarcity forcing Apple's hand on pricing, and EU regulatory overreach blocking Apple Intelligence in the EU — directly feed the RDCO chip-cycle thesis and the broader platform-regulation landscape.
Issue contents
Schedule note: Thompson flags this is the last regular Update before the summer break schedule (no Updates June 29–July 2 and July 27–30; no Stratechery Interviews until September).
Apple Price Increases
Apple is preparing imminent device price increases driven by surging memory and storage chip costs. Tim Cook confirmed to the WSJ that price increases are "unavoidable," citing unsustainable cost pass-through from chip suppliers. TechInsights estimates the memory crunch would add ~$270 to the iPhone Pro price to hold margins flat.
Thompson's read: Apple held out as long as possible hoping the memory crunch would resolve, then ran out of runway. He frames this as structurally correct — Apple has already captured price-insensitive buyers, so eating margin for share doesn't work. More importantly, Apple has been quietly lowering prices in real terms for years (the 2025 $1,099 iPhone Pro equates to ~$892 in 2020 dollars), so nominal increases now come after years of inflation absorption.
The Services Narrative once gave Apple pricing cover; now it's the AI narrative — except this time the input cost is exogenous and out of Apple's control.
Apple Intelligence and the E.U.
Apple announced at WWDC that Siri AI will not launch in the EU initially. The stated reason: EU regulators, under a maximalist DMA reading, would require Apple to give any virtual assistant unrestricted direct access to user data and the ability to control all installed apps — with no privacy intermediary. Apple proposed a "Trusted System Agent" intermediary with an 18-month rollout; the European Commission rejected every proposal.
Additional technical observations from Thompson:
- Memory shortage hits Apple twice: higher unit cost AND higher on-device RAM requirement for large models (≥12GB for full Siri AI; 8GB runs only basic models).
- Apple's "powerful server models" with usage limits run on Nvidia hardware — Nvidia's Grace Blackwell supports Confidential Computing at multi-node rack scale, which aligns with Apple's Private Cloud Compute privacy architecture. This is likely why Apple returned to Nvidia after years of distance.
Thompson's verdict: He understands Apple's product integrity argument against EU's design mandates, but faults Apple for years of App Store rent-seeking that made regulators hostile. The likely outcome: Siri AI never launches in the EU. His preferred resolution is competitive pressure (rival hardware platforms, open AI access), not government-mandated interoperability.
Mapping against Ray Data Co
Memory capital-cycle thesis — strong confirmation. This piece is a live data point: Apple's CEO giving an emergency WSJ interview about memory costs in Q2 2026 is exactly the Phase 2 supply-squeeze signal the Markov phase-tracker is designed to capture. The $270 per-unit headroom estimate from TechInsights is a concrete pricing signal worth tracking.
AI hardware implications. The 8GB vs 12GB RAM dividing line for on-device model capability is a useful framework when clients ask about edge AI deployment constraints — it's a real-world RAM-tier segmentation with measurable capability consequences.
Platform regulation as capability risk. Apple Intelligence not shipping in the EU is a case study in regulatory friction as product gap. For RDCO clients in regulated industries building AI-integrated products, the DMA interpretation precedent (any AI system must get unrestricted device access) is a threat pattern worth understanding. This is a counter-argument to EU-first deployment strategies.
Anthropic footnote. Thompson mentions the Sharp Tech episode covering "the ongoing Anthropic export control saga" in the same breath as this newsletter. Anthropic is named as an upcoming Siri AI partner (alongside ChatGPT and Gemini). Monitor for any Anthropic-specific EU distribution constraints that could affect RDCO's model choices for EU-deployed agent work.
Related
- [[2026-03-09-stratechery-macbook-neo-thin-macbook-memory]] — Thompson's earlier piece on Apple/memory pricing; he lost a bet to Gruber on Mac price increases that this newsletter's price-increase confirmation now partially vindicates
- [[2026-06-11-stratechery-bajarin-apple-ai-compute]] — Bajarin interview on Apple's AI compute architecture and the chip-supply/inference dynamics; directly upstream context for the memory-drives-Siri-AI-limits argument here
- [[2026-05-17-cernbasher-your-ai-agent-called-it-wants-more-memory]] — memory-as-governor framing for agentic AI; RDCO MU position anchor that this Apple pricing signal reinforces