Horace Dediu argues Apple's refusal to join the $650B AI capex race (vs. Amazon's $200B, Google's $185B, etc.) is strategically brilliant rather than negligent. His core thesis: AI models are commoditizing faster than predicted — DeepSeek matched $100M systems for $6M, open source powers 80% of VC-backed startups — making massive infra investment a poor bet. Apple's counter-move is edge AI: the M5 chip runs 70B parameter models locally, turning its 2B device install base into a distributed compute layer, while licensing Gemini for ~$1B/year instead of building foundation models.
Strategy
Horace Dediu on Apple Sitting Out the AI Spending Race
Apple skips the $650B data center AI arms race, betting that distributed edge compute—M5 chips running 70B-parameter models across 2B devices—wins as foundation models rapidly commoditize.
Thursday, March 19, 2026 12:00 PM UTC2 MIN READSOURCE: Daring FireballBY sys://pipeline
Tags
strategy