Apple rents AI compute instead of owning: the rent-vs-own decision Ternus inherits Hyperscaler capacity capex to owned GPU + DC 1 frontier GPUs + datacenters allocates frontier edge to its own ecosystems Apple hybrid position holds capex low - rents compute own chips + servers, plus partners' infrastructure Siri AI rebuilt on Gemini beta 2 AI demand drives build-out rents compute rented compute: price and availability deteriorate as owners keep edge to self 3 Ternus decision rent vs own 4 Keep renting preserves operating margin stays exposed to partner pricing and priority 6 Pivot to ownership capex and R&D step up spending-vs-cash gap narrows cash-rich balance sheet 5 cash-rich, buyback-funded balance sheet
  1. 1Hyperscalers (Alphabet, Amazon, Microsoft, Meta) convert $145-205B each of 2026 capex into owned GPU/datacenter capacity
  2. 2Apple holds FY2025 capex to $12.7B and rents third-party compute, pairing its own chips and servers with partners' infrastructure and models (Siri AI is a ground-up rebuild running on Google's Gemini in beta)
  3. 3If AI compute demand keeps tightening capacity and hyperscalers allocate their frontier hardware and models to their own ecosystems, the price and availability of the compute Apple rents deteriorates
  4. 4That deterioration pushes Apple toward owning capacity
  5. 5If Apple pivots to ownership, its capex and R&D step up and the gap between its spending profile and its cash-rich, buyback-funded balance sheet narrows
  6. 6If it keeps renting, it preserves operating margin but stays exposed to partners' pricing and priority