TL;DR
Apple's privacy-first AI runs on Google's models at $1B/year — MSM covers the partnership, X questions whether privacy branding survives the plumbing.
MSM Perspective
Bloomberg's Mark Gurman first reported the $1B annual figure, framing it as a pragmatic shift.
X Perspective
X frames the deal as Apple renting intelligence it refuses to train, raising questions about where Apple's models end and Google's begin.
Apple's rebuilt Siri and next-generation Apple Intelligence features run on Google's Gemini models under a deal valued at approximately $1 billion per year [1]. The partnership, first reported by Bloomberg's Mark Gurman, represents a seismic shift in Apple's model strategy: own the interface, rent the intelligence.
The $1B figure names what privacy costs when you refuse to train your own frontier model. Apple has positioned Apple Intelligence as privacy-first, processing data on-device. But the underlying models that power Siri's new capabilities come from Google's infrastructure [1]. The plumbing contradicts the branding.
On X, the deal reads as Apple admitting it cannot compete at the frontier of model training while maintaining its privacy constraints. MSM covers the partnership as a pragmatic business arrangement. Neither frame captures the structural tension: Apple's privacy promise depends on Google's willingness to supply the intelligence at a price that scales with adoption [1].
The question is not whether the deal works technically. It is whether consumers perceive the difference between Apple's interface and Google's models, or whether "Apple Intelligence" becomes a brand name for someone else's work.
-- THEO KAPLAN, San Francisco