Apple does not replace Gemini with an in-house frontier model for Siri before end of 2027

The prediction

Through December 31, 2027, Apple will continue to rely on Google's Gemini — not an in-house Apple foundation model — as the frontier-reasoning backend for Siri's most advanced features. The roughly billion-dollar-a-year arrangement unveiled at WWDC 2026 will still be in force, and Apple will not have shipped a generally available, Apple-built frontier model that displaces Gemini for Siri's hardest queries within the window.

Why I believe this

The Gemini-Siri deal turns Apple's most strategic capability into a recurring line item. That is not the kind of decision a company reverses in eighteen months. Once you have surrendered build-vs-buy on the highest-stakes capability you own, the forces that produced the surrender do not vanish — they compound:

1. The capability gap is structural, not temporary. Apple did not rent frontier reasoning because it lacked time; it rented because closing the gap on its own would have cost more, and shipped later, than the deal. Those economics hold for years, not quarters.
2. The deal buys breathing room, which reduces urgency. With Siri's advanced tier already working on Gemini, the internal pressure to crash-build a replacement drops. Dependencies that work tend to persist.
3. Switching is expensive and risky. Re-platforming Siri's reasoning onto an unproven in-house model — and eating the quality regression that a first in-house generation usually carries — is exactly the risk Apple just paid a billion dollars a year to avoid.

What would falsify it

This prediction is false if, before January 1, 2028, Apple ships a generally available Apple-built foundation model that serves as the primary frontier-reasoning backend for Siri's advanced features — replacing Gemini for those queries on shipping devices — or if Apple publicly terminates the Gemini arrangement in favor of its own model. A research demo, a beta limited to developers, or a model that only handles on-device lightweight tasks while Gemini still answers the hard queries does not count.

What would confirm it

The prediction is confirmed if, at the end of 2027, Siri's advanced features still route to Gemini (or the deal is still in force) and Apple has not shipped a generally available in-house frontier model that displaces it.

Confidence

Medium-high (68%). The strategic and financial logic strongly favors continuity, and two years is a short window for reversing a capability surrender of this size. The risk to the call is that Apple treats the Gemini deal explicitly as a bridge and races an in-house model to ship by late 2027 — possible, but it would mean closing on Apple's own timeline a gap it just declined to close on its own terms, which is why this stays a tier-2 call.