For an AI loser, Apple had plenty of wins last year.
The Apple Intelligence implementation mess was certainly embarrassing, but the company nonetheless did what it does best: sell iPhones. With news this week that Apple will be using Gemini models to power the long-awaited, smarter Siri, it appears that Apple has made huge strides in the overall AI race. However, it still faces a major challenge, and Apple is not yet out of business.
Apple Intelligence got off to a well-documented rocky start in 2024. iPhone 16 was “built for Apple Intelligence” but shipped without it. Features arrived over the next few months, but the so-called smarter Siri never arrived. Apple’s management admitted it went back to the drawing board, the people responsible for the project were shuffled around, and it all looked like a major failure on Apple’s part.
However, it doesn’t look like people are ready to ditch their iPhones for Gemini-powered Android phones. According to IDC Q3 2025 Report“Demand for Apple’s new iPhone 17 lineup has been strong, with pre-orders exceeding orders for the previous generation.” Counterpoint Research Calls Apple Global Smartphone ‘Market Leader’ in 2025 10% market share increase year on year. Meanwhile, Apple Intelligence is much less perceptible in the marketing of the iPhone 17 than it was with the iPhone 16; you have to scroll halfway down the iPhone 17 product page before you get to the first mention.
The delay tactic has proven effective, but now investors panic if AI isn’t mentioned every five minutes. Apple had to come up with something something by strategy, and in the second half of 2025 we started hearing reports that yes may look for external partnersinstead of building your own models from scratch. This wouldn’t be entirely unprecedented; Apple already allows users to access ChatGPT directly in iOS, and has promised from the beginning that it will add more external LLMs this way. But this week’s offer isn’t about adding a quick way to chat with Gemini on iPhone. You can already do this in the Gemini app. The idea is to build a smarter Siri on Google models and run it all on Apple Compute’s private cloud. If and when a smarter Siri arrives this year, it’ll have some solemn Gemini DNA.
You can argue that Apple made the right decision from a business standpoint, but was it the right decision Apple transfer? To consider Tim Cook’s own words on the 2009 earnings call: “We believe that we must own and control the core technologies used in the products we make…” This was the basis of the company’s efforts to develop its own silicon, which turned out to be an absolute winning strategy. But either Apple believes that AI models are not a core technology at all – they are more like a core service on which products will be built – or it has seriously misjudged AI in the face of another platform change and risks being left behind. Low stakes stuff!
That’s the challenge: transforming Apple Intelligence into a product that people actually want, not one they are indifferent to
And you need to be sure that Apple doesn’t control the fate of every part of the iPhone. It didn’t go out and build a search engine, its own wireless network, or an algorithmic social media platform. All of these things work on the iPhone, but they’re not a core part of the iPhone’s identity, and the AI could very well end up the same way. Perhaps there’s a clue in the way Apple has seemingly moved away from encouraging developers to adopt its own App Intents framework using Anthropic’s MCP as the basis for agentic features. If AI just needs to find the right hooks to get the job done, the specific models it runs on are less vital. But it all depends on the product Apple builds around artificial intelligence — and that starts with Siri.
This is the real challenge facing Apple: transforming Apple Intelligence into a product that people actually want, not one they are indifferent to. It needs to turn Siri into something the company promised all along, not a glorified timer-setting machine. Apple can make a pretty product – no doubt about it. Can it do this without having control over its own models? Can it do it faster than Google, Jony Ive or any other competitor ready to attack the garden walls? The contract may be signed, but the real work begins now.