Great piece. You really nailed the divide between personal and enterprise AI. I’m curious, though—what do you think the price tag will look like for an iPhone that can actually handle on-device AI? Are they going to have to lean into a dedicated chip to keep up? Also, how do you see Google playing this with the Pixel? Are they going to stay the course, or shift to mirror what Apple’s doing?
On pricing, I expect that the next iPhones will hold their current price points. The memory and silicon costs are going up, but Apple has more flexibility to absorb that than most analysts assume. They will take a smaller margin to defend their market position rather than risk pricing themselves out of the upgrade cycle, which matters more strategically right now than the per-unit margin. If a foldable arrives, that is where I expect the premium tier to land.
I guess that Apple expects services growth driven by Siri AI to more than offset any margin compression on the hardware side. I would be surprised if Apple charged a subscription for on-device AFM 3 access, since the on-device argument loses force the moment you put a paywall in front of it. What I do expect is that the more capable Private Cloud Compute features, the heavier reasoning, the larger context handling, the agentic work, get bundled into the Apple One tiers as a premium layer above what runs locally. That preserves the on-device privacy story for everyday AI tasks while giving Apple a recurring revenue line for the demanding ones. That is my read. We will see.
The dedicated chip question is one of the more interesting parts of this story. Apple Silicon was built for machine learning from the start, well before the current AI moment. The neural engine has been in iPhones for years, and the architectural decisions Apple made then are paying off now in a way that would be expensive for anyone else to replicate. AFM 3 Core Advanced, the 20-billion-parameter sparse model that only runs on more recent devices, only works because the silicon was already designed to do this kind of work efficiently. The strategy from Apple has been device infrastructure first, on-device personal context second, and the chatbot last. That ordering is what makes the architecture coherent.
Google has the same structural advantages with the Pixel. The TPU work goes back even further, and Gemini Nano running on a Pixel is, in principle, the closest thing to what Apple is doing. The complication is Android itself. The AI experience on a Pixel is one thing, on a Samsung something else, and on a Motorola something else again. Fragmentation cuts against the closed-loop ecosystem argument the way Apple has built it. I expect that Google will increasingly treat the Pixel as the reference implementation for what a true Google AI device looks like, and the rest of the Android ecosystem will continue to vary by manufacturer choices around model integration and privacy posture. If you want the full Google AI experience, you will want a Pixel.
The broader point is that the ecosystem argument cuts both ways. Apple's advantage is end-to-end control. Google's advantage is the model and search infrastructure. Whether Android's openness becomes a structural weakness in this race, or whether Google finds a way to enforce coherence across the ecosystem, is one of the more interesting questions for the next few years.
Great piece. You really nailed the divide between personal and enterprise AI. I’m curious, though—what do you think the price tag will look like for an iPhone that can actually handle on-device AI? Are they going to have to lean into a dedicated chip to keep up? Also, how do you see Google playing this with the Pixel? Are they going to stay the course, or shift to mirror what Apple’s doing?
Thank you, glad it landed.
On pricing, I expect that the next iPhones will hold their current price points. The memory and silicon costs are going up, but Apple has more flexibility to absorb that than most analysts assume. They will take a smaller margin to defend their market position rather than risk pricing themselves out of the upgrade cycle, which matters more strategically right now than the per-unit margin. If a foldable arrives, that is where I expect the premium tier to land.
I guess that Apple expects services growth driven by Siri AI to more than offset any margin compression on the hardware side. I would be surprised if Apple charged a subscription for on-device AFM 3 access, since the on-device argument loses force the moment you put a paywall in front of it. What I do expect is that the more capable Private Cloud Compute features, the heavier reasoning, the larger context handling, the agentic work, get bundled into the Apple One tiers as a premium layer above what runs locally. That preserves the on-device privacy story for everyday AI tasks while giving Apple a recurring revenue line for the demanding ones. That is my read. We will see.
The dedicated chip question is one of the more interesting parts of this story. Apple Silicon was built for machine learning from the start, well before the current AI moment. The neural engine has been in iPhones for years, and the architectural decisions Apple made then are paying off now in a way that would be expensive for anyone else to replicate. AFM 3 Core Advanced, the 20-billion-parameter sparse model that only runs on more recent devices, only works because the silicon was already designed to do this kind of work efficiently. The strategy from Apple has been device infrastructure first, on-device personal context second, and the chatbot last. That ordering is what makes the architecture coherent.
Google has the same structural advantages with the Pixel. The TPU work goes back even further, and Gemini Nano running on a Pixel is, in principle, the closest thing to what Apple is doing. The complication is Android itself. The AI experience on a Pixel is one thing, on a Samsung something else, and on a Motorola something else again. Fragmentation cuts against the closed-loop ecosystem argument the way Apple has built it. I expect that Google will increasingly treat the Pixel as the reference implementation for what a true Google AI device looks like, and the rest of the Android ecosystem will continue to vary by manufacturer choices around model integration and privacy posture. If you want the full Google AI experience, you will want a Pixel.
The broader point is that the ecosystem argument cuts both ways. Apple's advantage is end-to-end control. Google's advantage is the model and search infrastructure. Whether Android's openness becomes a structural weakness in this race, or whether Google finds a way to enforce coherence across the ecosystem, is one of the more interesting questions for the next few years.