Apple's AI chip and the vertical integration play

Apple is designing AI chips for servers.

Not the Neural Engine in iPhones and Macs, which has been doing on-device AI inference for years. Server chips. The kind that go in data centers. The kind that compete with NVIDIA.

This is Apple doing what Apple always does: controlling the entire stack.

The vertical integration argument

Most AI companies rent their compute from someone else. OpenAI runs on Microsoft Azure. Anthropic uses Amazon and Google cloud. They’re dependent on hardware they don’t control, manufactured by companies whose priorities aren’t perfectly aligned with theirs.

Apple doesn’t rent. Apple builds. The M-series chips in Macs. The A-series chips in iPhones. The Neural Engine for on-device AI. And now, server chips for cloud AI inference.

When you control the chip, the operating system, the device, the model, and the integration between all of them, you can optimize in ways that nobody else can. NVIDIA makes general-purpose AI chips that work for everyone. Apple can make AI chips that work perfectly for Apple.

The best model doesn’t always win. The best integrated experience wins. The iPhone camera isn’t the best camera sensor on the market. But the integration between sensor, ISP, software, and neural engine produces photos that beat phones with better sensors. The same principle applies to AI.

What this means for everyone else

If Apple runs its own AI inference, it doesn’t need NVIDIA GPUs for its server fleet. That’s a significant chunk of demand removed from the most supply-constrained market in technology.

It also means Apple can offer AI features that are faster and cheaper to run than competitors, because the silicon is designed specifically for their models. Latency goes down. Cost per query goes down. And the features that depend on server-side AI (Siri improvements, intelligent search, generative features) get better without getting more expensive.

The company with the deepest pockets, the most vertically integrated stack, and the largest consumer device install base (2+ billion active devices) is building custom AI silicon.

I don’t know if Apple will have the best AI model. Probably not. But I think they might have the best AI experience. And for consumers, experience is what matters.

The chip team in Cupertino just became the most important team at the most valuable company in the world. I’m watching what they ship.

Related thinking: