Apple's Abandoned Self-Driving Car Quietly Built the AI Chip Inside Every iPhone

Key points

• Apple's Neural Engine, the dedicated AI processor inside every modern iPhone and Mac, grew out of work on a self-driving car project that was later scrapped.
• The Neural Engine first appeared in the iPhone X and its A11 Bionic chip in 2017, handling face recognition, Animoji and augmented reality.
• Apple plans to skip the Pro and Max and Ultra variants of the M6 chip and move straight to the M7.
• The M7, due in the first half of 2027, will carry major Neural Engine upgrades.
• Apple's M7 Ultra is expected to power a new server product supporting up to 1.5 terabytes of RAM.

Apple spent years trying to build a self-driving car. The car never arrived. The engineering problem it posed, how to run powerful artificial intelligence directly on a vehicle's hardware without leaning on a distant data centre, planted a seed that grew into something far more lasting.

That seed is the Neural Engine, a dedicated processing unit built into every Apple chip since 2017. Think of it as a specialist worker inside the chip: while the main processor handles general tasks, the Neural Engine does nothing but AI calculations. As first reported by The Verge, citing Bloomberg's Mark Gurman, Apple engineers working on the car realised standard chips weren't going to cut it. The car processor was never finished. What survived was the Neural Engine.

It debuted inside the A11 Bionic in the iPhone X, with two jobs: Face ID and Animoji, the animated characters that mirror facial expressions. Modest beginnings, but they established something important. Apple carried the Neural Engine into its M-series chips, giving MacBooks and desktops the same on-device AI muscle as an iPhone. Running AI locally rather than sending data to Apple's servers is also why Apple can credibly claim stronger privacy than rivals who process everything in the cloud.

Now Apple's betting bigger on that foundation. According to Gurman, the company's skipping the Pro and Max and Ultra versions of the M6 chip entirely, accelerating straight to the M7. Expected in the first half of 2027, the M7 will bring substantial Neural Engine upgrades. The M7 Ultra, the top-end variant, should also underpin a new Apple server product capable of holding up to 1.5 terabytes of RAM. That's an enormous pool of fast memory, the kind needed to run the largest AI models without sending anything off-device. Memory supply at that scale is its own story: when we covered SK Hynix's $26.5 billion Nasdaq debut on 13 July 2026, the headline driver was exactly this kind of demand for chips that make AI work.

What does this mean for people who own Apple devices?

If you bought an iPhone or a Mac in the last few years, you already carry a Neural Engine. Features like voice transcription and photo organisation happen on the device itself, not in a data centre. The M7 should make those noticeably faster when it lands in 2027 hardware. For now, nothing changes, and that's actually the point: Apple's been building toward this quietly for nearly a decade, starting with a car nobody ever drove.