Source: Apple is working on Apple Neural Engine, a dedicated chip to power AI on devices
Mark Gurman / Bloomberg :
Context & Ripple Effects
In May 2017, Bloomberg's Mark Gurman reported that Apple was designing the Apple Neural Engine, a dedicated processor for AI tasks on its devices rather than running them on general-purpose silicon. At the time it read as one feature roadmap item; the subsequent coverage shows it was the opening move in a much longer silicon strategy.
The through-line is visible across the years that follow: by 2024 Apple had extended the same approach into the data center with Project ACDC, a server chip for running AI software, and bought DarwinAI, whose specialty is shrinking AI models so they run faster on constrained hardware. By 2026, Apple planned to frame fifteen years of chip design as its edge for running AI locally at WWDC, including a distilled Gemini model.
First-order effects
- Apple gains a dedicated path for on-device AI workloads, moving inference off shared main processors and onto purpose-built silicon inside iPhones and other hardware.
- The report signals Apple's intent to own the AI layer of its devices end-to-end, reducing reliance on third-party components for machine-learning performance.
Second-order effects
- The device-side bet compounds into infrastructure: Project ACDC shows the same in-house silicon logic pushed into Apple's servers, so suppliers who once sold Apple general compute now face a buyer designing around them.
- DarwinAI's model-compression technology feeds the same loop — smaller, faster models make dedicated on-device hardware more useful, which raises the bar for rivals selling cloud-dependent AI experiences.
Third-order effects
- If the pattern holds, the industry splits between companies that integrate their own AI stack — chips, models, devices — and those renting each layer, with Apple's 2026 plan to demo local AI built on fifteen years of chip design as the template.
- Vertical integration of AI silicon becomes a moat measured in years of accumulated design experience, not capital — hard for late entrants to replicate regardless of spending.
The trend: Apple's 2017 neural engine report is the first data point in a decade-long shift from general-purpose mobile chips to vertically integrated AI silicon spanning devices, servers, and compressed models.