Apple Discusses Its Advantage in On-Device AI, but the Interview Comes After Price Hikes for Mac mini and Studio
In cutting-edge AI labs, Macs are everywhere. This is what Doug Brooks, senior product manager of Apple silicon, shares in an exclusive interview with The Deep View, conducted by Jason Hiner ahead of the June 2026 WWDC and published on July 2. At the heart of the conversation is a statistic that Apple didn’t fully expect: the demand for Mac mini and Mac Studio has become, in Brooks's words, "incredible," and the two systems have turned into the preferred platform for AI-based agent workloads.
The reason, according to Brooks, lies in the type of need that AI agents bring with them. Those who run them autonomously seek "a system under their control," isolated from the main machine and capable of running 24/7. The compact and silent Mac mini fits this profile well. The trend is further supported by the fact that many AI tools are Mac-first or even Mac-only, a detail that has solidified the Mac as the reference platform among developers working on these projects.
It's Not Just About the GPU
Brooks emphasizes a point frequently reiterated in Apple's rhetoric about AI, presenting it in a more precise technical formulation: agent-based artificial intelligence is not just a GPU problem. "It’s no longer just about the GPU processing an LLM," he said. "It involves the entire chip, contributing to various parts of the task: tool-calling and everything happening around these workflows." In this view, the CPU, GPU, unified memory, and Neural Engine work together, each on a different portion of the process.
Apple Silicon features a Neural Engine designed for efficient matrix math, neural accelerators integrated into the CPU (originally born as "ML accelerators" for low-latency tasks like voice recognition), and, more recently, neural accelerators even within the GPU. It’s an architecture designed to distribute the load, not concentrate it on a single block of silicon.
The issue of power consumption enters the conversation with a figure that measures the ongoing change: agents are markedly increasing the quantity of processed tokens. Brooks cites estimates ranging from "three to ten times" higher than standard loads. A leap that, if confirmed by actual adoption, changes the type of hardware needed on both the cloud and client sides.
Towards a Hybrid Future Between Cloud and Device
On the local front, Brooks notes that models with 70 and 120 billion parameters are already running on laptops, thanks to modern optimization and quantization techniques. The trend toward on-device AI, he explains, is driven by three factors: privacy, security, and the rising costs of cloud inference. His prediction is of a hybrid future, where the agents themselves will decide, on a case-by-case basis, what to execute locally and what to delegate to the cloud.
Brooks describes this direction with the term "transparent AI": functionalities distributed within the operating system and third-party apps that work silently, without explicitly presenting themselves as artificial intelligence. He cites examples such as Draw Things, for image generation on iPhone, iPad, and Mac, and SwingVision, which analyzes tennis and pickleball matches in real-time through the iPhone's cameras.
However, there’s a detail (or rather, a true elephant in the room) that weighs on the reception of the interview. Recorded before WWDC, it was published after Apple had already raised the prices of Macs and iPads by several hundred euros. At this point, Brooks's assertion about the "compelling price-performance" of the Mac mini and Mac Studio sounds considerably less convincing today: the Mac Studio with M3 Ultra has risen from €5,099 to €6,399, and the Mac mini with M4 Pro has seen a €250 increase, now priced at €1,929. While the reasoning about the architecture may remain valid, the new pricing makes the offer decidedly less accessible.