Despite anticipation for the presentation of the new foldable smartphone, the iPhone Ultra, attention should be paid to the development of Apple's smart home solutions. Although new iPhone models, including the iPhone 18 Pro and Pro Max, are expected at the September event, and the standard iPhone 18 is reportedly set for a later release, the company's true direction lies in home automation.
Apple has long shown indecision regarding the smart home, despite the existence of HomeKit since 2014, as well as the use of Apple TV and HomePod speakers as home hubs. The company actively supports the Matter standardization, but compared to Amazon and Google, whose AI displays have been products from the outset, Apple's ambitions seemed restrained.
The situation is changing: it has become known that a new Apple TV is being prepared, an updated HomePod mini, and most intriguingly, a smart home display unofficially named HomePad. This display, presumably about seven inches in square shape, will be equipped with a speaker, camera, and presence sensor, running on an operating system derived from tvOS, and will be available in both wall-mounted and desktop versions. HomePad shipments may occur sometime between October and the end of the year.
The Missing Piece
The key element that long hampered the development of Apple's home hub was the Siri system. The previous voice assistant proved insufficiently advanced to serve as the primary interface. This problem was solved in June with the release of Siri AI, which was rebuilt based on foundational models, partially running directly on the device, while more complex tasks are delegated to the Private Cloud Compute infrastructure.
The first wave of consumer AI boom primarily relied on cloud computing, where users entered highly personal information into massive models running in remote data centers owned by OpenAI, Google, and Anthropic. However, many tasks required for a smart home, such as turning off the kitchen light, checking the status of the garage door, or determining who is in the room, do not require the use of advanced models.
Most of the intelligence needed for a smart home is narrow, repetitive, and deeply personal. This includes knowledge about residents, their routines, data from cameras and microphones, the status of household appliances, and energy consumption. These are precisely the data people are least willing to transmit to a public cloud.
This leads to the hardware Apple unveiled on Tuesday without a special event. In the press release, Apple describes the new Mac mini as a means of 'enhancing the leading desktop computer for continuous agent computing.' Notably, the emphasis is placed specifically on continuously running AI agents, rather than video editing or photography.
The base model uses the new M6 chip with a dual 16-core Neural Engine—a first in history—and neural accelerators in every GPU core. Apple claims this computer processes large language model requests 4.8 times faster than the M4 mini and 13.5 times faster than the M1 released six years ago. The M5 Pro model can use up to 64GB of unified memory at 307GB/s, which, according to Apple, allows running 'even larger local AI models.' Furthermore, Thunderbolt 5 allows multiple Mac minis to be combined for working with large models entirely on the device.
Although LM Studio is not an Apple product, but an application for running other developers' language models on one's own hardware, Apple is now promoting the desktop computer in terms of its speed in handling third-party local AI. It is important to note that for this device to be significant, it does not necessarily need to be the official brain of Apple's smart home.
It is interesting what this device says about the trajectory of hardware development. For six consecutive years, Apple has steadily increased the energy efficiency of its own silicon. Central processors, graphics processors, neural processors, and specialized accelerators now use very fast unified memory in low-power and minimal heat-dissipating devices. The Mac mini is ideally suited for continuous operation at home, running local models.
However, there are some drawbacks: these machines are no longer as cost-effective as they once were, and paradoxically, AI is to blame for this. When the Mac mini with the M4 chip was released in October 2024, its price was $599 USD. Apple raised this price to $799 in June, and the new M6 model starts at $899—a 50% increase in less than two years. The M4 Pro model started at $1,399, and the M5 Pro machine costs $1,699.
Memory Shortage
The same trend of expanding data centers, stimulating demand for Nvidia accelerators, has led to memory shortages. According to TrendForce data, prices for standard DRAM rose by 90–95% quarter-over-quarter in the first three months of this year and another 58–63% in the second quarter, with further growth projected at 13–18% in the third. Tim Cook compared this to a '100-year flood of memory prices.'
Local model operation is primarily a memory issue, not a computational power issue, and AI is making the computers necessary to exit the AI cloud more expensive. Upgrading from 16GB to 32GB in the new Mac mini will cost $400, and 64GB for the M5 Pro costs $1,000.
In the era of generative AI, Apple was often portrayed as a lagging company. Siri was considered a joke in the industry, and the first launch of Apple Intelligence was imperfect, with the promised revision of Siri occurring with a noticeable delay.
Nevertheless, Apple has never earned the majority of its money from selling chatbots. Its strength lies in transforming technologies into hardware ecosystems that ordinary people actually buy. If AI moves from browser tabs and mobile applications into the physical environment around us, Apple will find itself in a much more familiar territory. It controls silicon, operating systems, phones, watches, TVs, speakers, and computers, possesses a well-established focus on privacy, and has hundreds of millions of customers owning several elements of this system.
The next Apple TV is expected to feature a significantly faster processor and Siri AI support. HomePad can place this intelligence on walls and desktops. HomePods provide microphones and speakers, while watches and iPhones provide identification, location, and personal context. Mac computers provide serious local computation when necessary.
For fifteen years, the smart home remained a hobby for people willing to tinker with hubs, automation, incompatible standards, and poorly designed applications. AI finally offers a way to hide this complexity behind the help of plain language. Instead of programming the house, the user communicates with it.
If Apple can ensure reliable operation of this system and keep a significant part of the interaction within the home, the smart home could finally move out of the niche of enthusiasts and become the main platform for mass-market computing. Therefore, while it is certainly worth watching the September 9th presentation for the foldable iPhone, one should pay closer attention to how the presentation shifts away from the phone. The most important product Apple creates may not fit in your pocket at all.