As new Apple desktops begin shipping this week, the company's leadership is putting forward an unusual proposition to corporate buyers: these devices are cheaper than renting data centers.
Updated models of the Mac Mini and Mac Studio from Apple, which can cost nearly $20,000 (325,000 rubles) and are capable of performing artificial intelligence tasks locally, will compete with new desktop machines released by Nvidia and PC manufacturers. These machines are expected to be a central theme at the Microsoft Windows event in San Francisco next month.
The new Macs are aimed at more intensive AI tasks, such as coding or complex business work, without the need to pay 'tokens'—the basic unit of AI computation—to large cloud providers like OpenAI or Anthropic.
Apple hopes that its experience optimizing performance on battery-powered devices, such as the iPhone, will help it capture some market share traditionally held by Microsoft. However, the task is quite difficult: according to IDC data, Apple's share of the corporate desktop market is about 4.6%, while Windows holds a 91.3% share.
Apple co-founder Steve Jobs was known for his ambivalent stance on enterprise computing, as end-users could not choose their preferred products. Nevertheless, thanks to a long-standing commitment to energy efficiency, Apple is in a strong position to develop AI desktops.
When Apple released its first Apple Silicon chips in 2020, it combined two types of chips—compute and memory, which were separate in PCs—into a unified memory architecture to improve battery life. This close coupling between compute and memory, which Nvidia and others have only recently adopted, has resulted in the high efficiency of Macs in AI workloads. Apple began selling the Mac mini when the open-source tool OpenClaw became popular in markets, including China.
Exotic AI features
Although initially designed for content creators involved in video or music editing, over the past two years, Apple has subtly added exotic AI features, such as a specialized inter-chip network called RDMA via Thunderbolt.
At its presentation event this month, Apple demonstrated four Mac Studios connected together to run a trillion-parameter AI model—a measure of complexity—for the purpose of detecting and correcting errors in graphic code. Such tasks usually require a central server, but the Mac stack runs off a single outlet.
Jony Ive, Apple's Chief Hardware Officer, stated: 'Once you have a machine on your desk, you paid for it. And I believe we provide absolutely excellent value not just in terms of performance, but also in cost. No token costs. You just use the machine again and again.'
Microsoft aims to capture the same market that CEO Satya Nadella calls 'unlimited intelligence' of local AI. Nadella also announced that Microsoft plans to integrate many of its AI features into a Windows 'super app.'
However, Microsoft's historical dominance in enterprise computing means support from a vast number of hardware vendors, which can place an additional burden on Windows developers striving to maximize the capabilities of the chip.
In response to Microsoft's inquiry, the company stated that it works with chip manufacturing partners to optimize AI performance using Windows ML tools, and that features like RDMA are an active area of investment. Nvidia declined to comment, but at the launch of its new PC chip this year, CEO Jensen Huang downplayed intentions to directly compete with Apple, stating that Nvidia is focused on expanding capabilities that can be provided by Windows PCs.
Nvidia's strength remains in the data center sector. Ive noted that Apple offers corporate buyers the idea that AI models developed on its devices can scale up to the most expensive Mac Studio or down to the most accessible iPhone and iPad, since their chips are based on common principles and designs.