Apple Pitches Corporate Buyers on Macs Instead of Per-Token AI Bills

Apple is asking corporate buyers to consider a different way to pay for AI: buy Macs that run models locally rather than pay a cloud provider for each unit of use. In a Reuters interview, hardware chief Johny Srouji made the cost argument as the new Mac mini and Mac Studio reached customers on September 22.

The bill moves from usage to hardware

Srouji’s argument is about repeat use. Once a company owns a Mac, he told Reuters, it can keep running AI tasks without a per-token charge. Tokens are the units cloud AI providers use to measure work. Apple is targeting jobs such as writing code and handling complex business tasks—work that can otherwise produce an ongoing usage bill.

That is a pitch, not a demonstrated saving. The new Macs can cost nearly $20,000 in some configurations, Reuters reported. Whether ownership costs less depends on how much useful work a buyer gets from the hardware over time compared with what the same work would cost in the cloud. Srouji’s per-token point establishes a difference in how customers pay; it does not settle that comparison.

How a desk becomes a larger AI system

The hardware case starts with memory. Macs use unified memory, which places memory and computing resources close together. Apple says the new Mac Studio with M5 Ultra supports up to 512GB, giving it room for large models on one machine. That maximum-memory configuration is due in late October; it was not among the configurations arriving on September 22.

For work that needs more than one computer, Mac Studios can connect over Thunderbolt 5 using RDMA, a method that lets linked machines exchange data directly. Apple says this setup can make AI inference—running a trained model—up to three times faster than on a single Mac Studio. That is Apple’s performance claim, not a measure of savings against cloud services.

At its launch event this month, Apple linked four Mac Studios to run a trillion-parameter model that found and fixed a graphics coding bug, Reuters reported. The demonstration shows the kind of task Apple wants buyers to associate with a small local cluster. It does not show how that setup would perform, or what it would cost, across a company’s day-to-day workload.

The harder sale is the corporate desktop

Apple is not entering an empty market. IDC’s Linn Huang puts its share of enterprise desktops at about 4.6%, against 91.3% for Windows. Microsoft told Reuters it is working with chip partners to streamline AI work through Windows ML tools and investing in capabilities such as RDMA. Apple therefore has to sell a computing approach as well as a machine into a market where Windows already dominates.

Srouji also says models developed on Apple devices can move between its higher-end Mac Studios and cheaper iPhones and iPads because their chips share design principles. That range broadens Apple’s enterprise argument beyond a single desktop purchase. For buyers weighing it, the central question is more concrete: which recurring AI tasks can their own hardware handle well enough to justify the upfront bill?