Apple pitches new Macs as no-cost-per-token AI machines

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22, 2026, Apple unveiled a new Mac lineup explicitly framed as AI infrastructure hardware, with high-end models priced at around $20,000. Apple's hardware chief used the phrase "no cost per token" to describe the machines, positioning them as a direct alternative to cloud-based AI services that charge ongoing fees for inference. The company named Microsoft and Nvidia customers as its primary targets, arguing that running AI workloads on local hardware eliminates the recurring data center rental bills those customers currently pay.

The core dispute is whether a $20,000 upfront purchase actually beats long-term cloud spending for a typical AI workload. That calculation depends heavily on usage volume, the types of models being run, and how quickly the hardware becomes obsolete, none of which Apple's pitch settles. The move also signals a broader strategic bet that on-device processing, built around Apple's unified memory architecture, can compete with the dedicated GPU clusters that currently dominate enterprise AI infrastructure.

The Pro-AI camp points to technical reasons why Apple hardware may be well suited to this moment. The account newhercules.bsky.social noted that the tight integration between compute and memory in Apple's chips, a design choice that predates the AI boom, had the incidental effect of making Macs capable AI machines, citing sell-outs of Mac Minis as the open-source agentic tool OpenClaw gained traction. The Anti-AI camp is less interested in Apple specifically and more focused on the broader infrastructure picture. Ed Zitron argued that the majority of AI capital expenditure, which he put at $791 billion, is unproductive, with roughly half of expensive Blackwell GPUs being warehoused rather than deployed, and estimated between $50 billion and $100 billion in warehoused AI equipment industry-wide. Accounts including spockresists.bsky.social dismissed AI itself as rebranded autocomplete used to inflate stock prices. The Middle Ground camp offered little direct engagement with the Apple announcement; the account donnha.bsky.social remarked only that the timing of related speculation seemed strange, which is consistent with what that camp typically does, reserving judgment until concrete performance and cost data are available.

The argument will sharpen once independent benchmarks compare these Macs against equivalent cloud inference costs at realistic workload volumes, and once Apple releases adoption figures for its new hardware line. Any update on whether enterprise buyers are actually switching infrastructure, rather than supplementing it, would also clarify whether Apple's pitch is landing.