The latest strain on Apple's Mac supply isn't coming only from consumers. OpenAI and Anthropic are using compact Macs for AI agent work, and Apple has had to refresh the line while buyers are still waiting.
OpenAI has spent the past several months buying tens of thousands of Mac minis and Mac Studios for reinforcement learning and computer-use agents, according to The Information. Anthropic has also been renting Mac minis through Amazon Web Services for similar work, the report said. That is the story. Two of the companies spending hardest on AI infrastructure are leaning on the same small Apple desktops you can put under a monitor at home.
The point is not that a Mac mini replaces an H100. It does not. Nvidia still owns the market for the large training runs that require huge parallel compute. But agents that watch a screen, keep a desktop environment in memory, click through a browser and learn from repeated attempts create a different kind of demand. You need machines that can run many local sessions at once, and Apple's unified memory gives developers a practical way to keep large models and desktop tasks on the same system.
That is not a footnote. A Mac mini with enough unified memory can run local models too large to fit in a consumer GPU's own VRAM. Nvidia's RTX 5090, for example, lists 32GB of GDDR7 memory, while a 70 billion parameter model at common 4-bit quantization often needs roughly 40GB or more. The Mac is slower for some jobs, but capacity matters when the model has to load at all.
Apple Felt It #
Apple was already warning investors before this week's product refresh. On the company's fiscal second-quarter earnings call on April 30, Tim Cook said the Mac mini and Mac Studio could take several months to reach supply-demand balance. "Both of these are amazing platforms for AI and agentic tools," Cook told analysts, adding that customer recognition was happening faster than Apple had predicted.
OpenAI Fires Back at Apple Trade Secret Lawsuit With Its Own Email Evidence Apple's July 10 lawsuit accuses OpenAI's hardware chief Tang Tan and former engineer Chang Liu of stealing iPhone and Apple Watch trade secrets to build a competing AI device. OpenAI's August 3 rebuttal blog post disputes Apple's timeline, saying a lawyer's email mix-up, not stonewalling, explains a five-month gap Apple cited as evidence. - OpenAI Apple trade secret lawsuit response - how Apple botched corporate espionage claims
The shortage was visible in ordinary buying terms. MacRumors reported in April that some shipping waits had stretched into months, that Apple had stopped selling the Mac Studio with 512GB of unified memory, and that the base Mac mini was listed as currently unavailable in Apple's online store. Apple also said Mac revenue rose 29% year over year to $10.4 billion in the June quarter, making it the fastest-growing major product segment at the company.
Developers have helped pull the same rope. OpenClaw, the open-source agent project stewarded by Peter Steinberger, has become one public example of the demand for local agent work on Macs. Steinberger's GitHub profile says he is now at OpenAI working on agents and keeping OpenClaw open and independent, and the project's star count has run into the hundreds of thousands. You don't need every developer to buy a fleet for that to hurt supply. A few thousand people buying two or three machines each changes the shelf pretty quickly.
The Refresh Came Early #
Apple's answer arrived on August 25, earlier than its usual late-year Mac rhythm. The company announced a new Mac mini with M6 and M5 Pro options, with the M6 model starting at $899, and a new Mac Studio with M5 Max and M5 Ultra options. The M5 Ultra version starts at $5,499, while the broader Mac Studio line starts at $2,499. Pre-orders opened immediately, with most configurations scheduled to ship September 22.
The timing matters. Apple normally lets the iPhone own this part of the calendar, but the company put fresh desktop hardware into the market before September because the old line had become too hard to buy. The new M6 is Apple's first 2-nanometer Mac chip, according to Apple, and the M5 Ultra brings a quad-die architecture with up to 512GB of unified memory and 1.2TB/s of memory bandwidth. Those numbers are exactly why AI buyers care.
Here's the thing: Apple did not set out to become an AI infrastructure supplier in the way Nvidia did. It built compact desktops for developers, studios and business desks. Now those machines sit inside a wider fight over where AI work runs, in the cloud, on a rented GPU cluster, or locally on hardware a company can actually buy and control. Google has TPUs, Amazon has Trainium, Nvidia has DGX Spark, and Apple suddenly has labs buying Mac minis by the truckload.
That still leaves Apple with a hard operational problem. If enterprises can't get high-memory Macs when they need them, they will look at other hardware. The Information quoted David Stout, CEO of webAI, warning that enterprises could move to different choices if Apple can't supply the machines. That is a direct challenge to Cook's greatest strength: Apple is supposed to be the company that knows how to build enough hardware. For now, it cannot.
Also read: Anthropic Sued Over Claude Max Plans That Deliver Far Less Than Advertised • Twitch Streamer Sues Amazon Claiming It Trained AI on Streams Without Consent • Meta Is Testing Robots to Take Over Cabling and Server Jobs at Its Data Centers
Apple Sues OpenAI Over Claims It Stole Trade Secrets For Hardware Apple filed suit against OpenAI on Friday in federal court in Northern California, accusing the AI lab and two ex-Apple employees, including its own chief hardware officer, of stealing trade secrets to build consumer hardware. The complaint details interview show and tell sessions, a misappropriated metal finishing technique, and a former engineer... - apple sues openai trade secrets - openai hardware stolen apple confidential