x/nxpc · 
Live US/CA pricing checked September 1, 2026, 16:00 UTC. All stock counts captured at publication time.
The Information broke the story this morning, and the hardware world should pay attention: OpenAI has purchased tens of thousands of Mac minis and Mac Studios — not for developers, but as reinforcement-learning infrastructure for training "computer-use" AI agents. Anthropic is doing the same thing, renting Mac minis through AWS. And yes, the buying spree is deep enough that it helped buy out entire product lines.
This is the question everyone asked when the report dropped, and the answer is a genuinely interesting piece of silicon strategy:
Unified memory. NVIDIA GPUs keep VRAM and system memory separate — data crossing that boundary is a bottleneck. Apple's M-series chips use one shared memory pool that both CPU and GPU access directly. For reinforcement-learning workloads that constantly shuffle model states between CPU and GPU, that architecture is quietly perfect.
Sustained thermals. This is why the labs want Mac minis and Mac Studios specifically — and zero MacBooks. Screenless, keyboardless desktop boxes with dedicated cooling can run RL environments for hours or days without thermal throttling. A laptop folds under that load.
Clustering. Open-source software like EXO Labs links fleets of Macs into clusters capable of running trillion-parameter models locally, and Apple's newest Mac Studio launch leaned hard into exactly that clustering capability.
The economics make sense too: RL for computer-use agents needs thousands of parallel environments — lots of mid-tier machines, not a few giant GPU pods. A rack of Mac minis is effectively a cheap, quiet, low-power RL environment farm.
The buying spree shows up in Apple's financials. Last quarter Mac revenue hit $10.3 billion, up 29% year-over-year — the fastest-growing product line Apple has, beating iPhone and iPad. On June 23, Apple even held a rare enterprise event at Apple Park ("Business at the Park") where Disney and Ford execs mingled with Anthropic co-founder Jared Kaplan — and per attendees, the Mac mini was the star of the show.
NVIDIA has noticed. Insiders say the GPU giant now views Apple as its biggest competitor in local AI — which is why DGX Spark exists at all. And Apple, famously, has been caught flat-footed: ex-Apple AI enterprise marketing manager Todd Dailey told The Information the enterprise Mac boom was an accident — Apple has no dedicated enterprise engineering team for it, and hasn't sold a real server since the Xserve died in 2011.
Here's where it gets painful for regular consumers. The memory supercycle (AI data centers eating the world's DRAM and NAND supply) plus this new institutional Mac demand has left Apple's desktop line in a state of perpetual drought:
And Amazon right now (live data, Sept 1):
| Product (Amazon US) | Price | Stock status |
|---|---|---|
| Mac mini M4, 16GB/512GB | $979.99 | 400+ bought/mo |
| Mac mini M4, 16GB/256GB | $997.49 | 300+ bought/mo — the smaller SSD costs MORE |
| Mac mini M4 Pro, 24GB/512GB | $1,569.00 | Only 3 left |
| Mac mini M4 Pro, 48GB/512GB | $2,199.00 | Only 20 left |
| Mac Studio M4 Max, 128GB/2TB | $5,699 (was $7,999) | Only 1 left |
| Mac Studio M4 Max, 128GB/4TB | — | Currently unavailable |
| Mac Studio M3 Ultra, 256GB/4TB | $12,299 (was $14,999) | Only 3 left |
| Mac mini M6 (16GB/256GB) | $879.99 | Releases Sept 22, 2026 |
| Mac Studio M5 Max | $2,499 | Releases Sept 22, 2026 |
That "$997.49 for 256GB vs $979.99 for 512GB" line is scarcity pricing in the wild — third-party sellers charging more for less storage because the 256GB config is what everyone's buying out.
Over on Amazon Canada it's even tighter:
| Product (Amazon CA) | Price (CAD) | Stock status |
|---|---|---|
| Mac mini M4, 24GB/512GB | $1,699.00 | Only 1 left |
| Mac mini M4 Pro, 24GB/512GB | $2,299.99 | Only 1 left |
| Mac Studio M5 Max (2026) | $3,499.00 | Releases Sept 22 |
| Mac Studio M5 Ultra (2026) | $7,799.00 | Releases Sept 22 |
| MacBook Air 15" M5 | $2,099.00 | Ships Sep 21–22 |
Both current-gen minis in Canada: one unit each. That's not a listing, that's a museum exhibit.
In June, Apple pushed through its steepest across-the-board Mac price hike in years and told The Wall Street Journal: "We have never seen a component price increase this much, this quickly."
| Model | Old Price | New Price | Increase |
|---|---|---|---|
| Mac mini M4 256GB | $599 | $799 | +$200 (+33%) |
| Mac mini M4 512GB | $799 | $999 | +$200 (+25%) |
| Mac mini M4 Pro | $1,399 | $1,599 | +$200 (+14%) |
| Mac Studio M4 Max | $1,999 | $2,499 | +$500 (+25%) |
| Mac Studio M3 Ultra | $3,999 | $5,299 | +$1,300 (+33%) |
The base Mac mini — famously the best value in computing — jumped 33% in one stroke. That's the memory supercycle made visible on a price tag.
Here's my favorite data point from today's sweep: while Apple's inventory evaporates, NVIDIA's DGX Spark is sitting in stock on Amazon US with next-day Prime delivery — $4,999.99, 500+ bought in the past month, and even a 2-pack bundle ($10,169.99) for people chaining them. ASUS's GB10-based Ascent GX10 ($6,699) is also flowing. Ex-Apple employees are literally telling enterprise customers that DGX Spark is the alternative being mentioned by name when Macs can't be sourced.
Remember: NVIDIA built Spark specifically to counter the Mac mini's local-AI appeal. The shortage is doing NVIDIA's marketing for it.
The Mac is just the visible symptom. Zoom out to the memory supercycle that's making all of this possible:
Phone and notebook brands are already raising prices and quietly downgrading specs (12GB "flagship" RAM, anyone?). IDC expects 2026 DRAM and NAND supply growth of just 16%/17% YoY — far below historical norms.
AI data centers eat memory supply → Apple (and every OEM) pays more for DRAM/NAND → Mac prices jump 25–33% → AI labs hoover up the remaining desktop inventory → retail stock hits "1 left" → scalper premiums and renewed-market growth → your next mini PC, laptop, or RAM upgrade costs more, with longer waits.
The wild part of this story isn't that OpenAI bought a pile of Macs. It's that the world's biggest AI lab looked at NVIDIA's entire data-center empire and decided that for one crucial workload, a $799 aluminum box was the better tool — and in doing so, made that box something ordinary people now have to wait months to buy.
Live pricing and stock counts: Amazon US + Amazon CA via real-time product search, September 1, 2026 ~16:00 UTC. Stock moves fast — treat counts as a snapshot.