OpenAI Buys Tens of Thousands of Mac Minis as Nvidia’s RTX Spark Sells Out Pre-Launch

Elena Sokolova31/08/20264 min read

Somewhere in an OpenAI data centre, a stack of Mac minis is quietly grinding through reinforcement learning runs instead of sitting in someone’s home office. According to a new report from The Information, OpenAI has bought tens of thousands of Apple Mac minis and Mac Studios over the past few months, and it isn’t the only AI lab treating Apple’s desktops like server racks.

OpenAI has reportedly acquired tens of thousands of Apple Mac minis and Mac Studios to bolster its reinforcement learning and computer-use agent training efforts, according to The Information. Anthropic, meanwhile, is renting the same class of Apple devices through Amazon AWS. Two of the biggest names in AI, buying and renting Apple desktops instead of racking up more Nvidia GPUs, is not a small detail.

Gamers and PC builders reacted fast once the numbers started circulating, arguing over whether this is really squeezing consumer stock or just AI-hype noise:

Why Mac Minis, Not More GPUs

Apple’s Mac mini and Mac Studio lineup has reportedly been flying off shelves this year, with their unified memory architecture and Thunderbolt 5’s low-latency, TCP/IP-bypassing PC-to-PC links making them a strong fit for on-device agentic orchestration. That’s a mouthful, but the short version is simple. A Mac mini can hold a large model in memory and talk to a cluster of other Macs fast, without the cost of a full Nvidia server rack.

For a company like OpenAI running thousands of parallel reinforcement learning and computer-use agent tests, that’s cheaper than it sounds. Our view: this isn’t Apple suddenly becoming an AI hardware giant, it’s AI labs finding a shortcut and Apple happening to own it.

ASUS Says Its First RTX Spark Batch Never Reached a Shelf

On the Windows side, Nvidia’s RTX Spark chip is having its own supply squeeze before it’s even shipped. ASUS’s ProArt P16 and P14, among the very first laptops built on the platform, are already gone. The ProArt P16 and P14, among the first RTX Spark notebooks, are fully sold out on pre-order, positioning ASUS to capture the first wave of replacement demand at premium prices.

ASUS co-CEO Hu Shu-pin addressed the RTX Spark line directly on the company’s August 12 earnings call. He stated that channel pre-orders have exceeded expectations, with the originally planned quantity fully booked and additional orders being placed continuously. That’s a distributor pre-order, not proof every unit is walking out of a retail store, but it’s still a notable start for a chip nobody outside Computex crowds had heard of three months ago.

Wccftech reports, citing Taiwan’s Economic Daily, that MSI has burned through its own initial RTX Spark allocation too, alongside Dell, HP, Lenovo and Microsoft, and is now asking Nvidia for more supply ahead of the fall launch. The same report puts pricing on the flagship N1x system at roughly NT$110,000, or about $3,400.

What’s Actually Inside the N1x Chip

RTX Spark isn’t a laptop GPU bolted onto a normal chip. It pairs a 20-core Grace CPU with a Blackwell GPU carrying 6,144 CUDA cores, claiming up to 1 petaflop of FP4 AI performance and roughly 600GB/s of NVLink-C2C bandwidth between CPU and GPU. It also runs the full Nvidia software stack: CUDA, TensorRT, DLSS, Reflex and G-SYNC included.

ASUS is pitching the ProArt machines as creator and AI-agent tools rather than gaming rigs, and there’s no ROG-branded RTX Spark laptop yet. But the DLSS and Reflex support baked into the platform means gamers will care eventually, especially once the tech trickles down to cheaper SKUs. A GR1X desktop on the RTX Spark platform is slated for Q4.

Memory Prices Are Squeezing Everyone Else

None of this is happening in a vacuum. DRAM prices have climbed through 2026, driven largely by AI data centre demand pulling supply away from consumer devices, and it’s hitting laptops, workstations and servers all at once. ASUS itself flagged the same pressure on its earnings call, warning that tight memory supply could limit PC shipments even where server allocation holds up.

We’ve been saying it for months on this desk: if you’re building or upgrading a gaming PC right now, the memory line item is where your budget actually gets wrecked, not the GPU itself. Watch RAM and VRAM pricing over the next few months more closely than any benchmark chart.

RTX Spark’s first desktops and wider laptop rollout are still pegged for the fourth quarter of 2026. Whether Nvidia can get ASUS, MSI and the rest a second batch before then is the number worth watching next.