Apple reportedly clears Nvidia eGPU driver for Apple silicon, opening Macs to external AI GPUs
What happened
It has been reported that Apple has approved an external-GPU (eGPU) driver for Nvidia on Apple silicon, according to Tom's Hardware and a social post by the developer Tiny Corp. Reportedly the driver lets users attach high-end GPUs to Macs and use them for large-language-model (LLM) workloads without the old hacks — Tiny Corp says the install flow is now simple enough that “even Qwen” can be run locally. The company also stresses the driver is focused on AI model compute, not gaming.
The hardware and the seller
Tiny Corp — developer of the tinybox AI appliance — has been at the center of this move. tinybox configurations on sale include a “red v2” with four AMD 9070XT GPUs priced at $12,000 (roughly ¥82,700) and a “green v2 Blackwell” with four Nvidia RTX Pro 6000 Blackwell GPUs at $65,000 (roughly ¥448,000). Tiny Corp reportedly plans an “exabox” in 2027 with some 720 RDNA5 GPUs and about 1 exaflop of FP compute, a project it values at around $10 million. The company previously clashed with AMD over driver issues — reportedly prompting personal involvement from AMD CEO Lisa Su.
Why it matters
For developers this could be a big change. Until now, running large models on macOS often required disabling system protections or buying dedicated accelerators like tinybox. If ordinary Macs can meaningfully run training or inference jobs with attached GPUs, the barrier to local AI experimentation falls. That matters globally — and for China in particular, where access to the newest data-center GPUs has been affected by U.S. export controls and geopolitical friction. Could Macs become a practical fallback for AI teams cut off from cloud GPUs?
Caveats and context
The driver in question was developed by Tiny Corp, not Nvidia or Apple, and the rollout and real-world performance remain to be independently verified. It has been reported that Apple’s approval removes earlier technical workarounds, but we should expect testing and clarifications about driver support, stability, and software licensing. For Chinese AI practitioners watching supply chains and export rules, this development is one more variable in an increasingly contested compute market.
