You’re staring at your M2 MacBook Pro, then at the Nvidia RTX 4090 sitting in its eGPU enclosure on your desk. For years, these two pieces of hardware have existed in parallel universes—one running Apple Silicon, the other packed with CUDA cores that your AI workflows desperately need. In 2026, Apple finally approved a driver that bridges this gap. Plot twist: it didn’t come from Nvidia.
Tiny Corp, the company behind the project, got Apple’s blessing for a driver that enables both AMD and Nvidia eGPUs to work with Arm-based Macs. According to their announcement, installation is now simple enough that “a Qwen could do it, then it can run that Qwen.” For those keeping score at home, that’s a reference to running large language models—exactly the kind of workload that makes AI toolkit reviewers like me care about external GPU support.
Why This Matters for AI Work
Let’s be direct about what we’re dealing with here. Apple’s Metal framework is solid for many tasks, but the AI and machine learning space has been built on CUDA for over a decade. Training models, running inference at scale, testing frameworks—most of the tooling assumes you have access to Nvidia hardware. Mac users have been working around this limitation with cloud instances, which gets expensive fast, or by maintaining separate Linux boxes.
The eGPU solution isn’t perfect. You’re still constrained by Thunderbolt bandwidth, which means you won’t get the full performance of that GPU sitting in the enclosure. Think of it as having a sports car but only being allowed to drive it in a school zone. The hardware is capable of more, but the connection is the bottleneck.
The Tiny Corp Factor
What makes this story interesting is who made it happen. Tiny Corp stepped in where the big players didn’t. Nvidia hasn’t exactly been rushing to support Apple Silicon—there’s years of bad blood there, dating back to when Apple stopped using Nvidia GPUs in their machines. Apple, for their part, has been pushing developers toward Metal and their own GPU solutions.
This driver represents a third path: community-driven solutions that work around corporate standoffs. Tiny Corp built something that both companies probably wish didn’t need to exist, got it approved through Apple’s notoriously strict driver certification process, and released it to the public.
What This Means for Your Toolkit
If you’re running AI workloads on a Mac, this opens up some practical options. You can now test CUDA-dependent frameworks without spinning up cloud instances. You can run local inference on models that perform better on Nvidia hardware. You can prototype on your laptop and know that the code will translate to your production environment.
But let’s not oversell this. The Thunderbolt limitation is real. For serious training runs or production deployments, you’re still better off with a dedicated machine or cloud resources. This is a development tool, not a replacement for proper infrastructure.
The installation process being straightforward is actually the bigger win here. Previous attempts at Mac eGPU support involved kernel extensions, unsigned drivers, and the kind of terminal commands that make you double-check you have backups. If Tiny Corp delivered on making this accessible, that’s valuable.
The Bigger Picture
This approval signals something about Apple’s current stance on external hardware support. They could have rejected this driver. They could have slow-walked the approval process into oblivion. Instead, they certified it.
Maybe Apple recognizes that developers need flexibility. Maybe they’re less concerned about eGPUs now that their own silicon has matured. Maybe someone in Cupertino just decided it wasn’t worth the fight. Whatever the reason, Mac users who need Nvidia GPU access now have an officially sanctioned path forward.
For AI toolkit reviewers and developers, this is a practical expansion of what’s possible on Apple hardware. It’s not a perfect solution, but it’s a real one. And sometimes, that’s exactly what you need.
🕒 Published: