Summary
Matthew Berman makes a skeptical case about local AI today. Even as a technical creator with abundant compute, he still turns to ChatGPT and Claude when he needs to get real work done, and he questions whether the average person should cancel a subscription to run models at home.
The video first separates local AI, which runs on your own device even without internet, from open-weights AI, which labs release for anyone to use and which cloud providers and neoclouds often serve at scale. The strongest open-weights models, he notes, tend to be one generation behind the frontier and are too large for consumer hardware.
Berman then walks through the math. Hardware such as RTX 4090 GPUs and Mac Studios requires large upfront spending, often thousands of dollars, and only pays off if the machine is kept busy for years. He also argues that setup remains difficult even with tools like LM Studio, in which he has invested, because of differences in chips, operating systems, and models.
He still sees real benefits in data sovereignty, control, and independence from frontier labs, and sketches how local AI could become viable in the future. It is a candid reality check for anyone weighing a local AI setup.
📺 Source: Matthew Berman · Published October 23, 2026
🏷️ Format: Opinion Editorial
