Martin Ships AI

I build AI systems and run them in production. This is the gear behind the videos.

Everything listed here is hardware I paid for and use. No review units, no sponsorships. The specifications below are read off the running machine, not copied from the invoice.

The render box

Mini-ITX, built March 2026, $959 in parts

Every video on the channel is rendered on one small machine that runs as a GPU node in my home Kubernetes cluster. Avatar generation and video compositing both happen locally on it. It replaced a cloud avatar API that cost $150/month, so it paid for itself in about six months and the marginal cost of a video is now electricity.

What I would change

The 8GB of VRAM is the constraint that actually binds. It fits the avatar model fine, but leaves no room to run a local language model alongside it, so script generation still goes out to an API. If you are building this now and the budget stretches, VRAM is the upgrade that matters. Nothing else on this list is holding the machine back.

The 450W power supply looks undersized for a discrete GPU and is not. The whole system draws roughly 233W under render load, which leaves real headroom. Sizing a PSU off the GPU vendor's recommendation assumes a high-end CPU you are not buying here.