Tinybox ships for local AI workloads

The HN title makes it sound like a pocket offline gadget, but tinybox is really a brute-force AI server for people who want control over their stack.

• –This is serious infrastructure, not a novelty box: rack-mountable chassis, high-wattage PSUs, and GPU-heavy configs make that clear
• –The local-first pitch is compelling for teams that care about data control, latency, and avoiding cloud usage fees
• –Shipping with both tinygrad and PyTorch broadens it beyond one framework’s fanbase
• –The tradeoff is obvious: power, cooling, and setup complexity, so this only makes sense when you really need owned compute
• –The strongest buyer story is in-house model work, offline inference, and experimentation where cloud lock-in hurts