My Whole AI Stack Went Chinese

A founder went viral bragging that swapping his entire AI stack to Chinese open models cut costs 87% with revenue untouched. The reasoning brain moved off Opus onto a Kimi-class model, and each task got routed to whatever was cheapest. On the spreadsheet, it’s a clean win. But something quietly gets blurred. Sovereignty means the models, the data, and the infrastructure sit under your control, and on-prem means running all of it inside your own walls rather than someone else’s cloud. Trading a US API for a Chinese one lowered the invoice. The keys still belong to somebody else.

Source: RT @DeRonin_: My entire AI stack is now Chinese 🇨🇳 · twitter

What this means for ThakiCloud

We drew this because “cheap” and “mine” get confused far too often. ThakiCloud handles both on one board. Paxis runs the per-task model routing on your own on-prem hardware, and Metis trains and serves those models directly. It mirrors the savings from that viral tweet, except the work happens inside your walls instead of on a rented API. The bill drops and the keys stay in your hand. That’s an actual stack you own, not a fresh sublet.

---

An auto-generated comic riffing on this week’s industry news.