That's different, though. If 20 other companies can host these models, you still have to trust them. The end result should be cheap hardware that's good enough to large a solid, mature LLM that can code comparably to a fast junior dev.