I built a server with 768GB VRAM for frontier, but all new frontier open source models are likely to be two trillion or above now, including next GLM 6, am I cooked?

A user built an Epyc server with 768GB VRAM (twelve 64GB cards) and 256GB RAM for frontier open-source models. However, they are concerned that new models like GLM 6, Qwen-max, Kimmi, and Deepseek V4 Pro are becoming too large, potentially exceeding two trillion parameters, making their server insufficient. They question whether to abandon this frontier dream and switch to smaller flash models, noting the server is not for business use.