Kimi K2.5 on 2 Mac Studios (<$20K) running at 24 TPS (twitter.com)

Very impressive, would love to see the results from the verifier for this fully local setup: https://github.com/MoonshotAI/K2-Vendor-Verifier

Not to be glib, but how can people witness this and still shovel countless billions into GPU data center buildouts? Are there just two vastly different approaches supported here by the market?