Everyone can buy the models; what creates the gap is the AI productivity system.

What happened: An article from qbitai reports that Hayma (海马云), a real-time cloud rendering company founded in 2014, is positioning its RunningHub platform as an AI productivity system built on pooled GPU compute, post-trained vertical models, and an agent harness for image, video, and 3D Why it matters: The article argues that bought models are not a moat, and that compute scheduling, toolchains, and data are what separate AI products. It cites MiniMax purchasing H3 RH Enhanced through RunningHub as evidence that post-trained third-party models can flow back to model vendors.

An article from qbitai reports that Hayma (海马云), a real-time cloud rendering company founded in 2014, is positioning its RunningHub platform as an AI productivity system built on pooled GPU compute, post-trained vertical models, and an agent harness for image, video, and 3D

The article argues that bought models are not a moat, and that compute scheduling, toolchains, and data are what separate AI products. It cites MiniMax purchasing H3 RH Enhanced through RunningHub as evidence that post-trained third-party models can flow back to model vendors.