Conversely, has Oxide discussed in depth how they see GPU compute?

I’m guessing for their current target market of existing enterprise software it doesn’t matter, but the ownership and economics stories are at least as compelling for matmul code.