GPU compute markets: vertical integration or derivatives?

A user who says they recently moved from Lambda to SF Compute argues that there isn't enough public data to reliably price GPU indices and compute futures. They say many neoclouds have favored large private deals and clusters that boost valuations over adding liquidity to the market and their public-cloud businesses. They ask whether vertical integration, as with a neocloud, or a conventional derivatives market offers a better approach to the physical compute market.