Also they should have had a policy of "not one transistor for the national labs" and "not one transistor for hyperscalers", particularly because anything they do to appease hyperscalers just saves them money that they'll spend on their custom silicon transition.

There's something prestigious about HPC but the story of how commercial data processing went parallel in the oughts shows how out of touch the HPC community is.

In the meantime Intel has botched the deployment of SIMD, slowplaying to the point where maybe 7 or 8 years from now some mainstream software might support AVX512, maybe. Somebody should be tarred and feathered for introducing the idea that features should be fused off for all but the highest paying customers. If the customer is not paying for the fused off functionality, the shareholders are. It sounds really ruthless and avaricious and might impress some business analysts as a form of vice signalling but it's pure waste.