IDK, even in operations the size of the cloud providers, executing some of the same things billions of times a second, there's still very few specialized chips. TPU is an example, and I think they all use some for network packet filtering and such. But you don't see anything for DB acceleration or paxos acceleration or maps acceleration or anything like that. They've had years to figure this out, so if they're not finding it worth it at hypercompetitive cloud scale then I don't see the revolution going too awful far anytime soon.