I suspect there’s a few factors preventing that - cost of server capacity, the UX of higher latency (it would be for possibly all ML stuff?)