GPUs in the cloud aren't targeted at gamers. They're targeted at people doing things like running render farms and training deep learning models.
Fortunately the amount of additional latency introduced is likely to be negligible (another comment cites PCI-E switches incurring <= 1µs).