Unarguably, deploying within the same physical location should yield the lowest latency.

In the majority of use cases however, the added few milliseconds in return for a multiple increase in resilience are worth it.

Try to span your deployment across two availability zones at least.