In the early days of the rise of cloud computing (or the reincarnation of mainframe computing, as some would have it), it was common to put different customers' workloads on the same physical machine, relying on the hypervisor to maintain a firewall between them.
But that was before Specter/Meltdown, which allow code to see through the firewall and eavesdrop on other code on the same machine. There are mitigations for this class of vulnerability, but they have some performance cost, and there doesn't seem to be a consensus on whether they are really one hundred percent leakproof.
Does AWS still allocate different customers on the same physical machine at the same time? (Not talking about e.g. spot instances where someone else was using the machine five minutes ago, and someone else again may be using it five minutes from now.) If so, does this happen across the full range of instances, or only at the low end of the price range?