As regards PC performance, it certainly is slowing down from where it had been. CPUs first; my i7 940 is now regarded as "4th tier" hardware by Tom's, but the physics in most racing games run fine within the 'tick window' on something which is 2.93GHz (as opposed to the latest, which will turbo boost to 4.7GHz). GPU has up until 2 years ago been doing the Moore's Law thing of doubling in performance each generation; but nVidia stopped that with the 10 series, and as
@stonec says, they are now milking the cash cow (was crypto, now they have a big backed up demand from gamers and THIS is why their CEO said this week that next gen won't be coming "soon", because they have capacity they need to maintain use of and sell now so he wants everyone to think and upgrade is not imminent, because that depresses demand in the short term). AMD are just capacity constrained, and they've had power/heat issues for quite some time. The upcoming trends on both CPU (seen in Threadripper, 4x modules) and GPU (not yet seen) is for multiple 'modules' within the unit bridged together, since they can't keep making bigger and bigger chips with the yields they get. Whether these will eventually have internal bridging bottlenecks isn't clear. But this is a fairly major structural change.
Most of all though, the problems I think we have is with software structure. The physics calcs in all sims at the moment seem to all require high single-thread performance, since they are single threaded. This is surely solving multiple equations simultaneously, but the algorithms seem to be recursive and not parallel. The single-thread performance needs more GHz, and we aren't going to be seeing massive leaps with this. This means we can't take advantage of multiple CPU cores. The place where you can is with the graphics: issuing draw calls to the GPU. But, this needs DX12. I'll be frank and say, I don't know why S397 targeted DX11 and not DX12 when they had to do their rewrite; they are a PC, Windows-only solution; and 70% of Steam users already having DX12-class GPUs. We've already seen the limits of a DX11 engine - not sure anyone is going to get more out of that than SMS have with PCars2. Windows still has internal structural problems: for example, it 'supports' HDR but you have to enable a specific mode and if you do, the desktop looks like complete crap. MS are sleeping on the job. For things like VR, there's a lot of CPU-to-GPU interaction, no direct memory accesses across both, its across a relatively slow bus. The APUs available use 'slow' DDR3/4 shared memory, not high spec memory. I still maintain we need "one GPU per eye" or one per viewport, but I see no progress in making this viable (there is a spec, its unused). I don't see S397 as being a technology innovator in any of these areas, they'll use "tried and true" approaches because they can't afford to take the risk or requiring a unique software/hardware combo; they are "mainstream". Hopefully we'll see more improvements in the rain implementation and general performance, but it won't be orders-of-magnitude.
I went from upgrading a PC every 3 years to upgrading every 10 (CPU/motherboard; 3+ years now for GPUs). That's how much progress has slowed down, or at least, that which impacts a sim racing gamer such as myself.
TL;DR summary: yes, we are limited by hardware and especially so in 'mid range', but the manufacturers aren't exactly helping matters with their pricing/profit taking (Intel/nVidia and MS, I'm looking at you).