It makes sense for AMD to be trying to get out a GPU with a smaller die.
For RDNA2 they pulled a 180 over Vega to prioritize a big cache with slower memory over expensive HBM. Unfortunately for AMD, they did not anticipate that TSMC would become a major bottleneck for their production. Cache consumes a lot of die space. Their 6700XT die still has a large cache, so the chip isn't nearly as small as they need it to be. So, their current lineup is not well suited to today's manufacturing limitations. But lower powered GPUs need far less memory bandwidth, meaning an exponentially smaller cache. I would imagine that 6700XT production is pretty limited right now, with priority going to the higher margin 6900XT/6800, and to the smaller chips that they will launch soon. There may not be enough faulty chips available to support launching a 6700 for awhile. But, AMD is definitely working to solve the problem. The chiplet-on-chiplet cache that they recently demonstrated is going to be very important for their future GPUs, assuming they are similar to RDNA2.
To Nvidia's credit, Samsung's 8nm is not nearly as efficient as TSMC's 7nm, but Samsung had capacity to spare. It looked bad at launch... But in retrospect, I think that decision is working out very well for Nvidia, right now. There's still production bottlenecks, but that's one less for them to deal with.