I said before that if 18A didn't pan out, Intel is sunk. I had no predictions whether it would or not, as I had no special information everyone else didn't have.
Well, Intel had to back off of doing foundry services with 18A and won't tell us why, but there can only be one real answer: Whatever problem they are having with it takes so much specialized work to achieve that nobody else's designs can practically be fabbed in it.
Intel has had a long line of expensive, high-profile failures. Perhaps most notably given current trends in computing, they failed at ARM. They bought the fastest ARM implementation around from DEC, but they couldn't get the design to scale down enough to match the power consumption of other implementations, so they sold the result of their messing with it (XScale) off again — to Marvell, who they had licensed some bits for XScale from. And let us not forget iTanic, which never achieved performance parity with amd64 because Intel's magic compiler never materialized.
At this point Intel probably cannot be saved with a new process, as their customers are rapidly becoming AMD customers.