> instead of keeping their optimizations proprietary

There were no magic secret optimizations to release. It just straight up did not work. They had to add back dynamic branch prediction, and even then the load store latency was such trash that they had to put ginormous L3 caches on it to get even close to reasonable performance.