I don't have exact numbers but from my limited experience I would say about 30% of the price is the cost of setup and production (human labor, materials, capital depreciation) and the rest of the price minus profit is administrative, sales, and regulatory cost. Silicon design is getting cheaper, especially with FPGAs, and the designs are relatively flexible while the sheer amount of time required to set up a manufacturing line to produce even a single processor is very expensive. Like the RCA 1802 linked above, a single silicon design can be used for decades in mission critical applications as the revenue piles up and the costs to tweak the design fall. When you're the only one buying a setup for a rad hardened manufacturing process and producing only a dozen chips for your client, slipping in several experimental revisions on the mask can sometimes be almost free.

Most compliance issues are barriers to entry but are largely overblown. In fields like aerospace and nuclear weapons the most critical requirement is abundant reliability and lifetime data which is why, regulation or no, you will never put something like an Intel 60nm or lower processor into a critical $100 million satellite.