Why do ZK rollup provers stay as centralized as a single cloud region?

Every major ZK rollup in production today runs its proving layer through a small permissioned set of nodes operated by the founding team. Generating ZK proofs fast enough to keep up with block production requires specialized hardware and deeply optimized software that only a few entities can operate at production speed. A single outage of the proving service halts L2-to-L1 settlement and can freeze withdrawals for hours without any user-facing warning. Efforts to build decentralized prover networks are in early rollout but no major ZK rollup has yet handed over permissionless proof generation in production. The rollup's decentralization claims do not hold as long as one team controls the proving layer.