The supply limit is a constant in the software, and software can be edited by anyone. What makes the limit meaningful is the structure of who would have to accept an edit.
Nodes enforce rules independently
Every full node validates every block against its own copy of the rules, and rejects anything that breaks them regardless of how much work backs it.
A miner producing a block with an excessive subsidy would find that block ignored, and the reward would exist only on a chain nobody follows.
Enforcement is therefore distributed across everyone running a node, not delegated to the participants who produce blocks.
Changing the cap requires near-universal agreement
A rule change that permits previously invalid blocks is only adopted if nodes upgrade to accept them. Nodes that do not upgrade continue rejecting.
The result would be two chains: one with the original limit and one without, each with its own users and its own market price.
Anyone proposing the change must therefore convince essentially everyone, since a substantial holdout leaves the original chain intact and competing.
The people who would decide are the ones harmed
Additional issuance dilutes existing holders in favour of future block producers, so those being asked to approve it are the ones who lose.
Holders run nodes, exchanges run nodes, and custodians run nodes, which puts the enforcement in the hands of the parties with the strongest reason to refuse.
The incentive structure and the enforcement structure point the same way, which is the substance behind the claim that the cap is credible.
Credibility comes from precedent as well as design
Contentious rule changes have been attempted and have failed to displace the original chain, which established that the mechanism works under pressure.
Each episode where a change was resisted makes the next attempt less plausible, because the outcome is now something participants have seen.
Conservatism about protocol changes generally reinforces this, since a network that rarely alters anything is a harder place to alter something fundamental.
The open question is elsewhere
Nothing about the cap guarantees that fee revenue will be sufficient to fund security once the subsidy has effectively ended.
That is a genuine debate, and it concerns whether the arrangement remains viable rather than whether the limit will be respected.
Conflating the two produces confusion: the supply constraint is among the strongest commitments in the system, and the funding model beneath it is the part still being worked out.