The gap between what US frontier models cost and what Chinese alternatives charge has grown wide enough that developers are making practical choices that cut against the policy grain. Teams that can't get budget approved for GPT-5 or Claude tiers are routing workloads to cheaper options — not out of ideology, but out of spreadsheet math. It's a slow defection, not a revolt. Most shops run hybrid stacks anyway. But the pattern is real enough to watch: cost pressure is doing what geopolitical friction couldn't, nudging adoption curves in directions regulators probably didn't intend.