"If we slow down, China wins" sounds strategic. It describes a race that nobody can win.
The race story assumes the first superintelligence becomes a national asset. That only works if someone can direct it. A system smarter than its operators cannot be aimed like a weapon. It is a loss-of-control event that does not check passports.
Beijing has the same extinction problem Washington does. So does every other capital. Mutual vulnerability is why nuclear arms control worked between enemies. It is the argument for a verified limit, not for sprinting first.
Fear of cheating is real. The answer is inspection, chip tracking, and limits on large training clusters, the same family of tools used for fissile material. Defection risk is why verification exists. It is not a reason to drive off the cliff on purpose.
The US and USSR negotiated nuclear limits at peak hostility. The Outer Space Treaty barred weapons in orbit during the Cold War. Enemies can bind one dangerous activity without trusting each other on everything else.
"A race to build something nobody can control has no winner. The national-security move is a mutual, inspected ban on the systems that end control, not a contest to lose it first."