Nvidia sells the chips that run autonomous cars. So a Nvidia patent about running out of chip is worth reading twice. The grant US12649478B2 — "Method to estimate processing rate requirement for safe AV driving to prioritize resource usage" (issued June 9, 2026) — is about rationing onboard compute, which only makes sense if onboard compute is scarce.

The mechanism: the system estimates, moment to moment, how much processing the safe-driving task actually needs — denser traffic and higher speed demand more perception and planning throughput — and prioritizes resources accordingly. The CPC tags are pure functional safety: B60W 50/06, B60W 50/0205, B60W 60/0015. This is a patent about guaranteeing the safety-critical work gets the cycles it needs, even if that means starving something else.

Here's the quiet admission. The industry narrative is that compute is abundant — just add another Nvidia module. The patent says otherwise: power, thermal limits, and cost cap how much silicon a car can carry, so the real engineering problem is allocating a fixed budget to the tasks that matter most right now. You don't patent how to ration a resource that's free.

The risk factor is where the honesty lives, and this patent is the engineering equivalent. A robotaxi's bill of materials is dominated by sensors and compute; every extra chip is cost, weight, heat, and power draw that hurts unit economics and range. So even the supplier that profits from selling more chips is filing IP on using fewer of them more cleverly — because its customers' margins depend on it. The boring incumbent threatened here is "just add more compute."

What the grant does not disclose is the actual budget — how many TOPS safe urban driving needs, or how close current hardware runs to the edge. A method for estimating and prioritizing is not a measurement of headroom. The number that would settle the cost debate isn't in the claim.

For readers tracking robotaxi economics, this is the tell most coverage misses. When the chip vendor itself patents compute rationing for safety, it's confirming that autonomy is constrained not just by algorithms but by the silicon budget — and that the budget is tight enough to be worth a patent.