Nvidia’s head of automotive Xinzhou Wu openly acknowledged on The Verge’s Decoder podcast that his own division competes with Nvidia’s other units for access to scarce H100 GPUs. The admission underscores the company’s supply constraints despite its $2.2 trillion valuation and 80% market share in AI chips. Nvidia’s automotive arm which powers self driving stacks for Tesla BYD and others is now a victim of its own success as data center demand for AI training soars. H100s sell for $30k to $40k each but resale prices hit $70k due to shortages.

This is the clearest sign yet that Nvidia’s growth is constrained by its own capacity not demand. The fact that even internal teams must fight for chips exposes a structural flaw in the AI gold rush. If Nvidia can’t allocate enough compute to its own strategic divisions like automotive it risks ceding ground to AMD or custom silicon from hyperscalers. The bottleneck isn’t innovation it’s fabrication with TSMC’s 4N process maxed out and CoWoS packaging in short supply.

Nvidia wins as long as demand outstrips supply but its customers lose. Tesla which uses Nvidia DRIVE for FSD loses if automotive gets deprioritized. Cloud providers like AWS and Google which pay top dollar for H100s win at the expense of smaller players. TSMC wins as the sole supplier of advanced packaging. AMD wins if Nvidia’s constraints push buyers toward its MI300X chips.

Watch for Nvidia’s next earnings call to see if automotive revenue growth slows. If H100 allocations tilt further toward data center expect Tesla or others to accelerates custom silicon efforts. Also watch for AMD’s MI300X traction as a potential relief valve.