Even the company printing money from AI chips can't get enough of them for itself.
The Summary
- Nvidia's automotive division has to compete internally for GPU access, according to automotive chief Xinzhou Wu — Jensen Huang personally settles the allocation fights weekly
- Internal priority system balances near-term revenue against Huang's "zero trillion dollar businesses" — new markets that don't exist yet but could be worth trillions
- Autonomous driving remains strategic priority despite data center demand dominating supply
The Signal
Nvidia sells every chip it can make to hyperscalers building AI infrastructure. Microsoft, Meta, xAI, Amazon — they're all standing in line with checkbooks open. But inside Nvidia, teams building the company's own AI products are fighting the same scarcity battle as everyone else.
Xinzhou Wu runs automotive, one of Nvidia's long-term strategic bets on autonomous vehicles. His team needs GPUs to train and test self-driving models. So does every other division working on AI products. The allocation process happens weekly, with teams making their case for compute resources. When they can't agree, it goes to Huang.
"Sometimes we need Jensen to help."
This isn't just an interesting operational detail. It reveals Nvidia's actual resource allocation philosophy and what the company really believes about the future.
Wu says decisions aren't purely revenue-driven. Nvidia could allocate every spare GPU to projects that generate near-term cash. Instead, they're protecting capacity for what Huang calls "zero trillion dollar businesses" — markets that don't exist yet. Autonomous driving is one. These are multi-year, capital-intensive bets that won't pay off in the next quarter or the next year.
The weekly fights over compute show how seriously Nvidia takes these frontier bets:
- Teams pitch competing uses: training runs vs. testing infrastructure vs. product development
- Revenue potential gets weighed against strategic positioning in emerging markets
- Huang personally arbitrates when the trade-offs get too complex for standard prioritization
This is the real tell on where Nvidia thinks the agent economy is heading. If autonomous vehicles weren't a core strategic priority, Wu's team would just get whatever scraps are left after data center customers and high-margin product groups take their fill. Instead, they're getting enough allocation that it requires CEO-level intervention to balance.
The Implication
Watch what companies protect when resources get tight. Nvidia could stuff every GPU into data centers and let automotive starve. They're not doing that. The weekly allocation fights — and Huang's personal involvement — signal that Nvidia sees autonomous vehicles as foundational infrastructure for Web4, not a side bet.
If you're building in the agent space, this matters. Nvidia is rationing its own medicine to ensure strategic bets survive the scarcity. That's not charity. That's conviction about which markets will matter when agents start moving atoms, not just tokens.