NVIDIA is betting its next decade on a move from selling shovels to splitting gold, and the guy who called the 2008 crash thinks they're wrong.
The Summary
- NVIDIA launched a revenue-sharing AI compute model with cloud providers, shifting from hardware sales to partnership-based recurring income
- Michael Burry is shorting NVIDIA stock while this infrastructure play rolls out, creating a direct bet against the company's strategic pivot
- The model could deliver subscription-like stability but ties NVIDIA's returns to partner execution, not just chip demand
The Signal
NVIDIA isn't just selling GPUs anymore. The company's new AI compute framework lets cloud providers deploy compute capacity faster, but the economics have fundamentally changed. Instead of taking payment upfront for hardware, NVIDIA now shares revenue with partners over time. This is the difference between selling a car and collecting a percentage of every Uber fare it generates.
The revenue-sharing structure marks a strategic shift toward long-term financial stability, trading immediate hardware margins for recurring income streams. For a company that's spent two years as the poster child of AI infrastructure demand, this is a meaningful hedge. If AI compute becomes commoditized or growth slows, NVIDIA wants a stake in utilization, not just capacity buildout.
"NVIDIA is moving from arms dealer to coalition partner in the AI compute war."
But Burry's short position suggests he sees risk where NVIDIA sees opportunity. The same investor who spotted the subprime mortgage crisis is betting against a company at the center of the AI boom. His thesis likely isn't that AI compute doesn't matter. It's that NVIDIA's valuation assumes perfect execution on a model where they no longer control the primary variable: whether cloud providers can actually monetize the capacity.
Here's the operational reality:
- NVIDIA's old model: ship chips, collect payment, move on
- New model: ship infrastructure, share revenue, depend on partner success
- Risk transfer: from customer capex planning to NVIDIA's income statement
The partnership model makes NVIDIA's returns contingent on cloud providers successfully selling AI compute to end users. If demand for AI inference and training stays high, NVIDIA captures more value over time than a one-time hardware sale. If it doesn't, they've traded margin certainty for revenue risk.
The Implication
Watch how NVIDIA prices these partnerships. If revenue shares are generous to cloud providers, it signals concern about demand durability. If they're extractive, it means NVIDIA still sees itself as the indispensable layer and is willing to bet partners will pay up regardless.
For anyone building in the agent economy, this matters because compute costs are about to get more predictable or more expensive, depending on how these deals shake out. NVIDIA moving to recurring revenue could stabilize pricing for developers who need reliable cost structures. Or it could mean cloud providers pass along partnership costs, making inference more expensive just as agents go mainstream. Burry's short is a reminder that even the surest bets in AI infrastructure have execution risk.