Fordel StudiosNvidia has another record quarter amid record capex spends
What Happened
"The demand for tokens in the world has gone completely exponential," Nvidia CEO Jensen Huang said about the company's earnings.
Our Take
The 'demand for tokens completely exponential' quote is doing a lot of work here. Jensen's admitting that training's the easy part now (commoditized, solved-ish) but inference is the real margin game. We've flipped: tokens-per-second is where Nvidia makes money now.
Here's the thing: Nvidia's record capex on record revenue means they're doubling down on supply because they know demand is irrational. They're not building for today; they're building for 2027 when every Fortune 500 company is running inference at scale.
The actual problem nobody's talking about? Energy. Electricity's gonna be the bottleneck before silicon is. Nvidia's gambling that capex on data centers is safer than betting hyperscalers solve power.
What To Do
If you're pricing LLM inference costs, assume they'll stay high for 18+ months—data center capacity is growing slower than demand.
Cited By
React