Been following Nvidia's latest moves and Jensen Huang just dropped some pretty bold takes on where AI spending is really headed. The Vera Rubin platform they're rolling out in the second half of this year is genuinely impressive - we're talking 75% fewer GPUs needed for training compared to Blackwell, plus inference costs slashed by 90%. That's not incremental improvement, that's a significant leap.

What caught my attention more though was Huang's comment about the scale of opportunity. He basically said classical computing infrastructure historically ran around $400 billion annually, but AI workloads need something like a thousand times more capacity. That's... a lot. Back in 2024 he estimated AI data center spending could hit $4 trillion yearly by 2030. At the time it sounded wild, but when you think about what's actually required to power all these models, it doesn't seem unrealistic anymore.

Looking at the numbers: Nvidia pulled in $215.9 billion revenue in fiscal 2026, up 65% year-over-year. Data center business specifically was $193.7 billion with 68% growth. They're guiding for $78 billion in Q1 FY2027, which would be 77% higher than last year. The company is basically printing money right now, yet the stock trades at a 36.1 P/E - that's 41% below its historical 10-year average of 61.6.

Forward looking, Wall Street consensus expects Nvidia to earn $8.23 per share in fiscal 2027, which puts it at a forward P/E of just 21.5. The S&P 500 is trading at 24.7 trailing. So if earnings estimates hold and the stock doesn't move, Nvidia could actually become cheaper than the broader market. Huang's clearly betting that won't happen - and honestly, given the infrastructure buildout cycle he's describing, hard to argue against it.