Nvidia's $91 Billion Question: Is AI Spending Getting Narrower?

Is the AI trade broadening out, or is it really just a handful of companies buying all the GPUs? That's the question Nvidia answers today when it reports second-quarter earnings after the closing bell.

The company guided to roughly $91 billion in revenue. Let that number sit for a second. That's not a yearly figure. That's one quarter. Nvidia is the world's largest company by market value, and its chips still power most of the AI training and inference happening anywhere.

The raw numbers

Here's what I'm watching. $91 billion would be another record, up from $30 billion a year ago. The data center segment has been doing the heavy lifting, and analysts expect more of the same. Margins are the other tell. If gross margins hold above 70%, demand is still outrunning supply. If they slip, Nvidia is having to work harder to move product.

But the revenue number isn't the real story. Everyone knows Nvidia is selling a lot of chips. The question is who's buying them.

Concentration is the risk

The bear case has never been that AI is fake. It's that the spending is dangerously concentrated. Microsoft, Meta, Amazon, and Google have accounted for a massive slice of Nvidia's data center sales. Throw in a few sovereign AI programs and you've got most of the demand in maybe a dozen hands.

History suggests otherwise concentration. That's not how previous tech buildouts ended. But it's also not how all of them failed. To be fair, the hyperscalers have real cash flows and real reasons to build. They aren't burning money on vaporware.

Granted, the scary scenario is simple. If one of those giant buyers pauses or stretches out its spending, Nvidia's growth story takes a hit. And because nearly every investor in the S&. P 500 holds Nvidia in some form, that hit ripples through everything.

The real question worth asking: what happens when the biggest customers run out of room to spend even more? Capex budgets can't compound at 50% forever.

What the bulls are saying

The optimists on the sell side want signs of broadening demand. They're looking for enterprise customers, sovereign AI deals, and inference workloads that keep the chips busy long after training is done. Jensen Huang keeps talking up "AI factories." The market wants proof those are real orders, not just keynote slides.

Traders are also watching guidance for the current quarter. Anything below $95 billion could spook a market that's priced in perfection. Nvidia's stock carries a premium that assumes flawless execution, not just good execution.

Admittedly, the bulls have a track record here. They've been calling this a bubble for two years, and Nvidia has kept proving them wrong.

What to watch next

Four things tonight. One, Q3 revenue guidance. Two, any comment on Blackwell supply and whether lead times are shrinking. Three, gross margin direction. Four, how much of data center revenue came from just a few customers. If Nvidia gives us that breakdown, pay attention. If it goes quiet on the topic, that's also an answer.

My take? I'm not entirely convinced the AI buildout is a bubble. The demand is real, the revenue is real, the cash flows are real. But the setup is fragile. When the world's most valuable company depends on a handful of buyers for most of its growth, the market is pricing that risk at zero.

Tonight will tell us if that's still true.