Nvidia, reconsidered

Receive free US & Canadian companies updates

We’ll send you a myFT Daily Digest email rounding up the latest US & Canadian companies news every morning.

Good morning. The Atlanta Fed’s estimate of third-quarter GDP, which has looked improbably high, just keeps rising. It ticked up to 5.9 per cent yesterday. One contributor: a 0.5 percentage point boost from sold-out Taylor Swift and Beyoncé concerts plus the Barbenheimer phenomenon, according to Bloomberg Economics estimates. We await Jay Powell’s comments on this at Jackson Hole today. Email us: robert.armstrong@ft.com and ethan.wu@ft.com. 

Nvidia’s valuation, reconsidered

A week or two ago, I wrote a piece which suggested that Nvidia’s valuation might be overextended. After the company’s second-quarter earnings report on Wednesday afternoon and the market response yesterday, I feel half vindicated.

The earnings report was an absolute monster. The results were miles ahead of expectations and the outlook was too. The company expects $16bn in revenues in the third quarter; in the year ago period, revenue was $5.9bn. The company is growing like wildfire. But the stock responded by rising by all of a tenth of a per cent yesterday. That suggests that all the good news was already in the price, or in other words, that the shares might be near a valuation ceiling, for now.

I only feel half vindicated because Nvidia’s growth is so strong that I have to reconsider my argument that the stock is gripped by irrational exuberance, in something like the way Cisco shares were in 1999-2000. Cisco’s stock became so inflated that it has still not returned to the highs it hit back then, despite 23 years of great performance. But Cisco never grew the way Nvidia is growing now, so Nvidia’s valuation may make more sense than I assumed.

A little rough maths makes the valuation picture clearer. Let’s say the company makes $3.25 in net income per share this quarter, on a GAAP basis. Given the revenue guidance, this seems eminently possible. Annualise that, and the company is earning $13 a year. At the current price, that’s a forward price/earnings ratio of 36. Apple is at 27 and it is increasing revenue at a low single-digit pace. The S&P 500 is at 20. Relative to its growth, one might argue, Nvidia is cheaper than Apple or even the market. What this week’s earnings report showed was that while Nvidia’s valuation is high, it is absolutely not dotcom-crazy. (You could argue, of course, that Apple is overvalued, or that big tech is, or that the whole market is; maybe so; but what we are talking about here is relative valuation.)

The degree of Nvidia’s growth is undeniable. What about its stability?

It can appear that Nvidia’s competitive position in AI chips is unassailable, and not simply because they are better at the heavy parallel processing AI requires. Its “full-stack” products, which integrate chip design, hardware and software, are the only right-out-of-the-box solution for companies that want to add AI capabilities quickly. By the time rival chipmakers such as Intel and AMD catch up on parallel-processing chips, Nvidia’s installed base will represent an uncrossable competitive moat — something like Intel’s position in personal computing years ago with its x86 processor. 

A conversation with Gartner AI analyst Chirag Dekate convinced me that there is another way to look at it. In a certain sense, Nvidia’s principal competitors are not other chipmakers, but the big computing platforms: Amazon Web Services, Microsoft Azure and Google Cloud. These companies plan to both use AI in their own services and sell AI capabilities to customers.

Crucially, the platform companies compete on cost, and the way for them to deliver low-cost AI capacity and services is by using the computing infrastructure they already have in place. In other words, their incentive is to offer customers diverse options including both Nvidia’s out-of-the-box, highly specialised products and their own in-house technology, which can handle AI work and all the other kinds of computing work they and their customers do. And the platforms’ own purpose-built AI infrastructure could, potentially, offer more computing bang for the corporate buck. 

“If a platform uses Nvidia across the board, their cost advantage could dissipate. The only way they have a cost advantage over time is if they start eking costs from the infrastructure layers, exploring alternative pathways to efficiency,” Dekate says. He thinks that Nvidia alternatives will offer enterprises more flexibility and, potentially, superior price-performance trade-offs as soon as next year. The AI chip market is still up for grabs. “You don’t declare winners after the first inning,” he says.

Bearing this in mind, the issue is less whether Nvidia looks badly overvalued now, relative to peers or the wider market. I no longer think it does. The question is how it looks in, say, two years’ time. The answer depends on how AI services are being delivered in 2025. If you can forecast that with any confidence, by all means, email me. 

One good read

You really don’t want to be in the “quantitative eject seat”.