There are at least four strange things about Nvidia. 

The first is that the company has come from nowhere. Today it is 88 times more valuable than it was in 2015.

The second is the $220 billion overnight gap up in the share price two weeks back. I wrote about it last week. As far as I know, that has never happened before. The shares opened trading on Wednesday morning, after their earnings report, 26 per cent higher than they had the previous day.

The third is its multiple. Nvidia’s enterprise value is 132 times its Ebitda. And Nvidia is the sixth-most valuable company in the world by market cap. Among companies that size, no company trades on a multiple like Nvidia’s. It’s the kind of multiple you see in a very-promising micro-cap. Not a company already generating $7 billion quarterly revenue. 

And the fourth strange thing, which shouldn’t be forgotten, is the fact that this company makes its money from humanity's recent invention of artificial intelligence. Hard to overstate how strange that is.

The assumptions

I made a valuation model for Nvidia and shared it last week. You can access it here. If you want to fiddle with the inputs, to see how they feed through to the final valuation, just make a copy of it. The inputs are the numbers in the black box on the first tab. The following table shows all the inputs, or assumptions, I made to justify Nvidia’s current valuation of about $1 trillion. 

Some 13 numbers feed into the valuation. But if you fiddle with the model, you’ll note that two of them matter more than the others. One is the operating margin. And the other is sales growth. 

This is what has sent Nvidia’s valuation to the stars in the last fortnight. Nvidia is a company with high and apparently defensible operating margins of about 30 per cent. And its sales, in the last quarter, grew at a rate that’s difficult to believe. Investors aren’t quite sure what to make of sales growth like this. 

If you simply extrapolate sales growth from the most recent quarter you get an implausibly large number very quickly. Here’s what happens if you extrapolate Nvidia’s guidance for the next quarter, alongside Apple’s most recent quarterly sales growth. (Note that for Apple I extrapolated the trend based on q3 2022 rather than Q4 because Q4 is always Apple’s biggest quarter.)

Ok, that chart is silly. Nvidia is not going to maintain 59 per cent quarterly growth in sales. The question analysts are pondering is, how fast will it grow?

Above a certain point, when a company is forecast to gobble up a meaningful share of global GDP, the question is how big it’s possible for a company to get. Global Markets Insight, a market researcher, said spending on GPUs (the processors Nvidia specialises in) is forecast to grow 25 per cent per year for the next nine years. It is forecasting GPU spending will eventually reach $400 billion per year by 2032. If that’s approximately right, Nvidia can expect to take a big slice of that market. It’s the dominant producer of GPUs today. 

According to my model, the market has guessed Nvidia will grow sales by about 17 per cent per quarter. That’s the number that justifies the current valuation of about $1 trillion. 17 per cent per quarter is 49 per cent per year. That rate of quarterly growth would get it to about $150 billion in revenue by 2026.

The biggest companies — Nvidia’s peers in terms of market cap — take in $100 billion or more in annual revenue. That’s where it needs to get. It needs to be on a clear path to $100 billion or more.

The other point is, if this quarter turns out to be a flash in the pan, Nvidia will tank. From now on it will be running to stand still. It needs so much growth to justify its share price.

The competition

That’s the sales growth side of it. What about operating margins? Surely the rest of the chipmaking industry won’t stand around and let Nvidia claim hundreds of billions of dollars of profits from the rise of AI?

I touched on this last week. Nvidia’s thing is a bit like Apple’s. It’s a company that succeeds by integrating hardware and software. The software layer makes their hardware much more valuable and harder to compete with. Both companies have operating margins of about 30 per cent:

“Nvidia’s secret sauce is not a GPU technology but rather a software layer called CUDA. CUDA is a system that helps  programmers get the  most of out of Nvidia’s GPUs. It makes it easy for a non-expert to get  to work quickly. And of course, it only works with Nvidia’s GPUs.

“Artificial intelligence is hard. The software is progressing at a million miles an hour. AI experts don’t need to be stressing about how to get the most out of GPUs. They want a flexible easy to use system they can plug their models into. This is what Nvidia offers. Machine learning experts are expensive and it’s worth saving their time.

“Nvidia offers CUDA for free. But of course this isn’t out of charity: once a developer starts building their software stack on top of CUDA, they can’t leave. It’s not worth it to them.”

There’s no doubt that Nvidia’s products are what the market wants right now. In a recent presentation, Nvidia CEO Jensen Huang showed that for the same training cost, you can train a large language model 44 times faster on GPUs than CPUs. 

This efficiency is what gives Nvidia its pricing power. This is why it can charge so much.

The big question over Nvidia’s valuation is whether, in the future, AIs will live in the cloud (as they do now) or locally, on your phone. Nvidia’s AI chips live in data centres in the cloud. The cloud has the advantage of being much faster than the chips in your phone. 

But there are advantages of having an AI live on your phone. Unlike those finicky cloud-based AIs, it would help you with whatever you wanted. It would be private. It wouldn’t require an internet connection. And it could run faster. To the extent that the future of AI computation happens on your phone rather than in a data centre, it’s bad for Nvidia (and good for Apple).

You have to hand it to the Nvidia longs though. They’re thinking big, and they might even be right.