The Business Of

Semiconductors | Nvidia | What Are Their Margins?

Morning All!

So, yesterday we dug into how Nvidia makes money. And we saw that the company basically participates in 2 duopolies. In both (i) the discrete GPU market - for PC gaming and (ii) the enterprise GPU market - for supercomputers. Nvidia and AMD are the only two real players!

And in both these markets, Nvidia is number 1 by some distance. We saw that the company’s market share in the discrete GPU market was 83%. And their market share in the enterprise GPU market was an even more outrageous 91%.

But enough with revenues. Today, we’re going to switch gears and look at what kind of margins Nvidia has. And the chart below shows us what the company’s main costs are…

As we can see, cost of revenues makes up the majority of the cost base. And given the incredible innovation involved in designing these tiny transistors, it’s no surprise that R&D is also a significant cost bucket! So, without further ado, let’s dive in…!

ASPs To The Moon!

Okay, so usually I save the gross margin and EBIT margin charts for the end of the newsletter. But today I have to show you the gross margin chart early. Why? Because what’s happened with Nvidia’s gross margins is absolutely extraordinary. The chart below shows us that the company’s gross margin went from being 29% in 2004 to 65% in 2022!

I’m not sure I’ve ever seen another company grow gross margins like that. But it does beg the question, how on Earth have Nvidia managed to do that? Well, why don’t we dive into what costs make up their ‘cost of revenues’. And take a look…

So, firstly, let’s remember that pretty much all of Nvidia’s revenues come from selling GPUs. And so the cost of revenues = the cost of those GPUs. So, what is the cost of these GPUs for Nvidia? Well as we’ve seen already this week, Nvidia is ‘just’ a designer of GPU chips. They don’t actually build the chips themselves. They outsource these tasks to TSMC, Samsung and other suppliers. And pay these manufacturers a fee for making their GPU chips. This is the cost of revenues for Nvidia! (They pay a few other suppliers for memory chips and other components. But the main cost of the GPU chips is the fee they pay TSMC and Samsung).

But wait, TSMC and Samsung are huge aren’t they? Can Nvidia really dictate terms with them and increase their gross margin like that? Well, the short answer is no. They can’t. TSMC and Samsung are massive players. And in fact, TSMC are reportedly charging designers a whopping $20,000 to process one wafer on their 3nm node. And back in 2021, Nvidia actually paid TSMC ~$7bn to get priority for the cutting-edge 5nm chips.

So, that’s clear. The cost of GPUs hasn’t really been going down for Nvidia. But what about the price they’ve been selling them for? If the price they’re selling their GPUs has been growing faster than the costs they’re paying for the GPUs, gross margins would rise! And that’s what’s been happening!

And there’s 3 elements here. First, graphics card prices have grown at a blistering rate since 2020. Data from a German retailer tells us that the average selling price (ASP) of a Nvidia graphics card was $454 in Feb 2020. In Feb 2023, the ASP had gone up 95%(?!) to $879.

Okay, so the price inflation of graphics cards (and particular the Nvidia GPU chip powering them) has contributed to higher gross margins since 2020. But what about before then? Well the second factor we’ll look at dates back to around 2017. And it’s cryptocurrency mining.

Back in early 2017, Bitcoin’s price was less than $1,000. And the cryptocurrency boom was about to get underway. During this time, crypto miners realised that the graphics cards powered by Nvidia’s GPU could help them solve the complex algorithms to attain Bitcoin. And so the demand for Nvidia’s graphics cards went through the roof! As crypto miners created ‘mining farms’ - which are basically rooms stacked with graphics cards (see screenshot below). Demand was so high that some graphics cards were being sold for 10x the original retail price. Of course, when graphics card manufacturers are able to achieve such high prices. Nvidia will lift the prices they sell to the manufacturers. And this boosts their gross margin!

These crypto ‘mining farms’ do look a little like those supercomputers we saw yesterday

Okay, so we’ve talked about factors causing the higher gross margins since 2017. But what about before 2017? Well, the third factor we’ll look at dates back to around 2010 when Nvidia’s gross margin really started to take off. And this factor is the increased importance of the enterprise GPU market.

The last decade and a half has seen cloud computing. AI. Machine learning. And other technologies become integral parts of the modern world. And suddenly, in around 2010, Nvidia realised that their main end customers may soon not be PC gamers. But huge corporations! And there’s a key impact here on gross margins. Because Nvidia realised they could charge companies like Amazon a much higher mark-up for their GPUs. As opposed to what they were charging graphics card manufacturers like Asus. So, the cloud computing/AI boom since 2010 has been the main reason why ASPs of Nvidia’s chips have risen so much. And lifted gross margins to extraordinary heights.

Remember We’re Dealing With Tiny, Tiny Things Here!

Okay, so that’s quite a good deep-dive into Nvidia’s cost of revenues. And whilst cost of revenues makes up 54% of the company’s costs. R&D makes up a very substantial 34% of Nvidia’s costs.

Now, 34% might seem like quite a lot. But when you consider that Nvidia are designing chips that are 20,000x smaller than the hairs on our head, it makes more sense. How you can even design things as small as that is pretty mind-blowing for me. And I think it’s more for a specialist semiconductor newsletter to get into the real detail of the R&D Nvidia does! But for those of you interested, check out this article on parallel programming and CUDA. Because this creation by Nvidia was one of the most significant for them as a company, and arguably all of us as electronics consumers.

The Margin Machine

Okay, so we’ve seen what makes up cost of revenues. And we’ve touched briefly on R&D. But how have these costs trended over the last couple of decades? Well, the chart below shows us that Nvidia have been spending big in both these areas. But particular in R&D! R&D expenditure has risen 18% CAGR since 2004. Whilst Cost of Revenues has grown 12% CAGR over the same period.

However, as we saw when we analysed Netflix’s costs. It’s not the total amount that matters the most. It’s how costs have grown RELATIVE to sales that matters the most. The chart below gives us this info. Showing us how Cost of Revenues and R&D have changed as a % of sales since 2004. And I’ve also thrown SG&A into the chart too…

And the chart makes for very interesting reading! Because whilst cost of revenues have continued to fall as a % of sales. (Which we should’ve already known because of the gross margin increases we saw earlier). R&D costs have done the opposite! Since 2004, R&D expenditure has outpaced revenue growth for Nvidia! Which proves just how important R&D is to the firm. As they continually look to find breakthroughs in transistor innovation.

Okay, final bit for today! And all we need to do now is put all these costs together. And have a look at how Nvidia’s EBIT margins have progressed over the years. The chart below shows us this info…

… but wait, is that right?! Nvidia’s EBIT margin was 37.3% in 2022! Isn’t that super, super high? Well, yes that is remarkably high. And tomorrow we’ll see where Nvidia’s EBIT margin ranks on our TBO margin tracker!

But equally remarkable is the huge drop-off in margin in 2023 to 16%. And on Friday, we’ll really dive into the cyclicality of the semiconductor industry. And how this impacts the margins of players like Nvidia.

3rd May 2023

Nigel Jacob CFA