NVDA 的风险是缓慢侵蚀，而非崩塌：占其营收 60% 的客户正被 75% 毛利率、65% 营业利润率吸引而自研芯片。CUDA 与 80%+ 的份额限制了近期流失，但其定价能力以及在仍在增长的市场中的份额可能缩水；高预期可能使该股受到不成比例的冲击。

Suddenly, Nvidia's biggest customers became its biggest threats. the statistics behind Nvidia are truly insane. Not only has the company experienced incredible growth over the past five years, but its profit margins have also expanded.

The figure that always catches my attention is that about 60% of revenue comes from just four customers. These customers are Alphabet, Amazon, Microsoft, and Meta.

But if you put together all these giants and cloud companies, you'll find that they represent the vast majority of Nvidia's business. Nvidia was the only company to win in the chip sector thanks to riding the wave of artificial intelligence.

a gross profit margin of 75% and an operating margin of 65% for Nvidia chips. What if we made a chip that might not be exactly as good as an Nvidia chip, but good enough for us to keep a portion of those profits for ourselves?

This appears to be a major problem for Nvidia. Yes, I think it is definitely an issue they should be monitoring. As I said, the biggest reason to get into this field is the incredible profit margins that Nvidia enjoys.

The profit margin is what each customer pays. They also lack control.

Also, as I said, even if these chips aren't the absolute best, we're still in an open debate about how much we need the best. And the adequacy of what is currently available in the market. The answer lies somewhere in the middle.

Even if you manage to reduce your spending on Nvidia by a certain percentage and you still need it for your most complex tasks, you will save a lot of money if you can do it internally.

In this case, on the other hand, Nvidia's profit margin is their inspiration to try to find a way to do at least some of the work internally. What surprised me, especially with the big cloud computing companies, is that they are pushing more of their own chips to their customers.

As you know, it really started with: "Hey, we'll provide everything you want." "But if you want to work on Nvidia chips, that's perfectly fine." But the tools they've built now with so many of these models, and the fact that you can now choose all sorts of different models, basically make it irrelevant whether you're using an Nvidia chip, a Tranium chip, or a Tensor Processing Unit (TPU).

Incidentally, Anthropic trains its models on Tension Processing Units (TPUs). I think that in the long run this will have an advantage over Nvidia, which is still an external supplier.

Let's be clear about that too. This can be said to be a good thing for everyone who is not an Nvidia user. Because this is a competition. This is a diversity of ideas. All of these things will hopefully create more efficient and less costly models for the user in the long run, rather than relying on a single type of gatekeeper.

There are many ways this can go wrong, but this is one of Nvidia's problems here: their problem looks like a solution to almost everyone who isn't an Nvidia.

Let's talk about some of the strategic moves Nvidia has made to try to counter this competition, because observing Nvidia over the past few years, I think it's easy to say, "Oh, look, they've invested in this company and that company is doing great."

Or they have already made this deal. But we are talking about circular financing. It is very clear that Nvidia needs an ecosystem of companies that build the infrastructure and that are not the big cloud computing companies, not Meta, not Alphabet or Microsoft, companies that can make their own chips.

They need all those companies to build on Nvidia's infrastructure, so they have a customer base that can't do things on its own. I think this is a very interesting strategic dynamic, because they basically created these customers themselves in many ways.

And look, at the top of their competitive advantage comes the CUDA platform. CUDA, the software that supports it, is deeply rooted. It takes a real effort to move on to other things, and frankly, some companies that fall below the level of giant corporations cannot afford that.

They invest in some open source, and they work on proprietary chips, so you know, you can't move them outside the clouds. They still control 80% or more of the market here.

There is a lot of pessimism about how that will collapse, but manufacturing capacity is currently limited as well, and they control a large part of it. I don't think the sky is about to fall.

The interesting question here is: Have we reached peak growth? Will this be an obstacle to growth going forward? And how will we, as investors, deal with that?

I don't think artificial intelligence will be built without Nvidia chips anytime soon. Let's talk about the metrics we look at to see if this strategic position is weakening. The two that come to mind are the revenue growth rate.

The complicating factor is that they are expanding beyond just graphics processing units. You have all kinds of networking equipment and all kinds of other things. So there is no direct relationship that says "Hi, we sold 50% more graphics processing units than last year."

Therefore, keep in mind that they may not capture the market share you expect if the revenue growth rate is high. But that rate will be important.

And then the margins as well. Are these the two biggest things investors should be watching? I think it's an excellent view, because look, go back a few years and you won't find that Nvidia has always been a company that has achieved what it has in the last two years.

We are paying what seems like a reasonable price now, but we are paying based on two years of tremendous and unprecedented growth.

Again, as investors, Wall Street pays for growth, but can they sustain it? Can it continue from here? This question seems more important to me than asking whether the company is about to collapse.

I would argue with anyone who says it will collapse and its value will reach zero. But this may not matter for the stock if the momentum starts to fade.

Yes, and here is a chart of revenue growth since 2010. What I find interesting here is that there were three years, full years, in which revenue was negative. There was almost a full year in 2023 in which revenues were also negative.

So this is a company that goes through these cycles, and we may be heading towards one of them in the future.

As for the other part of the margins side, I only look at their operating margin. If their operating margin returns to a level 30% lower than it is today. Suddenly, you have a stock trading at 20 times its future earnings, which seems very cheap.

But if those margins are falling and trading is now at 40 times forward earnings, with a company that isn't growing as fast, the stock won't look as cheap as it once did.

How do you distinguish between these two things? Because it looks like the slowdown will be bad, but margins are what investors should focus on, because a historically cheap stock with high margins may not actually be that cheap.

And this...you know, might not make for exciting headlines, but I think this kind of slow, boring erosion of competitive advantage is the most likely outcome from now on. So, yes, I... how do you see this?

I think those statistics are what you should really focus on. The danger is not in losing customers overnight. The danger is not in the collapse. The risk lies in reduced pricing power, a smaller share of a still-growing market, and a company that does not perform as well as it did in the past year or two.

I think this will affect the stock disproportionately to its impact on the business, simply because we have raised expectations too high here. It is certainly one of the strongest companies, and I believe it is still the most valuable company in the world, but when you rely on four, five, or six customers for the majority of your revenue, that poses a real risk to investors.

This is the case for many business-to-business (B2B) companies, where they sell to a company and not to a hundred million individuals or a billion people like Meta or Apple.

There is a real danger when these customers have the ability to design their own chips, change their ways of working, and make different purchasing decisions. Tell us what you think about Nvidia's future, and whether its biggest customers will actually become its biggest threats.