E077: NVIDIA, Intel, and the Illusion of Moats: Rethinking Strategy in the Age of AI Velocity

(Transcribed by UniScribe (https://www.uniscribe.co). Upgrade to remove this message.)

Hello, welcome to the Big Strategy Podcast. I'm your host and fellow strategist, Jeff Ayad. This is your go-to spot for real talk with the masterminds of business strategy. It's about sharing stories, insights, and those little nuggets of wisdom to help you unlock big growth.

Let's turn the key and spring to action. Welcome back. Last week, I published a piece on LinkedIn called NVIDIA is Intel. Over 12,000 of you read it, so I thought I would go back and do a little deeper dive.

The argument of my article is simple. Platform lock-in always looks permanent until it isn't. CUDA, which is the software layer on top of NVIDIA's GPU, is today what x86 is on top of Intel's. It's the thing everybody wants to build on.

It's the thing nobody wants to leave. If you were a software developer and you didn't write for x86, you were out of business. Same thing in the agentic AI age. If you are not writing on top of CUDA, you won't be able to interface with NVIDIA's chips, which is the dominant hardware platform.

I got so much feedback that I said, let me just go back and reread history because the kind of moat that Intel had ultimately eroded. And it starts at the edges first. When, for example, ARM was coming in building chips for... lower power applications like smartphones instead of high power applications like PCs.

So I went back and I listened to a recent interview with Jensen on the Lex Friedman podcast, which I highly recommend. And I figured he'd dance around the lock-in question, do the CEO thing, and then change the subject. But that's not what happened. He read the same story I did.

He looked at history, the same history books that I did. and he used it to argue the exact opposite of my blog. So this episode is not about me taking my piece back. It's me doing the thing I tell every executive and also my students to do, and that is to take your best idea, go find the smartest person who disagrees with it, and see what's still standing after that.

One, company versus architecture. First thing that stopped me cold. Again, I assumed that Jensen would get a little nervous about lock-in. Instead, In that episode with Lex, he leads with it.

He says the single most important thing NVIDIA has is the installed base of CUDA. And I would imagine Intel would have said the same thing about x86 architecture. He's not defensive of it. He's proud of it.

But then as I listened a little bit deeper, take a look at what he does, how he frames this argument. He calls x86 one of the ugliest architectures ever shipped. Now I'm not a technologist. I'm not here to say yes or no.

But from my experience, both in business and now teaching strategy, I understand that the best technology isn't always the one that wins. He lands on this line. He says, quote, install base defines an architecture. Everything else is secondary.

That is not a man who's worried about my argument. That's a man who built a$40 billion company on top of my argument. So here's where my blog was sloppy. I treated Intel as one thing, and in reality, it's two.

There's Intel, the company, and then there's x86, the architecture. Intel, the company, stumbled. It lost its thread. It missed low-power chips.

It made mistakes in terms of the shift to fabulous semiconductor companies, and it lost a decade or more of time. But x86, that operating level, that won. And it's still the foundation of computing 40 years later. The company faded and the architecture is immortal.

So the real question was never, is NVIDIA the next Intel? It's a little sharper than that. Is CUDA the company or is CUDA the instruction set? Because one of those gets disrupted someday.

And we're seeing that now. Apple's going to announce their partnership with Google and Google is building their own TPUs, lower power AI chips. Amazon and XAI are building their own chips as well. So we can see that all of these players are trying to disrupt the hardware level.

But ultimately what Jensen's arguing is that if you're sitting in a strategy seat, when you name a risk, name the right layer. We're too dependent on CUDA sounds like one risk, but it's actually two. And they do not have the same shelf life as we saw with Intel and the x86 architecture. The second piece is position versus velocity.

The second crack in my own piece says that moats erode because some clever challenger finds an edge. And that's half true. But here's the half that I skipped. Intel didn't just get outflanked by ARM, for example, or AMD as a fast copycat.

Intel slowed down. As I mentioned earlier, it missed mobile. It tripped over its own factories. It went flat while the whole world kept moving.

And this is classic Clayton Christensen disruption model, right? Where companies get to a scale where they become so focused on defending their core business that they lose sight of lower margin, lower profitability, new entrant markets where startups take the place and ultimately disrupt the incumbent. So let's listen to how Jensen talks about speed. He got this idea he calls, quote, speed of light.

Every decision measured against the physical limit. Not against last year's version, but against the limit. He won't ever accept incremental improvement as the starting point. He says, don't come to me and say you got from 74 days down to 72 days.

Tell me why it isn't six, because six is the limits of physics. And here's the line that flipped for me. From a developer's chair, his pitch is almost unfair. He says, build on CUDA.

In six months, it will be 10 times better, and you don't have to do anything. You just waited. trusted Jensen and continued to build on that architecture. Jensen would argue that's not a moat.

He's building a treadmill that nobody else can run on. So I realized I asked the wrong question in the blog. I asked, when does the moat erode? The better question is, what would have to slow NVIDIA down?

Because if velocity is the moat, then Jensen spent 30 years building the antibody to the exact thing that killed Intel. He was in the business at that time. He had a front row seat. He watched it happen.

And whether that antibody holds, that's the real bet. It's not the lock-in, it's the speed. So stop tracking your competition's position. Track their rate of change.

Tracking your competition is just an X and Y axis graph. It's two-dimensional. You have to check the rate of change. That's the Z axis.

And that's where disruption occurs. Three, share versus creation. And this is the one that actually got under my skin because it questions the whole frame I was using. My blog is built around the language of moats, incumbents, challengers.

You either take share or you defend share, gain share, defend share. It's a fixed pie that gets cut into pieces. Jensen says flat out he's not in the market share business. He says there's nobody he could even take share from.

Interesting. His argument is that he's not in the business of filling and draining moats. He's in the business of velocity where no one else is playing. And here's the uncomfortable part of my argument.

The whole disrupted at the edges. Then center story assumes that there is a center, a settled market with an incumbent sitting in the middle of it. So think of the castle in the middle of that moat. But what if the center hasn't been built yet?

And I would argue in AI and in these GPUs that power the AI revolution, I don't think that they've been built. So here's where I actually land. Both things can be true at once. The pie can explode and the eventual winner can still get disrupted someday.

It just means the clock is longer and the stakes are way bigger than I wrote about. The Intel cycle took 40 years and we're not an NVIDIA in 2005. We might be at its 1995. That one company shift completely changes what you should be doing about it today.

Imagine velocity. changing the ground so quickly that you can't build a castle and thus the moat is irrelevant. So to wrap up, no, I'm not taking back the blog, but I found three places that it was thinner than I wrote. First, you have the company versus the architecture.

Second, position versus velocity. And third, share versus creation. And honestly, that's worth more to me than being right. Because the job, your job, my job is not just to have the perfect take.

or to have the perfect strategy. It's to be able to iterate, to have a strong, well thought out hypothesis and have the courage to test it, knowing that if you're not right, you're willing to adjust. Jensen read the same history I did, and he refused to assume the ending. Which look, as an instructor, you have to end your class, whether it's one class or an entire semester with a takeaway.

And in this case, I ended the class without going to the end of the story. And maybe that's the whole move. Hold the framework, question the conclusion. That's it for this one.

Short, sharp, which is how I want to do this season going forward. I'm Jeff Iatt. This is the Big Strategy Podcast. Next week, we're going to dive deeper into the strategy disruptions caused by AI.

Look forward to seeing you then. (Transcribed by UniScribe (https://www.uniscribe.co).

Upgrade to remove this message.)