Nvidia is having an “Intel Inside” moment
The 1990s were a chaotic and exciting period in computing history. Not only was the World Wide Web finally emerging from the lab, but also PCs were becoming more useful outside the workplace. To bring order to the chaos, Microsoft and Intel forged the “Wintel” partnership, which provided a common platform for software developers, improved software compatibility, and delivered a relatively predictable consumer experience. The focal point of home computing in the 90s was the “Intel Inside” branding campaign.
There was a time when you didn’t have a “real” computer if it wasn’t a Pentium. These days, nobody cares.
Brands and retailers need each other, but wish they didn’t. Heinz ketchup needs Walmart to get their products in the hands of customers, and Walmart needs Heinz to offer shoppers the brands they expect when they enter the store. But there exists a constant tension: Walmart would prefer to sell everyone their own private-label ketchup to capture higher margins, and to strengthen the Walmart brand in the mind of the consumer. Heinz wants the same thing, but has no option but to sell through Walmart because laying out the capital to build ten thousand ketchup stores is a dubious proposition.
We oscillate between periods of retailer power and brand power. From time to time, customers may prefer buying the cheaper store brand, which weakens the negotiating position of Heinz. During other times, consumers want The Real Thing, and will demand branded products by name — crucially, during periods of extreme brand power, customers will drop their shopping cart and walk right out the door if they don’t find the brand they expect to see. This does happen occasionally, and these vicissitudes are as natural and reliable as any other cycle of human behavior— it always reverts.
We are experiencing a rare period where enterprise customers are walking into the store and demanding Nvidia GPUs by name. Whereas Heinz sold to Costco and Walmart, Nvidia sells to the hyperscale cloud providers: Google, Microsoft, and Amazon. If Amazon doesn’t have enough H100 GPUs for all their customers who want them, the customer will take their cloud workload to Google — this is the first time this has happened in the Cloud computing business. (And because you don’t want this to happen, you might buy a few extra racks of GPUs even if they aren’t needed, just to keep them out of your competitors’ hands). Just like Heinz and Walmart, the cloud providers have their own competing offerings that they’d rather sell: Google offers Cloud TPUs, which arguably offer many advantages over Nvidia’s GPUs; Amazon offers Titanium ML Accelerators; and Microsoft offers ML-accelerated FPGAs.
Apple’s incremental move toward building their own chips hammers home this desire of brands to maintain and strengthen their mindshare with customers. In addition to any technical benefits, Apple fundamentally wants to be able to say that this is an “Apple” computer from top to bottom, not an Intel-powered computer with an Apple-designed veneer. The Cloud providers have a similar desire to differentiate in this way and avoid commoditization. They may not be able to make it all happen today, but this will be the tendency over time.
The primacy of the Pentium lasted a little more than five years: demand ramped up with the launch of Windows 95, and began to subside in the early 2000s with the disappointing launch of the Pentium 4. By the mid 2000s, not only were Apple and Linux beginning to attack the market share of Windows and weakening Microsoft’s half of the partnership, and the mobile phone became a larger part of American life, but offerings from peer competitors such as AMD became much more competitive against the Intel lineup. Likewise, Nvidia faces peer competitors such as AMD and Intel racing to capture market share; Nvidia also faces latent competition from Cloud providers who make up the bulk of Nvidia’s “datacenter” segment customers.
An additional complicating factor here is that Nvidia, AMD, and many others compete for the productive capacity of a single firm: TSMC. Economics tells us that scarce resources are always put to their highest valued uses — that is, whoever delivers the most profitable end-product will have the advantage in securing scarce production capacity. And on this front, Nvidia is competing with the likes of Apple, Google, and the United States military. So, an investor looking to capitalize on temporary ML hype should think very hard about who really has the pricing power in this equation.
Cyclicality is inherent in the semiconductor industry as a whole, and it is inherent in the retailer/brand tension dynamic. There was a time when you didn’t have a “real” computer if it wasn’t a Pentium—these days, nobody cares. And now Nvidia is being valued by equity investors like Intel was in the late 1990s: for perpetually-increasing profits and world domination. That, of course, didn’t happen.