How Nvidia Became The Apple of AI, Just 10 Times Faster

"Nvidia is the Apple of artificial intelligence, one bull explained to me this past week. I don't want to sound like the Microsoft of stating the obvious, but this seems like a good thing. It means Nvidia, whose split-adjusted stock has multiplied nearly 18 times in five years, and whose data-center revenue is estimated to have tripled over the past year, might yet find ways to beat even euphoric expectations.

Let's hope so. Just this year, Nvidia blew past Amazon.com and Alphabet to become the third-largest company in the S&P 500 index, behind Apple and current leader Microsoft. That gives it plenty of pull on index fund returns-Nvidia's $1.8 trillion market value is more than 4% of the S&P 500. Its rise is a key reason that the index has returned 22% over the past year. If there are other companies in the market's top 20 with the potential to gain 224% in a year, like Nvidia just did, I'm not seeing them. Home Depot? That would be a lot of screws. JPMorgan Chase? Not even Jamie Dimon could believe that.

The same bull who compares Nvidia to Apple also tells me that Nvidia will have to "thread the needle" when it reports fourth-quarter financial results this coming Wednesday after the market closes. This is concerning. From what I vaguely remember about needle-threading, the bestcase scenario involves twisting, a bit of saliva, and hopefully no blood.

Let me start with the Apple thing and then come to the needle business. The bull is Ben Reitzes, head of technology coverage at independent Melius Research. He keeps a framed copy of his report from January 2004, when he worked at UBS and upgraded Apple stock from Neutral to Buy, where he kept it for many years. The upgrade, at what Reitzes calculates was a split-adjusted 17 cents per share, was years before the iPhone, and based on factors like iPod momentum and retail profitability.

Reitzes' first task after joining Melius 10 months ago was to launch coverage of Nvidia at Buy, but the initiation report took three months to prepare, with the stock rising the whole time. "Every day, just ta-tick, ta-tick, ta-tick," he says. "But anyway, it still worked." Today, his price target of $920 implies 24% upside from recent levels, with Nvidia entering what he calls its 30/30 season-marching toward more than $30 in earnings per share over the next few years, more than double its estimated take last year, and deserving a stock price of 30 times that figure.

The first key parallel with Apple is an end market that's turning out to be bigger than previously expected. The iPhone replaced cameras, camcorders, answering machines, and personal digital assistants, and made smartphones ubiquitous. 

Nvidia makes chips that accelerate servers, which is a must-have for AI applications, which are steadily creeping into software and services, in part because one of the things AI can do is chomp through vast oceans of data to figure out new ways to make itself useful.

That raises the question of whether all servers, or nearly all of them, will need to be accelerated in the future. "An unaccelerated server will be the flip phone of a data center within 10 years," says Reitzes. In December, Nvidia rival Advanced Micro Devices raised its estimate of the total addressable market, or TAM, for AI accelerators, including memory, to $400 billion by 2027-up from $150 billion six months prior. The new figure provides more than enough room for both companies to thrive.

A second parallel is CUDA. If you're wondering why a chip company is taking over the S&P 500, even though internet pioneer Marc Andreessen told us in a 2011 essay that "software is eating the world," it's because Nvidia, like Apple, is a software company in disguise. CUDA is a software layer that optimizes Nvidia's chips and allows customers to build applications. Its model libraries and network of experts aren't easily replicated. I can't tell you the processor speed or pixel count on my iPhone, but I can tell you that I get one every three years, and I'm not migrating 23,000 family photos and 4,300 videos to anything new. I'm sure I'm understating how data scientists feel about Nvidia's impressive chip specs, but you get the idea.

CUDA, by the way, isn't short for barracuda. That would be cooler than its actual origin story, which is that it's an acronym for "compute unified device architecture."

A third, related parallel is that Nvidia's software allows it to bring in a rising stream of revenue from recurring services, much the way Apple cashes in on application sales and subscriptions. "This whole Apple analogy for Nvidia is sort of happening in two years versus what took 20 years for Apple," says Reitzes.

Now for the needle-threading. Wall Street's consensus has Nvidia reporting $20.3 billion in fourth-quarter revenue, made up mostly of $16.8 billion in data-center business and $2.7 billion from gaming. But Nvidia has topped sales estimates by double-digit percentages in each of its past three quarters, and firms are now raising estimates last-minute. UBS analyst Timothy Arcuri this past week went to $23 billion from $20.8 billion. Customer discussions, he wrote, led him to conclude that lead times on Nvidia chips have "come in substantially."

Lead times, or delays in customers getting their stuff, are a tricky business. They can be caused by manufacturing delays, or by ravenous demand, or, in the case of Nvidia, both. There has been a dire shortage at Taiwan Semiconductor Manufacturing and other manufacturers of machines for a type of circuit packaging called chipon-wafer-on-substrate, or CoWoS. I'd explain it in detail, but I have a photo library in dire need of curating.

The important thing is that new CoWoS capacity has come on-line, and lead times are believed to have collapsed. But Nvidia must be careful to explain to investors that this is the good kind of collapse, not the bad kind. Blowout revenue and a nod toward more of the same should help-UBS predicts guidance of $25 billion to $26 billion for the first quarter. For the sake of our index funds, here's hoping that Nvidia talks up its big TAM and lucrative services, too." [1]

1.  How Nvidia Became The Apple of AI, Just 10 Times Faster. Hough, Jack.  Barron's; New York, N.Y. Vol. 104, Iss. 8,  (Feb 19, 2024): 8.