Jensen Huang and The Total System
Years ago, Nvidia’s co-founder and CEO outlined what I would call “The Total System,” where computers talk only to computers, with no humans in the loop. It’s one of the reasons I believe he’s looking beyond OpenAI and Anthropic and other AI leaders to the enduring value of chips.
Tiernan Ray explains a vision of chip-to-chip communications propounded by Nvidia CEO Jensen Huang years ago, one that goes beyond current AI activity.
I wanted to take you through a thought exercise about how Nvidia CEO Jensen Huang sees the current and future market for AI. It’s just under ten minutes long. I’ve added a brief promotional plug at the end, I hope readers will forgive me that! Feedback most welcome either in comments or at tiernan@thetechnologyletter.com
TRANSCRIPT
( 0:10 - 1:04) I'm going to state the obvious, which is that artificial intelligence depends on computer chips. I think we all know that. But I think sometimes people don't really think through what the implications of that are. This week, we saw some very, uh, impressive numbers from both Nvidia and Marvell, which also participates in the computer chip market for AI. And that lifted the shares of Nvidia, Marvell, not so much. What people tend to miss who look at OpenAI, Anthropic, SpaceX, and focus on the “frontier model” race in AI models, is just how much the computer chip companies such as Nvidia have this whole market in their grasp. Because nothing you can do with frontier models can exceed what the chip provides you.
(1:47) The semiconductor, such as a GPU chip, is the basis for all of this. It always determines what AI scientists can do at any point in time, given what is the capability of the latest chip. And in the case of Nvidia, I think that Jensen Huang, the co-founder and CEO, has been thinking for many years far beyond the present moment. And it’s one of the reasons that his company has been willing to put up its balance sheet to backstop OpenAI's projects, and projects of many others; to go to financial firms such as BlackRock and Blackstone and Goldman Sachs to securitize future chip sales, which seems a highly risky prospect.
(2:26) Huang is looking beyond the moment of these AI companies to the chip ultimately being the dominant force here. And the crux of his thinking, in my view, is that there will be lots of activities of agentic AI that communicate with one another outside of the involvement of individual human beings. That automated systems will be generating huge amounts of traffic. And he spoke to this somewhat on Wednesday evening's conference call for earnings, talking about all these agents running around doing all this stuff.
(3:14) And that talk about agents has been in the air for a while now, but I think people don't appreciate how complete Huang believes this agentic activity will take over much of computing activity. And to give you the perspective, I want to go back to something I wrote six years ago. This was in October of 2020. This was before ChatGPT, and before the prominence of OpenAI, before Anthropic existed, before SpaceX's Grok was a thing, when large language models were, kind of, still a science project. Huang was talking to a bunch of journalists, including myself, about the direction for the company's GPU chips.
(4:18) This was at a GTC conference. And Huang expounded on how computing is limited by humans. If I can quote from the write-up that I did on The Technology Letter at the time, Huang said that "The total amount of computing that the world can use is limited by humans today." Limited in what respect? Well, it's because humans just take too long. He said, "There are only a few billion of us, humans, and it takes nine months to gestate, and then it takes years to raise them, and then when you finally get them to a point where they're somewhat intelligent," meaning humans, "they poke at these things called smartphones, and every time you poke at them, it creates queries in the cloud." So, activity in data centers moves at the speed of humans doing this, and doom scrolling.
(5:00) It's too slow, in Huang's view. And, so, what's the solution? "Well," he said, "in the future, it won't take nine months to build a new intelligent being, and it won't take years to raise them. It literally takes an hour to manufacture a BMW car. It takes seconds to download the artificial intelligence, and it's right on the Internet within seconds after that. There'll be billions of these things." [I mix-spoke. He said “trillions.”] But he was basically talking about agents, agentic AI, these autonomous computer processes that compute together with one another.
(5:52) And, in conclusion, he said, "And so the next Internet, and the next set of data centers are going to be much larger than this one. Orders of magnitude larger than this one. Many orders of magnitude." All coming to pass now, of course, with projects; for example, Ports Pike is the five hundred billion dollar data center in Ohio to which Huang has committed his balance sheet. You put it together, what you see is Huang looking to a time when, no matter what happens with these frontier AI labs, chip-to-chip communication via autonomous processes, software that is talking to software, survives whatever current incarnation there are.
(6:29) And you could see elements of this in this week's earnings call. Both Huang and his CFO, Colette Kress, spoke about their technology being fungible. Fungible means, can be adjusted to operate in a different context. And, in this limited sense of the conference call, they meant, can operate for different applications. The same GPU chip can do different stuff. But, in a sense, I also think that, uh, perhaps metaphorically, fungibility means that the customers at present of Nvidia are fungible.
(7:02) Today, it's OpenAI, Anthropic, SpaceX, right? Elon Musk at SpaceX has committed his company to only going with Nvidia. Huang is looking past all of these customers to the activity that will continue on his chips decades into the future so that any one customer account is fungible. You could lose all of these customers, and there would still be an activity of computer chips, and therefore a market that supports computer chips. Now, of course, this has some implications near term.
(7:33) If you lose gigantic customers worth a trillion, two trillion or more, each, you lose a lot of business. And, I think, in a sense, Huang is looking past, through, the possibility of a downturn in his market. You could lose all these companies, stocks could go down, and, on the other side of it, you continue to have this chip market and chip activity because, Why? It is inevitable that these AI processes will talk to one another.
(8:10) And, for common stock investors, that's kind of chilling because that prospect is one of a market downturn, a big market downturn, where lots of financial bets come apart. Huang saw his company's stock decline by 60% in 2022. So, he has been through, over decades, many market downturns, and he has come out okay on the other side. And, so, I think, if you could say, the long bet that he's placing is the total system.
(8:53) That's the term I've given to it. The total system is chips talking to each other. And current-generation large language models, diffusion models, transformers — all these AI things — are one moment of that, one moment of the use of the chip. But, what is inevitable is this total system of chips talking to chips, software processes communicating with software processes, and the slow pace of human activity extracted, removed as the impediment. And that survives after whatever the current market participants are who happen to be his gigantic customers.
(9:24) It's just my theory. I can't prove it. But from watching Jensen Huang for years, decades, I think that he views the chip, and particularly his chips, as the ultimate answer to activity, productive activity, and the ultimate kind of technological cultural artifact that survives whatever is the current moment of all of these AI companies that are the focus of everyone.
(9:48) So, don't forget the chips. Don't forget that at the end of the day, they really drive everything. And, so, the intention of the people running the chip companies, such as Jensen Huang, really matter.