Nvidia CEO Jensen Huang’s view of generative AI’s hyper growth | interview

Jensen Huang will be giving a keynote speech at the Computex 2023 event in Taiwan over the weekend, but he’s already riding high from the resurgence in AI demand thanks to the popularity of generative AI such as OpenAI’s ChatGPT.

Huang reported earnings on Wednesday that beat Wall Street’s expectations for revenues as well as expectations for the second half of the year. That in turn fueled a broad tech rally on Wall Street.

The AI and graphics chip company’s stock price rose 27% in the past couple of day from $305 a share to $389.25 today. With a market cap of $963 billion, Nvidia’s valuation is closing in on $1 trillion.

Thanks to strength in chips for the data center, Nvidia reported revenues of $7.19 billion for the first fiscal quarter ended April 30, down 13% from a year ago but above expectations. Data center revenue in the first-fiscal quarter was a record $4.28 billion, up 14% from a year ago and up 18% from the previous quarter.

I spoke briefly with Huang after the earnings announcement this week and he filled me in on his confidence in the AI surge as well as Nvidia’s ability to line up manufacturing to meet this demand. But he didn’t think there was a surge coming in the broader economy in the second half. And he noted that gaming, while down from a year ago, is back to quarterly growth from previous quarters.

Here’s an edited transcript of our interview.

GamesBeat: It looks like you guys have a cheerful report today.

Jensen Huang: We always have cheerful reports! It’s keynote time. I’ll say something fun.

GamesBeat: I wondered how the transition would work as we go from stepping on the gas, having shortages during the pandemic, stepping on the brakes, and now stepping on the gas again in the wake of ChatGPT’s launch and the growth of generative AI. How do you feel the pattern is going to be for meeting these higher expectations now?

Huang: The good news is we were already flooring it on Hopper. We went into production in August of last year. Our timing was impeccable. Ampere, of course, is still in high demand, and it’s already in volume. The first part is our supply chain is very large, and our supply chain flow is already very high. On top of that, we still have quite a big step up in demand. We’ve procured substantially more supply for the second half of this year. We’ve responded very quickly and placed very substantial orders. We’ve procured a bunch of supply incoming.

GamesBeat: Do you see something like an economic revival accompanying this? Does the second half of the year look good for that reason as well? Or is this really just an AI phenomenon?

Huang: I think it’s an AI phenomenon. The reason for that is because the rest of the data center is still down. Data center, enterprise computing, as you know, is muted. But generative AI is doing incredibly. I think it’s very focused on generative AI. For the first time, people can see how they’re going to make money with generative AI. All these APIs can be connected to all these services and applications. It’s a lot easier to invest when you can see that return on investment. That’s number one. Number two, people finally realized, when everything clicked together, that accelerated computing saves them money and saves them power.

GamesBeat: Why are large language models so expensive to train?

Huang: Well, they’re not. They’re actually not.

GamesBeat: The impression I got from ChatGPT is that it was very expensive.

Huang: It doesn’t cost that much at all. The reason for that is–it just depends on where you started. If you’re a software engineer, you used to be able to write an application just by buying a Macbook. Now, all of a sudden, the fact that you need a supercomputer to help you develop the model seems like a lot. But let’s take it into perspective. If you were building a chip company and you were taping out a chip, the tapeout of a chip is around $100 million, just the tapeout. Not to mention the tools, which are probably another $100 million, and not to mention all the engineers, all the systems you’re bringing up, things like that. In order to build one of our chips, it’s a few billion dollars. And we’re just one chip company. There’s a whole bunch of chip companies. When they tape out a chip it’s no less than $25 million. Writing, developing a large language model–taping out a chip these days, what the software industry is learning is that building these large language models is kind of like taping out a chip.

GamesBeat: Is there much worry that we could still see shortages of some kind because of this rapid change in demand? Or are you not worried so much about that?

Huang: I think the shortages for the services are quite severe at the moment. But I think it’s going to improve tremendously in just a few months, as all the systems are delivering in real time. This is going to get better in real time.

GamesBeat: Gaming seems softer. Is that for any particular reason?

Huang: No, I thought gaming was terrific. We’re seeing sequential growth.

GamesBeat: I guess it’s still down from a year ago, though.

Huang: Yes, but that’s because of all the things from a year ago. The yearly comparisons are tough. But the quarterly results are good. The channel inventory correction is now behind us. We’re ramping Ada now across the board. I’m quite excited about that. There’s a new application for creatives in town as well. It’s called generative AI. GeForce is for gamers and creatives. Now you have generative AI to help you create things. That’s the buzz. People are very excited about that.