Analyst View on AI Demand, Inference Scale, and Hardware Cadence
Diane King Hall: Morning Movers. I’m Diane King Hall, live from the floor of the New York Stock Exchange. Joining us now is Austin Lyons, senior analyst at Creative Strategies. For a look inside the mag seven Austin, thank you so much for joining us today.
Diane King Hall: All right. We’ve had the beginning of Mag seven report earnings this week. You had Alphabet, Google and Tesla. Alphabet Google obviously had a tailwind. And it looks like the AI story is still there. But let’s talk Nvidia. We’re ways out from its results. But Nvidia holds the crown as the world’s most valuable company. Well above 4 trillion now approaching 4.25 trillion. What does that say about where we are in the AI adoption cycle to you, Austin?
Austin Lyons: Hey good morning Diane. Thanks for having me back. Happy Friday. Yes Nvidia first company to crack $4 trillion. They’re well past it like you said. And what this says about the AI era is that it’s definitely here. It wasn’t just a fad.
You know, for the first couple years we had AI labs and hyperscalers spending a ton of money buying mostly all Nvidia to train these models. But now that they’re being deployed and we’re in the inference era, we’re seeing consumers and enterprises really find a lot of value in these trained models.
So ChatGPT, for example, on the consumer side, is processing 2.5 billion queries per day from almost a billion weekly active users. Consumers are finding value, and yet enterprises are finding value, too. It’s no longer just trying little projects, but it’s actually deploying.
You know, we see nice articles from the Wall Street Journal about RSM finding lots of value from these AI reasoning models. And even I think yesterday there was an article about Walmart using AI agents. So inference is here. Demand is skyrocketing, workloads are skyrocketing and everyone’s buying Nvidia. So we have left the training era. We’re into the inference era. And consumers and enterprises are finding value.
Diane King Hall: When you think about the real world use case, there was a lot made of that. I would be a jobs destroyer, but a former colleague of mine made this point, and I think some in the analyst community see this, as well as it being a jobs creator, do you at least especially in the early years, how do you see AI and its real world use usage, especially in the early years?
Austin Lyons: Yeah, that’s really the question, isn’t it? You know, a lot of what AI is helpful for is just supercharging employees today in the, in the enterprise and letting them be more productive and get more done.
It’s even tearing down barriers a little bit. Now, all of a sudden, developers can start to expand and write requirements or try prototyping a design. And so it’s really empowering people to do more and be more creative and reach beyond.
As far as, you know, new jobs. The interesting thing about technology is that it’s always creating new jobs. So, you know, for example, I actually used to work with autonomous tractors and now there’s jobs of people managing autonomous tractors like fleet logistics for autonomous tractors. That was obviously never a job before, you know.
So I definitely think there’s going to be opportunities to increase your productivity to do more per person. But then there’ll be new jobs that we haven’t even thought about, which is always just exciting to move into. You know, as we think about the next generation, they’ll probably have jobs that didn’t even exist when they went to college.
Diane King Hall: No, it’s a great point, you know, because I think about sometimes some jokes. My son says to me, to roast me about, you know, the times, like, did that even exist in your day?
Let’s talk a little bit about Nvidia and Blackwell. And you know, like we talk about the pole position often that Nvidia has in AI. What’s the significance in terms of the leaps it’s making with Blackwell? Blackwell Ultra. And how long of a runway does Nvidia have where you see more catch up?
Austin Lyons: Sure. So Blackwell, Grace Blackwell, for example, is Nvidia’s most recent system that they built. It’s hardware. It’s software. It’s the interconnects. It’s the whole AI data center, kind of from soup to nuts. And that has been ramping very quickly.
In the last earnings call, Nvidia said that they were they had hyperscalers that were standing up a thousand racks per week. And each rack has 72 GPUs. So 72,000 GPUs per week. Obviously that’s massive.
Now, what’s exciting for Nvidia and Nvidia followers is that they are currently ramping their next generation chip, Blackwell Ultra, which has 50% more memory and 50% more compute capacity than the previous generation. There’s obviously huge demand for that because from an AI perspective, from an inference perspective, that means more tokens per second per dollar. And that’s what the name of the game is in this scaled inference, when you’re running inference at scale for these billions of queries per day.
Now Nvidia has shifted to an annual cadence. And so they’ve got a roadmap for 26 and 27 where they’ve got Vera Rubin is their 2026 offering with the new CPU and new GPU. And then beyond that, the Rubin Ultra with more memory again. So they have this consistent cadence where Nvidia is pushing the boundary of what’s possible, and they’re delivering year after year.
And the economics, the tokenomics just continue to get better for customers. So customers continue to want to upgrade and get those new systems year after year.
Diane King Hall: All right. All right, Austin, we will have to leave our conversation there for today. We’re short on time because we have some economic data to get to. We always appreciate your perspective on.