Aaron Rakers: I want to expand on the networking question a little bit further. When we look at the Microsoft announcement today, we think about what Meta is doing on the AI footprint that they’re deploying. Jensen, can you help us understand like where your InfiniBand networking sits relative to like traditional data center switching? And maybe kind of build on that, how you’re positioning spectrum for in the market, does that compete against a broader set of opportunities in the Ethernet world for AI fabric networking?
Jensen Huang: Yes. Thanks, Aaron. The math is like this. If you’re going to spend $20 billion on an infrastructure and the efficiency of that overall data center is improved by 10%, the numbers are huge. And when we do these large language models and recommender systems, the processing is done across the entire data center. And so, we distribute the workload across multiple GPUs, multiple nodes and it runs for a very long time. And so, the importance of the network can be overemphasized. And so, the difference of 10% in overall improvement in efficiency, which is very easy to achieve, the difference between NVIDIA’s InfiniBand, the entire software stack with what we call Magnum IO, which allows us to do computing in the network itself, a lot of software is running in the network itself, not just moving data around.
We call it in-network computing because a ton of software is done at the edge of the — within the network itself. We achieved significant differences in overall efficiency. And so, if you’re spending billions of dollars on the infrastructure, or even hundreds of millions of dollars on the infrastructure, the difference is really quite profound.
Operator: Your next question comes from the line of Ambrish Srivastava with BMO.
Ambrish Srivastava: I actually had a couple of clarifications. Colette, on the data center side, is it a fair assumption that compute was down Q-over-Q in the reported quarter because the quarter before, Mellanox or the networking business was up as it was called out. And again, you said it grew quarter-over-quarter. So, is that a fair assumption? And then I had a clarification on the USG band. Initially, it was supposed to be a $400 million, really going to what the government was trying to firewall. Is the A800 — I’m just trying to make sure I understand it. Isn’t that against the spirit of what the government is trying to do, i.e., firewall, high-performance compute, or is A800 going to a different set of customers? Thank you.
Colette Kress: Thank you for the question. So, looking at our compute for the quarter is about flattish. Yes, we’re seeing also growth, growth in terms of our networking, but you should look at our Q3, compute is about flattish with last quarter.
Jensen Huang: Ambrish, A800, the hardware, the hardware of A800 ensures that it always meets U.S. government’s clear test for export control. And it cannot be customer reprogrammed or application reprogrammed to exceed it. It is hardware limited. It is in the hardware that determines A800’s capabilities. And so, it meets the clear test in letter and in spirit. We raised the concern about the $400 million of A100s because we were uncertain about whether we could execute, the introduction of A800 to our customers and through our supply chain in time. The company did remarkable feeds to swarm this situation and make sure that our business was not affected and our customers were not affected. But A800 hardware surely ensures that it always meets U.S. government’s clear tests for export control. .
Operator: Your next question comes from the line of William Stein with Truist Securities.
William Stein: I’m hoping you can discuss the pace of H100 growth as we progress over the next year. We’ve gotten a lot of questions as to whether the ramp in this product should look like a sort of traditional product cycle where there’s quite a bit of pent-up demand for this significant improved performance product and that there’s supply available as well. So, does this rollout sort of look relatively typical from that perspective, or should we expect a more perhaps delayed start of the growth trajectory where we see maybe substantially more growth in, let’s say, second half of €˜23?
Jensen Huang: H100 ramp is different than the A100 ramp in several ways. The first is that the TCO, the cost benefits, the operational cost benefits because of the energy savings because every data center is now power limited, and because of this incredible transformer engine that’s designed for the latest AI models. The performance over Ampere is so significant that I — and because of the pent-up demand for Hopper because of these new models that are — that I spoke about earlier, deep recommender systems and large language models and generative AI models. Customers are clamoring to ramp Hopper as quickly as possible, and we are trying to do the same. We are all hands on deck to help the cloud service providers stand up the supercomputers.
Remember, NVIDIA is the only company in the world that produces and ships semi-custom supercomputers in high volume. It’s a miracle to ship one supercomputer every three years. It’s unheard of to ship supercomputers to every cloud service provider in a quarter. And so, we’re working hand in glove with every one of them, and every one of them are racing to stand up Hoppers. We expect them to have Hopper cloud services stood up in Q1. And so, we are expecting to ship some volume — we’re expecting to ship production in Q4, and then we’re expecting to ship large volumes in Q1. That’s a faster transition than Ampere. And so, it’s because of the dynamics that I described.
Operator: Your next question comes from the line of Matt Ramsay with Cowen.
Matt Ramsay: I guess, Colette, I heard in your script that you had you talked about maybe a new way of commenting on or reporting hyperscaler revenue in your data center business. And I wondered if you could maybe give us a little bit more detail about what you’re thinking there and what sort of drove the decision? And I guess the derivative of that, Jensen, how — that decision to talk about the data center business to hyperscalers differently. I mean what does that mean for the business that is just a reflection of where demand is and you’re going to break things out differently, or is something changing about the mix of I guess, internal properties versus vertical industry demand within the hyperscale customer base. Thank you.
Colette Kress: Yes. Matt, thanks for the question. Let me clarify a little bit in terms of what we believe we should be looking at when we go forward and discussing our data center business. Our data center business is becoming larger and larger and our customers are complex. And when we talk about hyperscale, we tend to talk about 7, 8 different companies. But the reality is there’s a lot of very large companies that we could add to that discussion based on what they’re purchasing. Additionally, looking at the cloud, looking at our cloud purchases and what our customers are building for the cloud is an important area to focus on because this is really where our enterprise is, where our research is, where our higher education is also purchasing. So we’re trying to look for a better way to describe the color of what we’re seeing in the cloud and also give you a better understanding of some of these large installments that we’re seeing in the hyperscales.
Jensen Huang: Yes. Let me double click on what Colette just said, which is absolutely right. There are two major dynamics that’s happening. First, the adoption of NVIDIA AI in internet service companies around the world, the number and the scale by which they’re doing it has grown a lot, internet service companies. And these are internet service companies that offer services, but they’re not public cloud computing companies. The second factor has to do with cloud computing. We are now at the tipping point of cloud computing. Almost every enterprise in the world has both a cloud-first and a multi-cloud strategy. It is exactly the reason why all of the announcements that we made this year — this quarter, this last quarter since GTC about all the new platforms that are now available in the cloud, a CSP, a hyperscaler is both — are two things to us, therefore, a hyperscaler can be a sell to customer; they are also a sell with partner.
On the public cloud side of their business, because of the richness of NVIDIA’s ecosystem because we have so many internet service customers and enterprise customers using NVIDIA’s full stack, the public cloud side of their business really enjoys and values the partnership with us and the sell with relationship they have with us. And it’s pretty clear now that for all of the hyperscalers, the public cloud side of their business would likely — would very likely be the vast majority of their overall consumption. And so, because the world’s CSPs, the world’s public clouds is only at the early innings of their enterprise to — lifting enterprise to the cloud world, it’s very, very clear that the public cloud side of the business is going to be very large.
And so, increasingly, our relationship with CSPs, our relationship with hyperscalers will include, of course, continuing to sell to them for internal consumption but very importantly, sell with for the public cloud side.
Operator: Your next question comes from the line of Joseph Moore with Morgan Stanley.
