NVIDIA Corp (NVDA) Presents at Raymond James Institutional Investors Broker Conference Call - (Transcript Excerpts)
Colette Kress
Yes. So let's look at our data center business as we finish fiscal year '21. Fiscal year '21 was a record level for the total company. As we've discussed, it was a record number for gaming. But it was also a record finish for our data center business. It was the year of not only launching Ampere for gaming, but we launched the Ampere architecture for data center, starting in terms of that launch even during the lockdowns associated with the pandemic. So a true successful year for us.
Q4 was also a record. And in Q4, our data center business grew while the industry as a whole did not grow. Additionally, we are guiding growth in Q1, where our industry is also planning to be down sequentially 20%. So we feel really great about accelerated computing, AI, our overall inclusion of Mellanox and our networking business here. We are still ramping overall A100 and we have expanded and have a great set of diverse customers with the over launch -- with the overall launch of A100. Hyperscales and clouds are still ramping. But we have also added customers through our vertical industries. This is a case where we have broad-based strength across our CSPs, our vertical industries as they continue to support in our enterprises, the use of accelerated computing and AI. We've gotten great momentum from our DGXs, both at CSPs as well as our vertical industries. So we'll continue to see growth as we are in Q1 today. And I think our data center business will still be a growth driver for the years to come.
Christopher Caso
Colette, how -- what's the split between the on-premise versus hyperscale part of the data center business? And obviously, on-premise is one of the areas that has been COVID impacted. Enterprise spending in general has been ?weaker. Is that true for your business as well? And by extension, if we do expect enterprise to get better as people go back to offices, should that materialize in your data center business also?
Colette Kress
Yes. So our hyperscales are an important part of our business and have been with us for many years. And so some of the initial starts with AI and accelerated computing really was birthed with our overall hyperscale partners, and they are ramping A100. But our hyperscale business as a whole is less than 50% of our data center business. So our enterprise and verticals represent more than 50%.
But the important piece here is there are multiple ways to attract our vertical industries and our enterprises. We have cloud instances, which have been a new form to adopt computing platforms in this new wave of accelerated computing. Most of our hyperscale cloud providers have quickly deployed overall A100 into their cloud instances. This is a way for our enterprises to both test to get comfortable with the new architectures to continue to buy that on-premise basis or continue to keep it in the overall cloud.
So we have seen the strength from many years of working in terms of building out software SDKs to meet the needs of accelerated computing applications that are out there. So as we have moved into this era of Ampere architecture, the growth of those customer sets is now, not only just the hyperscales, but a very, very strong enterprise presence. We expect that we grew faster than the industry as well here on the enterprise as acceleration AI is still important, even though the economy may have some challenges with it at this time.
Christopher Caso
As a follow-on on the data center, and I'll read one of the questions that came in is about the DPU part of the data center market. And the question is, is that an entirely incremental opportunity? Or is there some cannibalization of the GPU in the data center if you're going to do more compute and sort of -- at the DPU.
Colette Kress
Yes. So let me start with discussing what is our DPU. We work together Mellanox plus NVIDIA, working together to create the opportunities for acceleration in the data center beyond GPU acceleration. And what we were looking to was include both networking and software-defined networking.
But there's a new piece as well, which is referred to as the DPU. Now the DPU provides data processing unit for that modern data center. If you step back and look at how modern data centers are being built right now. Modern data centers are disaggregating all of those different components that will exist in the data center to have racks of CPUS, racks of GPUs, racks of storage, memory. But important piece is also the DPU. The DPU is a data processing unit. It aids in the overall security of the data. The possible encryption necessary in the data, particularly on those cloud instances. So the data center as a whole is becoming the compute, not just the server. And so this DPU is a great opportunity for us to expand the market beyond our networking, beyond our overall GPU to serve the market with our overall DPU. Our BlueField-2 end market today continues to build awareness through the hyperscales, through cloud Internet companies as well. And this will likely be a part of our business and grow in the second half of this year as well as next year. We've added a new opportunity nearing about $10 billion in TAM by just working together, leadership team to leadership team, Mellanox and NVIDIA together. So we're really great to see how this will evolve as we go forward.
Christopher Caso
And how do you view the size of that DPU TAM as compared to the existing data center TAM, which is AI training, AI inference. And I think Jen-Hsun, your CEO, on the call made a comment that his view was over time, and it didn't sound like over a long period of time, that every workload -- server workload would be accelerated because of security. So I'd imagine that, that's quite a large TAM.
Colette Kress
Correct. We incorporated this as we spoke about our new size of the data center TAM as we moved forward with the inclusion of Mellanox. And so much of our work in terms of both edge computing, high-performance computing, inference training, all of these things together have allowed us to reach about a $100 billion TAM as we move forward. So just having more ability to influence the whole view of the infrastructure of the data center has been able to increase our TAM.
Christopher Caso
What about -- just moving on with that into the hyperscale part. You mentioned about half your business. And as we go forward there, how do you expect your growth rate in hyperscale to compare to hyperscale growth rates overall?
And we're seeing somewhere in the range, 20%, 25% hyperscale CapEx growth this year, a bit of an improvement from last year. Do you expect your business both in AI and DPU and everything else to grow at some multiple of that for some time?
Colette Kress
Our hyperscales are an important part of our data center business for sure. And our level of engagement with them, their support of our overall A100 architecture design has really allowed them to use A100 quite efficiently from the onset of our overall launch. We created the A100 as a system, a full system that allowed them an easy deployment, an easy ability to overall qualify for not only cloud instances, but also for their internal use.
An important piece of why A100 is such a great opportunity for these hyperscales is really about that strong performance and its universal use. Universal use of not only just doing training, but inferencing at the same time, virtual instances of both of those occurring in one overall system is a great form factor for them.
Now, why are they so interested? Well, the overall growth of what we've seen in terms of training models as we've been talking for more than 1.5 years about conversational AI. The complexity of those models, what that has driven in terms of their need for performance, their need for compute to do those AI models.
Additionally, around this work-from-home, everything on the web right now is about recommendators. That says how can you recommend the news? What I should have for lunch? Where I should go next in a safe environment? All of those are driving the interest in A100.
The ramp for A100 with our hyperscale still has a ways to go. A great end to the overall year of them both growing sequentially and year-over-year. But we will use this opportunity to continue to provide A100 to all of our hyperscales around the world.
Christopher Caso
All right. I think we've got about 8 minutes left, so this will be the last call for questions. But if I can move on to the acquisition of ARM, and you gave some updates on that on the earnings call. But I guess, my question is -- on that is what does ARM bring to NVIDIA that one you don't have already? And then secondly, why the need to buy ARM as opposed to just the licensee of ARM, which you already are.
Colette Kress
Yes. Thanks for that question. ARM is a great overall asset. And NVIDIA plus ARM together can really make the AI computing company the future together. ARM has been a 30-year tremendous success, the most energy-efficient CPU out there, a developer base that is quite enormous and serves a wide range of customers from mobile to PC to data center to IOT. The interesting thing is NVIDIA also can bring to them the overall capabilities of acceleration and AI, which that paired with an overall energy efficient CPU is a great opportunity for the overall market. So we look at this in terms of help infusing investment and understanding of the ecosystem around several very important markets, important markets of data center and PCs we can assist on in that path of creating an opportunity for a new type of CPU in the data center and CPUs -- data centers and PCs using the ARM CPU.
We love their licensing model. They're licensing -- their open licensing model is absolutely one of the greatest licensing models out there. Our ability to take some of our technology and license through their open model is absolutely one of the great things that we can do with the overall ARM acquisition. But let me step back and say, first, ARM, we are confident it will close, and we do promise to keep it open, keep its licensing open and make sure that our customers have the assurance of that existing business model that we will keep. We are excited about injecting ARM with a boost of AI and ecosystem through there. And I think that will provide more opportunities in the market, but most importantly, more opportunities for customers.
Christopher Caso
And in the event that the acquisition were unable to go through, I know that's not your intention. But as an investor looking at the risk, how would it change NVIDIA's road map? And again, you would -- if the acquisition were not to go through, you would still be a licensee. Would you still be able to pursue some of those things you talked about? And particularly in the CPU space and server, would that be something that you would be able to do if it didn't happen?
Colette Kress
Yes. So let's first start with, yes, we remain confident on the acquisition closing at this time. But the realistic look at this is NVIDIA will be a great large company, whether with it or without. But I think it's important to note that there is great optionality that we have as a licensee of overall ARM as well as continuing to assist them as we have been assisting them with CUDA, integrating CUDA into their overall work that they do today and we will continue that great partnership that we have had. But I will finish with we remain confident that the acquisition will close.
Christopher Caso
We will make a note of that, I'll finish -- we've got just a couple of minutes here, but I'll finish up with automotive. It's something that we've all talked about for quite a long time. I know it's a focus area for you. Maybe there's been disappointment in the market that Level 5 self-driving hasn't moved as fast as maybe the industry thought 3, 4 years ago. It still sees something that you're very committed to. What drives the auto space over the next couple of years? Is this still something that investors should pay close attention to?
Colette Kress
The automotive industry is working feverishly right now. It certainly has had probably one of its unique years of manufacturing lines coming to a halt due to the pandemic and then the quick surge back to those manufacturing lines and building those.
But aside from that, let's step back and look at what we're seeing in terms of autonomous. We have continued to win overall agreements for long-term Level 2+ or Level 4 robotaxis to fully deals. Probably the one -- the most transformational deals that we've put out there is our Mercedes-Daimler deal, a very important and all eyes on it just because of its unique overall structure. They have decided to use our end-to-end platform through its entire fleet that it will produce in 2024 and beyond. The important part of that is not only incorporating hardware systems within the fleet of cars and continuing to upgrade the software through over-the-air updates and keeping that current. But sharing the overall revenue that they achieve from selling that autonomous system within the car with us here at NVIDIA. So now we have an ability to have a software stream from the overall Daimler agreement. Very important deal for us, the trust we have of Mercedes-Daimler. As you know, we are also in their AI cockpit and their hyper screen that they just recently announced. And the second year of the MBOS product line that they have for their overall AI infotainment system.
We couldn't be more pleased about that deal, but we have many others. We also discussed deals in terms of -- with Li Auto. We have discussed with Xpeng. We have discussed with SAIC. All of them continuing to use the autonomous platform that we have for their cars and their electric vehicles that they will put on the road. We continue to work with many other start-ups, Tier 1s and other OEMs on Autonomous. It will take probably a couple of years for production level of these, whether it be robotaxis or passenger cars, overall hit the road. But we have clearly demonstrated that our both performance, our engineering of software, connected with our performance of our hardware is a true end-to-end performance for them for autonomous vehicles. So we couldn't be more pleased with the progress that we've seen.
Christopher Caso
Well, we'll look forward to sending an autonomous taxi to take you from the airport into one of the IIC conferences in Orlando, maybe, I don't know, maybe 2024. It sounds like it might be the reasonable, if not more, not earlier. I think that is all the time we have. |
