Advanced Micro Devices, Inc. (AMD) Earnings Call Transcript & Summary

Greetings. Welcome to the AMD Conference Call. [Operator Instructions] Please note, this conference is being recorded. At this time, I'll turn the conference over to Matt Ramsay, Vice President, Financial Strategy, and Investor Relations. Thank you, Matt. You may begin.

Thank you, everyone, for joining on such short notice this morning, and welcome to our call to discuss the significant new AI partnership between AMD and OpenAI. By now, you should have had the opportunity to review a copy of our press release and our Form 8-K filing discussing this partnership. If you have not had the chance to review these materials, they can be found on the Investor page of amd.com. Participants on today's conference call are Dr. Lisa Su, our Chair and CEO; and Jean Hu, our Executive Vice President, CFO and Treasurer. This is a live call and will be replayed via webcast on our website. Before we begin, I would like to note that AMD will be reporting our full Q3 financial results on Tuesday, November 4, after the market close and remind you that AMD will host our Financial Analyst Day in New York on Tuesday, November 11, and we look forward to seeing many of you there. Today's discussion will contain forward-looking statements based on current beliefs, assumptions, expectations, speaks only of today and as such, involves risks and uncertainties that could cause actual results to differ materially from our expectations. Please refer to the cautionary statement in our press release for more information on those factors. With that, I will hand the call over to Lisa.

Great. Thank you, Matt, and good morning, and thanks to all of you for joining us on this call. Today marks an important milestone for AMD as we announced a strategic partnership and definitive agreement with OpenAI that places AMD at the center of the global AI infrastructure build-out. AI is the most transformative technology of the last 50 years, and we are in the very early stages of the largest deployment of compute capacity in history. Over the last several years, we have been laser-focused on making AMD the trusted provider for the industry's most demanding AI workloads. And we've done this by delivering an annual cadence of leadership data center GPUs, significantly strengthening our ROCm software stack to enable millions of models to run out of the box on AMD, and expanding our Rack Scale solutions capabilities. Our differentiated strategy and strong execution are paying off with AMD Instinct GPU adoption expanding rapidly. Today, 7 out of the top 10 model builders and AI companies are using Instinct, including large-scale deployments with Microsoft, Meta, Oracle, Tesla, xAI, and others. In addition, there are more than 35 Instinct platforms now in production from the leading OEMs and ODMs, and we are actively engaged in a growing number of sovereign AI initiatives. Against this backdrop, I'm very happy to announce that OpenAI and AMD have signed a comprehensive multiyear multi-generation definitive agreement to deploy 6 gigawatts of AMD Instinct GPUs. AMD and OpenAI will begin deploying the first gigawatt of Instinct MI450 Series GPU capacity in the second half of 2026, making them a lead customer for both MI450 and Helios at massive scale. Today's announcement builds on our long-standing collaboration with OpenAI that has spanned across the Instinct MI300 and MI350 series, our ROCm software stack and open-source software like Triton. OpenAI has also been a key contributor to the requirements of the design of our MI450 Series GPUs and Rack Scale solutions. Under this agreement, we are deepening our strategic partnership, making AMD a core strategic compute partner to OpenAI, and powering the world's most ambitious AI build-out to train and serve the next generation of frontier models. To accomplish the objectives of this partnership, AMD and OpenAI will work even closer together on future road maps and technologies, spanning hardware, software, networking and system-level scalability. By choosing AMD Instinct platforms to run their most sophisticated and complex AI workloads, OpenAI is sending a clear signal that AMD GPUs and our open software stack deliver the performance and TCO required for the most demanding at scale deployments. I want to thank Sam, Greg and the entire OpenAI engineering team for their collaboration and technical partnership. We are proud to work side-by-side with one of the most innovative organizations in the world as we advance the future of AI together. This partnership creates a true win-win for both companies, enabling very large-scale AI deployments and advancing the entire AI ecosystem. As part of the agreement to strategically align the interest of both companies, we are issuing OpenAI a performance-based warrant for up to 160 million shares of AMD common stock. Now let me walk you through some of the specifics. The warrant vesting tranches upon gigawatt scale GPU deployments. The first tranche vests upon deployment of OpenAI's initial 1 gigawatt purchase and subsequent tranches vest at major deployment milestones with the last tranche vesting only after the deployment of all 6 gigawatts of GPU compute. Importantly, the vesting of each tranche is also directly tied to increasing AMD stock price milestones with the final tranche vesting at a price of $600 per share. And to further align our interest, exercises of warrants are tied to OpenAI achieving key commercial and technical conditions that are important to ensure the success of their AMD Instinct deployments. This unique structure tightly aligns OpenAI and AMD, driving significant revenue and earnings growth for AMD, while allowing OpenAI to accelerate their AI build-out and share directly in the upside of our mutual success. To provide some context on the financial aspects of the agreement. Overall, we expect this structure to be highly accretive to our revenue growth and earnings and create substantial long-term value for AMD and our shareholders. From a revenue standpoint, revenue begins in the second half of 2026 and adds double-digit billions of annual incremental data center AI revenue once it ramps. And it also gives us clear line of sight to achieve our initial goal of tens of billions of dollars of annual data center AI revenue starting in 2027. Overall, we expect it will be highly accretive to AMD's non-GAAP earnings per share immediately from first revenue. We also believe that with the massive scale of this deployment and the strong benefits to the overall AMD AI ecosystem, this partnership will enable additional revenue from existing and new customers deploying at scale and has the potential to generate well over $100 billion in revenue over the next few years. In summary, today represents a major milestone for AMD, OpenAI and the entire AI ecosystem. Our partnership with OpenAI accelerates our data center AI momentum, deepens our strategic alignment with one of the industry's leading AI companies and creates substantial long-term financial value for AMD. In addition to the work with OpenAI, we have a significant number of MI450 and Helios engagements underway with other major customers, placing us on a clear trajectory to capture a significant share of the global AI infrastructure build-out. This is a truly exciting time for all of us in the industry and at AMD. The pace of innovation in AI has never been faster, and it demands full partnerships and collaboration across the entire ecosystem to push the limits of what is possible. Today's announcement is a major inflection point for us as we expand the ecosystem of partners and customers who rely on AMD to power the global AI infrastructure. Before I hand the call back to Matt for Q&A, I want to note that our focus today is on the OpenAI announcement. And we are in our third quarter quiet period, so we will not be commenting on the quarter's results today. Our near-term business momentum is strong, and we look forward to updating you on our results when we report earnings on November 4. Now I'd like to turn the call back to Matt for the Q&A session. Matt?

Thank you, Lisa. Operator, we're going to start the Q&A session now. For the Q&A session, please analysts focus your questions on today's exciting announcement. Operator, please poll for the first question. We'll allow each caller to dial one question and one brief follow-up. Thank you.

[Operator Instructions] And our first question is from the line of Timothy Arcuri with UBS.

Lisa, I know that you don't like to talk about market share, but obviously, this is a big whale in the market, and you're doing 6 gigawatts and your main GPU competitor is doing 10 gigawatts. And of course, there is some ASIC business as well. But sort of, I guess, like a two-part question. What does it say about your competitive position overall in the market? And is this ratio a reasonable sort of milepost to use for what your share could be? Could you be -- could your business be more the path of theirs in terms of data center GPU? And just wondering if you can put some numbers around that for us.

Sure, Tim. Look, this is a huge milestone for us. We have been really focused on delivering a very competitive road map. MI350 is doing very well today in the marketplace, its ramp. MI450 is a significant, significant step-up, very competitive from a technology standpoint. And look, this is a huge milestone for us. We've said that, we believe that our data center AI revenue could be tens of billions of dollars going forward. You guys have often asked me when. I think, we have clear line of sight to this being achieved in 2027. I think at gigawatt scale, there's no question that the world needs lots and lots of AI compute. And so, there's a large TAM out there. But certainly, our view is that this deal, along with, let me say, we're having a very, very active conversations with a number of other customers who are also very interested in MI450 and Helios that gives us an opportunity to be a significant piece of the market as we go forward.

And I guess, Lisa, just as a quick follow-up. So would you expect them to hold the stock? Or as the warrants vest, they will sell the stock to basically pay for the CapEx? So this is kind of like a self-funding mechanism.

I would, let's call it that, this is really a way for us to align incentives and think about it as a win-win-win. It's a win for AMD shareholders. It's a large deployment for us, very significant, tens of billions of dollars of revenue over the next number of years. The deal is structured that OpenAI must -- the warrants vest as OpenAI deploys at scale with AMD. It's highly accretive to our shareholders. I think it's also an opportunity for OpenAI to share some of that upside if we're both as successful as we plan to be. And I think, it's up to them what they do. But my view of this is, this is a very nice structure for us to be incredibly aligned between our strategic objectives, OpenAI strategic objectives. And frankly, it's a big win for our shareholders.

Our next question is from the line of Vivek Arya with Bank of America.

Lisa, I think you mentioned the potential for about $100 billion of opportunity. Does that include networking as part of your Helios ranks? Or will that be something that OpenAI will supply separately? And if it is part of your rack scale, is it Ethernet? Is it UALink? Could you give us some more details about what exactly is included in the $100 billion opportunity that you mentioned?

Yes. Let me say a couple of things, Vivek. Thanks for the question. So, first of all, when we think about our ecosystem, our ecosystem is an open ecosystem. So, when we think about our MI450 plus Helios Rack, it includes our CPU, our GPU. We have also our networking solutions from a NIC standpoint. But we're also interoperable with networking solutions. And we view this as an opportunity, not just for us, but for the entire AI ecosystem as we come together. And in terms of what's included in the various numbers, I mean, let me just go through a couple of things to make sure that we state. So revenue for this deal starts in the second half of '26. We would expect for each gigawatt of compute, significant double-digit billions of revenue for us. And then, when you think about how that accrues to AMD going forward, you should think about it as adding double-digit billions of incremental AI revenue for us once it ramps. And there is -- certainly, as we look at things going forward, we look at revenue as certainly, there's direct revenue from this deal for MI450 and next-generation products. But there's also a compounding effect. I mean, this is clear validation of our technology road map, and it is tremendous learning for us with deploying at this scale, which we think will be very, very beneficial to the overall AMD ecosystem for everyone in the industry. And so, in addition to the OpenAI opportunity and the very significant revenue addition there, we expect to generate well over $100 billion in the next several years when we think about what this accrues to our ecosystem, our capabilities, our current existing customers and new customers who now can see that AMD can deploy at very significant scale, and that's the value of this partnership all in.

And for my follow-up, Lisa, maybe something on EPS accretion. I think, you mentioned that you expect the deal to be perhaps immediately accretive. Is there some simple way to think about what things like gross margins or EBIT margins might be relative to your corporate average by the time you start to ship this? So, is there like a simple EPS math we can keep in mind per gigawatt that is deployed over time?

Vivek, this is Jean. Thank you for the question. I think, as Lisa mentioned, the way to think about it is, when ramped, we're going to generate significant double-digit billions of revenue. And the gross margin for this business is very consistent with what we discussed in the past, given the massive and fast-growing market opportunities we have, our focus is really to drive the top line revenue growth. And when you have that significant double-digit billions of revenue, we are going to generate substantial gross margin dollars. And at this scale, if you think about it, that's all incremental. Our business model is going to drive very significant operating leverage to drop to the earnings per share. The warrant is only vested based on the performance milestone when revenue is recognized. It will be included fully diluted share count only one vested and exercised. So it is very highly accretive to our bottom line.

The next questions are from the line of Joshua Buchalter with TD Cowen.

Congratulations on the announcement. OpenAI has signed deals for compute with other vendors and obviously has very big ambitions. I mean, can you maybe speak to how you would expect them to allocate their compute capacity across workloads? I mean, it's probably oversimplifying it, but should we expect MI450 to be used for both training and inference on the initial gigawatt? And how should we expect that over time?

Sure. So Josh, thanks for the question. The way I would state it is, as you know, from our road map standpoint, I think we have really been focused on ensuring that we have a very flexible GPU. So, our GPU technology from an inference standpoint is excellent, and we've had significant advantages based on our chiplet architecture for memory and memory bandwidth that are really helpful for inference. We do expect that the growth of inference is going to exceed the growth of training, and we've said that in terms of what the overall TAM is. But I think it's really for our customers to decide how they deploy. And our view is our customers are looking for the flexibility in their infrastructure to use the same infrastructure for both inference and training. I think the inference story is a very, very strong one, but we expect MI450 to also be used for training as well.

And then, as my follow-up, you mentioned this a little bit before, but could you speak to the software work that was required to get the deal over the line? What was OpenAI looking for from ROCm? And how should we think about this as validation? And like how applicable is the work you've done for OpenAI to other customers?

Yes, Josh. So, this was a tremendous amount of work. I want to say, the OpenAI team has been deeply involved with our engineering team, both hardware, software, networking, all of the above. The work that we did together really started with MI300 and some of the work there to make sure that they were running our workloads and things work. And we've done a lot to ensure that the ROCm stack, software stack is capable of running these extremely advanced workloads. I think, there's very much a joint partnership approach to how we do this. They've given us a lot of feedback on the technology, a lot of feedback on one of the most important things to them. And then, on the OpenAI side, they've been big proponents of Triton from an open ecosystem standpoint. So that has also been something that we've worked on, which Triton is basically a layer that allows you to be, let's call it, much more hardware agnostic in how you put together the models. And so, the work that we're doing together absolutely accrues to the rest of the AMD ecosystem. You should think about the hardware work, the software work, all that needs to be done in terms of just bringing the entire ecosystem to the point where you can run a gigawatt scale is all there. And we are incredibly excited about getting to work with all of this to ensure that we bring that technology across the entire AMD AI ecosystem.

Our next question is from the line of Jim Schneider with Goldman Sachs.

Lisa, I was wondering if you could comment on the -- both the data center preparedness on OpenAI side to deploy multiple gigawatts. Can you maybe comment on whether you expect the AMD deployments to come in the form of Stargate, Oracle environments or some self-built data centers and then your ability to support them in terms of supply chain preparation over the next 2 years?

Yes. Thanks, Jim. So the choice of CSP, so we would expect that these deployments would be in CSPs and the choice of CSP is really OpenAI. So talking to them about their data center environment, I think, we are actively working with all of the hyperscalers to ensure that MI450 is ready in their environment and then OpenAI will decide how they will deploy the different tranches. On the supply chain thing, I mean, we've been working on this very, very actively. The MI450, the Helios rack, 2-nanometer technology, all of the rack scale solutions require a very detailed supply chain planning. So we are absolutely ready to ensure that we deliver all of this compute. And in addition, as I mentioned, we have lots of other very important and strategic customers who are interested in MI450, and we have the supply chain capacity to satisfy this strategic deal as well as many of the other strategic relationships that we have with our other large customers.

That's great. And as a follow-up, could you maybe comment on any additional supply chain -- supply agreement terms with respect to OpenAI in terms of either pricing or preferred availability of supply relative to some of your other customers?

Yes. In terms of relative to -- again, as I said, we have a number of strategic customers. This deal is very strategic to AMD, but I want to make sure it's clear that we have a lot of other very strategic relationships as well. There's nothing exclusive about this deal. We are well positioned to ensure that we supply everyone who is interested in MI450, and we intend to do that. I think, it's just a very strategic way of putting together sort of a long-term agreement. We expect it to start with MI450, but go beyond MI450, and that's another key aspect that I want to make sure is understood. That's the reason for, let's call it, the long-term nature of the agreement.

Operator, I think given the limited time we have this morning, we have time for one more caller, please.

The last question will be coming from the line of Ross Seymore with Deutsche Bank.

Congratulations. So the first question is just on kind of the duration and the shape. I know, you said, you're starting in the second half of next year, Lisa. But any idea on how long the agreement lasts and what sort of slope on per gigawatt or annual cadence you're expecting?

Yes. I think, Ross, the key point is, the first gigawatt, we are aiming to deploy as soon as possible. So we will start with the ramp of MI450 in the second half of '26. And I think from the standpoint of the overall shape, I mean, you know that there is tremendous demand for AI compute. So this is about how do we line up the power and all of the pieces of it. But I think the idea would be to deploy as soon as we can. And from an overall deal standpoint, if you look at the 8-K, I believe the details are there, the warrant structure is set up for 5 years.

Great. And I guess, that's a perfect segue to my quick follow-up. I just want to go back to the warrant side of things. Just what was the thought process of that, because we've seen your competitor kind of go the other direction where they were investing in OpenAI and this one, OpenAI is investing in you. So just talk a little bit about how that came into the negotiation as part of kind of the overall economic equation?

Sure, Ross. I think it's a good way for us to -- since it's the last question, maybe if I take a step back and just make sure that we frame the whole thing. I think, where we are today is, we are in a place where there's a massive demand for AI compute. Like people just want more compute. I mean, I think, you'll hear that from Sam and Greg. There is -- compute is a limitation in what can be done today. And so, our goal is to build out the AI compute infrastructure. And with this structure and the warrants, it was really around creating aligned incentives for long-term agreements. I mean, this isn't just about the first gigawatt or 2 gigawatts. This is about how do we align our road map with one of the leaders in the AI industry. And so, we wanted to set up a structure that, of course, benefits AMD. I mean, we love the fact that we get to deploy lots of GPUs. We get a tremendous amount of learning from that. And OpenAI actually has to do a lot of work to make sure that our deployments are successful. And we wanted to make sure that they were motivated in the sense of OpenAI would be motivated for AMD to be successful. And the more OpenAI deployed, the more revenue we get, and they get the share and part of the upside. I think, the important piece of it is, it is all performance-based in the sense that the upside is aligned when we get more revenue, when there are more deployments, there is awesome opportunity for our shareholders to significantly benefit and OpenAI will be able to benefit as well. So that was the reason for the structure. It's actually a pretty innovative structure. I wouldn't say it came lightly. We looked at a number of different things, but this is the way that we thought we could truly -- it's really a very significant deployment in terms of just the size and the scale, and it makes it quite special.

Thank you. Ladies and gentlemen, thank you for your participation. This does conclude today's teleconference. Please disconnect your lines at this time, and have a wonderful day. Thank you.