Advanced Micro Devices [AMD] Conference call transcript for 2023 q2
2023-08-01 19:42:09
Fiscal: 2023 q2
Operator: Greetings, and welcome to the AMD Second Quarter 2023 Earnings Conference Call. At this time, all participants are in a listen-only mode. A brief question-and-answer session will follow the formal presentation. [Operator Instructions] As a reminder, this conference is being recorded. It is now my pleasure to introduce to you, Mitch Haws, Head of Investor Relations. Thank you, Mitch. You may begin.
Mitch Haws: Thank you, and welcome to AMD's second quarter 2023 financial results conference call. By now, you should have had the opportunity to review a copy of our earnings press release and accompanying slideware. If you've not reviewed these documents, they can be found on the Investor Relations page of amd.com. We will refer primarily to non-GAAP financial measures during this call. The full non-GAAP to GAAP reconciliations are available in today's press release and slides posted on our Web site. Participants on today's conference call are Dr. Lisa Su, our Chair and Chief Executive Officer; and Jean Hu, our Executive Vice President, Chief Financial Officer, and Treasurer. This is a livecom, and will be replayed via webcast on our Web site. Before we begin, I would like to note that Jean Hu will attend the Jefferies Semiconductor, IT Hardware and Communications Summit on Tuesday, August 29, and the Deutsche Bank Technology Conference on Thursday, August 31. Dr. Lisa Su will attend the Goldman Sachs 2023 Communacopia & Technology Conference on Tuesday, September 5. Our third quarter 2023 quiet time is expected to begin at the close of business on Friday, September, 15. Finally, today's discussion contains forward-looking statements based on current beliefs, assumptions and expectations speak only as of today and, as such, involve risks and uncertainties that could cause actual results to differ materially from our current expectations. Please refer to the cautionary statement in our press release for more information on factors that could cause actual results to differ materially. With that, I'll hand the call over to Lisa. Lisa?
Lisa Su: Thank you, Mitch, and good afternoon to all those listening in today. We executed well in the second quarter, launching multiple leadership products, significantly expanding our AI engagements, and ramping our latest Zen 4, EPYC, and Ryzen product families. Second quarter revenue declined 18% year-over-year, to $5.4 billion. Sales were flat sequentially as Client and Datacenter segment growth was offset by expected declines in our Gaming and Embedded Segments. AI cluster engagements grew by more than seven times sequentially as multiple customers initiated or expanded programs supporting future deployments of Instinct MI250 and MI300 hardware and software at scale. Looking at the second quarter business results, Datacenter segment revenue, of $1.3 billion was down 11% year-over-year, and up 2% sequentially. Although market demand remains mixed, the 4th Gen EPYC CPU adoption accelerated in the quarter, with revenue nearly doubling sequentially as cloud provided expanded deployments to power their internal infrastructure and public instance offerings. In cloud, 30 new AMD instances launched in the second quarter, with multiple Genoa instances announced by AWS, Alibaba, Microsoft, and Oracle. Genoa delivers up to 1.9 times more performance in enterprise and cloud applications, and 1.8 times more performance per watt than the competition, making it, by far, the industry's higher performance and most efficient server processor. As an example, AWS announced its M7a Genoa instance, which is the highest performance and best price-performance general purpose x86 instance they offer. In total, there are now more than 670 AMD-powered cloud instances publicly available, and we expect that number to grow 30%, to nearly 900 by the end of the year, driven largely by new Genoa deployments. We also expanded our Zen 4 server product portfolio in the quarter with the launches of Bergamo and Genoa-X. Microsoft Azure announced the first Genoa-X HPC instances that offer more than five times higher performance in technical computing workloads compared to their prior generation. With Bergamo, we deliver more than double the performance than competitive offerings for cloud-native applications, while offering full x86 software compatibility. We were excited to be joined at our launch event by Meta, where they announced plans to deploy Bergamo broadly across their global datacenter infrastructure to power applications, including Facebook, Instagram, and WhatsApp. Looking ahead, Dell, HPE, Lenovo, Supermicro, and other large server providers are on track to begin launching their new Bergamo platforms in the third quarter. In enterprise, while macroeconomic uncertainty resulted in weaker customer demand year-over-year, sales of EPYC processors for enterprise servers grew sequentially as we closed multiple wins with large energy, technology, financial services, and healthcare companies. Overall, pull from large enterprises continued to grow. For example, Banco de Brasil, BNP Paribas, Petronas, Uber, and other large enterprises, all adopted EPYC processors in the quarter. And SAP selected EPYC processors to power RISE with SAP applications hosted on Google Cloud. We expect EPYC revenue to grow by a double-digit percentage sequentially in the third quarter, led by the expanding 4th Gen EPYC CPU ramp. In addition, Siena, our first EPYC processor optimized for leadership, Edge server, and teleco infrastructure is on track to launch this quarter. Turning to our broader Datacenter business, in networking, the largest cloud providers expanded their adoption of Pensando DPUs in the quarter, highlighted by new deployments with Alibaba and Oracle Cloud. In supercomputing, EPYC and Instinct processors continue to be the solutions of choice for the most powerful supercomputers in the world, powering 121 of the fastest systems on the latest top-500 list, and seven of the 10 most efficient systems on the Green500 list. In AI, we made strong progress in the second quarter as we met key hardware and software milestones to address the growing customer pull for our datacenter AI solutions. Our AI strategy is focused on three areas. First, deliver a broad portfolio and multi-generation roadmap of leadership GPUs, CPUs, and adaptive computing solutions for AI inferencing. Second, expend the open and proven software platform we have established that enables our AI hardware to be deployed broadly and easily. And third, expand the deep and collaborative partnerships we have established across the ecosystem to accelerate deployments of AMD-based AI solutions at scale. We delivered on all three fronts in the second quarter. On the software and partnership side, Hugging Face announced plans to optimize thousands of their models for AMD Instinct, Ryzen, EPYC, Radeon, Versal, and Alveo platforms. To make it easier for developers to tap into the full performance and features of our AI hardware, we delivered a significant performance and feature update in our latest ROCm software, and expanded support for AMD silicon across the leading frameworks, including PyTorch, TensorFlow, Onyx, and technologies like OpenAI Triton. We are receiving positive feedback on the improvements and the new capabilities of our latest ROCm software stack from our AI customers and ecosystem partners. As an example, leading AI software company, MosaicML, recently highlighted that our Instinct MI250 accelerator delivers competitive training performance with minimal or no changes to the underlying AI software. On the hardware side, we announced our new Instinct MI300X GPUs designed to be the world's most advanced accelerators for generative AI. MI300X combines our next-gen CDNA 3 architecture with the industry's largest memory footprint and fastest memory bandwidth. These are critical factors in AI inferencing performance. Customer interest in our Instinct MI300A and MI300X GPUs is very high. Engagements with top-tier cloud providers, large enterprises, and numerous leading AI companies significantly expanded in the quarter. We are providing early system access and sampling both products with our lead AI, HPC, and cloud customers now, and remain on track to launch and ramp production in the fourth quarter. Turning to our Client segment, revenue declined 54% year-over-year to $1 billion. Client segment revenue increased 35% sequentially as Ryzen 7000 series CPU sales grew significantly, led by the launches of new notebooks from the largest OEMs. We also launched new commercial offerings with our first Ryzen Pro notebook and desktop processors powered by our leadership Zen 4. More than 100 AMD-powered commercial PC platforms are on track to launch this year from HP, Lenovo, and other leading OEMs as we grow this important part of our client business. We expect our Client segment will grow in the seasonally stronger second-half of the year based on the strength of our product portfolio and increased adoption of our Ryzen 7000 CPUs, including the ramp of our Ryzen 7040 mobile CPUs that deliver leadership performance and energy efficiency, and are the industry's first x86 processors with a dedicated AI engine. Going forward, we see AI as a significant PC demand driver as Microsoft and other large software providers incorporate generative AI into their offerings. We are executing a multi-generational rise in AI processor roadmap which, together with our ecosystem partners, will fundamentally change the PC experience. Now, turning to our Gaming segment, revenue declined 4% year-over-year to $1.6 billion as higher semi-custom revenue was more than offset by lower gaming graphic sales. Sequentially, segment revenue declined 10%. Semi-custom SoC sales were strong in the quarter as Microsoft and Sony had healthy console demand based on improved retail availability globally, and the launches of new AAA games. In gaming graphics, we expanded our Radeon 7000 GPU series in the second quarter with the launch of our mainstream RX 7600 cards for 1080p gaming. We are on track to further expand our RDNA 3 GPU offerings with the launch of new, enthusiast-class Radeon 7000 series cards in the third quarter. Turning to our Embedded segment, revenue increased 16% year-over-year to $1.5 billion. Sequentially, revenue declined 7% as solid demand with industrial, vision and healthcare, automotive, and broadcast customers was offset by softness with communications customers as some operators slowed their infrastructure upgrades. We expanded our leadership adaptive computing product portfolio in the quarter, launching our new Versal Premium VP1902 adaptive SoC with advanced chiplet packaging, the industry's largest and most performance solution for emulating and verifying next generation ASICs and SoCs. In the low end, we announced our Spartan UltraScale+ FPGA family to address a new range of cost optimized, industrial, computer vision, healthcare and robotics applications. We also released enhanced versions of our Vivado and Vitis software platforms that make it easier for customers to develop highly performant applications for our Versal Adaptive SoCs. Embedded CPU sales grew in the quarter with the launches of new AMD powered security, storage and networking solutions from HPE, Fortinet and other leading vendors. Looking into the second-half of the year, after delivering six quarters of very strong year-over-year growth, we expect Embedded segment revenue to decline in the back-half of the year as lead times normalize and some customers reduce their inventory levels. We continue to be very pleased with our embedded design win momentum and in particular, the growing revenue synergy opportunities we see based on our combined adaptive and embedded processing product portfolio. In summary, we executed well in the quarter against our strategic priorities. Looking at the second-half of the year, we expect the PC market to grow seasonally with more normalized inventory levels across the supply chain. In the datacenter market, we see a mixed environment as AI deployments are expanding. However, cloud customers continue optimizing their datacenter compute and enterprise customers remain cautious with new deployments. Against this backdrop, we expect strong growth driven by higher fourth gen EPYC and Ryzen 7000 processor sales and initial shipments of our Instinct MI300 accelerators in the fourth quarter. Longer term, while we are still in the very early days of the new era of AI, it is clear that AI represents a multibillion dollar growth opportunity for AMD across cloud, edge and an increasingly diverse number of intelligent endpoints. In the datacenter alone, we expect the market for AI accelerators to reach over $150 billion by 2027. We have increased our AI related R&D, ecosystem enablement and go-to-market investments to capture a significant share of this emerging market. The strong progress we are making executing our AI roadmaps and the rapid pace at which we are expanding our ecosystem of AI hardware and software partners makes us very confident we can deliver leadership, training and inference solutions powered by our instinct EPYC, Ryzen AI, Versal and Alveo platforms for our customers and partners. Now I'd like to turn the call over to Jean to provide additional color on our second quarter results and our outlook for Q3, Jean?
Jean Hu: Thank you, Lisa, and good afternoon, everyone. I'll start with a review of our financial results for the second quarter and then provide our current outlook for the third quarter of fiscal 2023. We are pleased with our second quarter results with revenue of $5.4 billion and a diluted earning per share of $0.58. On a year-over-year basis, revenue declined 18% as growth in the Embedded segment revenue was more than offset primarily by lower Client segment revenue. Revenue was flat compared to the first quarter, as growth in both the client and the Datacenter segments was offset by expected declines in the Gaming and Embedded segments. Gross margin was 50%, down approximately four percentage points from a year ago, primarily driven by lower Client segment performance, partially offset by strong Embedded segment performance. Operating expenses were $1.6 billion, an increase of 3% year-over-year primarily due to higher R&D investments. Operating income was $1.1 billion, down $114 million year-over-year and operating margin was 20%. Interest expense, taxes and other was $120 million. For the second quarter, diluted earning per share was $0.58, compared to $1.05 in the same period last year. EPS declined on a year-over-year basis, primarily due to Client segment performance. Now, turning to our reportable segments for the second quarter, starting with the Datacenter segment, revenue was $1.3 billion, down 11% year-over-year, mainly due to lower third generation EPYC processor sales as enterprise demand was softer and inventory levels were elevated at certain MDC customers. Datacenter revenue grew sequentially with strong sales of our fourth generation EPYC processors, specifically Gen 1 partially offset by decline in adaptive SoC product sales. Datacenter segment operating income was $147 million or 11% of revenue, compared to $472 million or 32% a year ago. Lower operating income was primarily due to lower revenue and increased R&D investment to support future growth. Client segment revenue was $998 million down 54% year-over-year, due to reduced processor shipment resulting from a weaker PC market and significant inventory correction across the PC supply chain. On a sequential basis, revenue grew 35% as we ramped our Ryzen 7000 series processors and the PC market conditions improved. Client segment operating loss was $69 million, compared to operating income of $676 million a year ago, primarily due to lower revenue. We expect the Client segment to return to profitability in the third quarter. Gaming segment revenue was $1.6 billion, down 4% year-over-year. Semi-customer revenue grew year-over-year which was more than offset by lower gaming graphics revenue. On a sequential basis, gaming revenue declined 10% in line with our expectations. Gaming segment operating income was $225 million, or 14% of revenue, compared to $187 million or 11% year ago, primarily due to higher semi-customer revenue. Embedded segment revenue was $1.5 billion, up 16% year-over-year, primarily driven by strength in the industrial vision, healthcare, automotive, test and emulation market. On a sequential basis, Embedded segment revenue declined 7%, primarily due to weaker communication market demand. Embedded segment operating income was $757 million or 52% of revenue, compared to $641 million or 51% a year ago, primarily driven by higher revenue. Turning to the balance sheet and the cash flow, during the quarter, we generated $379 million in cash from operations. Free cash flow was $254 million. Inventory increased by $332 million to support the continued ramp of Advanced Technology products, we expect inventory to decline as we ship this product to customers in the second-half of the year. At the end of the quarter, cash, cash equivalents and short-term investment was drawn at $6.3 billion. Now, turning to our third quarter 2023 outlook, we expect revenue to be approximately $5.7 billion plus or minus $300 million and an increase of approximately 2.5% year-over-year and approximately 6.5% sequentially. Year-over-year, we expect revenue for the Client segment to be up, Datacenter segment to be flattish, and the gaming and Embedded segments to decline. Sequentially, we expect the Client and the Datacenter segment to each grow by a double-digit percentage and the Gaming and Embedded segments to decline. Non-GAAP gross margin to be approximately 51%, non-GAAP operating expenses to be approximately $1.65 billion, non-GAAP effective tax rate to be 13%, and the diluted share count is expected to be approximately 1.63 billion shares. In closing, I'm pleased with our second quarter top line and bottom line execution. We expect our new product ramps across the Datacenter and Client segment to drive sequential growth into the third quarter. Importantly, our leadership product portfolio, datacenter and AI investment priorities and financial strengths position us well for long-term growth. With that, I'll turn it back to Mitch for Q&A session.
Mitch Haws: Thank you, Jean. John, we're happy to poll the audience for questions.
Operator: Thank you. We will now be conducting a question-and-answer session. [Operator Instructions] And the first question comes from the line of Matt Ramsay with TD Cowen. Please proceed with your question.
Matt Ramsay: Yes, good afternoon, and thanks for taking my questions, and congrats on the results. I guess, Lisa, my first question is around the Datacenter business. I think we're all, across the industry, observing a shift in workload and spending patterns like maybe we've, arguably, never seen. And your company is in a great position to participate on both sides of that on the CPU strength, and obviously in the AI space. Last quarter, you had given us some metrics around potentially being able to grow your datacenter business by 50% in the second-half of the year versus the first-half. And maybe you could give us a little bit of an update on how you're thinking about that milestone and the drivers of growth across CPU and accelerator for the back-half? Thanks.
Lisa Su: Yes, sure, Matt. Thanks for the question. So, you're absolutely right. It's a very dynamic market right now in the datacenter. We certainly see -- let me go through some of the pieces. So, on the positive side, we certainly see that acceleration of AI demand. From our standpoint, we see it in a couple ways. We have a number of design wins in AI deployments as the CPU that goes with GPUs, as well as other accelerators. So, in the head nodes, we've seen that positive on the CPU side. We've also seen some strong interest in our MI250 accelerator, which is currently shipping right now. And we see very strong pull on the MI300 accelerators that are starting production in the fourth quarter. So, those are the positive market dynamics as we go into the second-half of the year. We also see some of the softer cloud spend that is happening outside of AI as some of the cloud vendors are optimizing their CapEx. And enterprise, I would say is still on the weaker side. But with all that in place, we are expecting a large ramp in second-half for our Datacenter business, and weighted towards the fourth quarter. And we are still looking at a zip code of, let's call it, 50% plus or minus second-half to first-half. So, it's a big ramp, but when we look at all the components, I think that the customer pull is certainly there. And it's exciting to be in this part of the industry.
Matt Ramsay: Thank you for that, Lisa. I guess as my follow-up, still sticking with the Datacenter business. Your company is aggressively trying to ramp both the hardware and the software side of the MI300 programs to support AI. There's been some conflicting reports as to whether all of those deployments are time. I think you've, in the prepared script, said what you guys think about that. I guess my question is really around the software work and the hardware itself that you're doing with your lead customers, maybe you could give a little bit about, firstly, how the customer feedback has been on the performance of the hardware itself? And secondly, how you think the software work you're doing with your lead customers will translate into other customer deployments as we work through next year? Thanks.
Lisa Su: Yes, sure, absolutely. So, if I give you just some color on how the customer engagements are going, there's very strong customer interest across the board in our AI solutions, that includes, let's call it, multiple tier 1 hyperscalers that we're engaged with. It includes some large enterprises. And it also includes this new category of some of these AI-centric companies that are sort of very forward-looking in terms of how they're deploying and building AI solutions. So, from that aperture, we made a lot of progress with our ROCm software stack. I'm actually -- there is a lot more to do, but I would say the progress that we've made has been significant. We're getting lots of feedback from those lead customers. We're seeing the benefits of the optimization, so working also on the higher-level model frameworks, the work that we're doing with the PyTorch Foundation, the work that we're doing with Onyx, with Triton. And the key is we're getting significant real-time feedback from some of these lead customers. So, we're learning at a very fast pace. In terms of the feedback on performance, a number of companies have now been able to look at MI250 across a broad range of workloads, and that's a good translation as you go to MI300, and the feedback has been quite positive. We have customers sampling either on our lab systems, they're accessing the hardware, or sampling in their labs. And I would say, so far very positive. The pull is there. There is a lot of work to be done, but we feel very good about the progress of our overall AI solutions for the Datacenter.
Matt Ramsay: Thanks, Lisa.
Lisa Su: Thanks, Matt.
Operator: And the next question comes from the line of Aaron Rakers with Wells Fargo. Please proceed with your question.
Aaron Rakers: Yes, thanks for taking the question. Just building on Matt's comments or question, I just want to go back to the implied revenue for the Datacenter business for the back-half of the year. Jean, I think, last quarter, you had alluded to, for the full-year, the expectation is still growing 10% or double digits, I should say, for the full-year the Datacenter business, just confirming that. And what I'm really trying to ask is, given the guidance of flat year-over-year growth in Datacenter in 3Q, it would seem, if my math is correct, you're implying a 50% or so increase sequentially into 4Q. I'm just trying to frame exactly how you're thinking about the cadence of what 4Q looks like, underpinning that expectation?
Jean Hu: Hi, Aaron. Thanks for the question. I think as Lisa just mentioned earlier, it's a very dynamic market. There are puts and takes. We have a tremendously strong momentum with our product portfolio, but there is continued softness in enterprise market, and also call it, the optimization is still ongoing. So, overall on balance, we think year-over-year it's probably more like a high single-digit. It's really strong ramp, not only in Q3, right, sequentially earnings double-digit -- strong double-digit. And the Q4, of course we're going to see continued sequential strong ramp.
Aaron Rakers: Yes, that's helpful, Jean. And then just following up on that as well, how have you guys managed through, with that ramp in mind, the supply chain side? I know that your manufacturing partners talked about expanding their capacity significantly. Just curious of what you're seeing as far as being able to fulfill that degree of demand as we look into, not just this quarter, but into 4Q?
Lisa Su: Yes, sure, Aaron. So, we have been really investing in our supply chain, the Datacenter growth is so strategic to us, that this has been part of the strategy. So, if you look at all aspects of the supply chain, from the wafers to the backend capacity, to some of the specific components that you need to do something of the class of MI300, we've worked with the entire supply chain. We feel that we have ample supply for an aggressive ramp in the fourth quarter and into 2024. But this is certainly one of the areas that we spent quite a bit of time to ensure that we do have that confidence.
Aaron Rakers: Okay. Thank you.
Lisa Su: Thanks, Aaron.
Operator: And the next question comes from the line of Toshiya Hari with Goldman Sachs. Please proceed with your question.
Toshiya Hari: Hi, thank you so much for taking the question. My first one is on the Datacenter business as well. And I just wanted to follow up on the Q3 to Q4 dynamic. And I do apologize if I missed this, but in the implied growth rate in Datacenter in Q4, can you speak to what percentage of supercomputing. I think there is a big project that's slated to ship in Q4. And is there any contribution from the Instinct series outside of supercomputing as well or is it primary your server CPU franchise?
Lisa Su: Yes, sure, Toshiya. Thanks for the question. So, as Jean said, into the third quarter, we expect double-digit sequential growth in Datacenter, that's primarily EPYC. So, that's primarily the Zen 4, let's call it the combination of Genoa and Bergamo, as that continues to ramp. As we go into the fourth quarter, there is an implied significant ramp in revenue. I think there are multiple components to that. So, there is -- the server CPU side will continue to ramp as we see Zen 4 ramp. There is a sort of large, call it, lumpy supercomputer win, so our El Capitan win will be in the fourth quarter primarily, with a little bit in the first quarter. And then we will have contribution from both MI300X going to large AI customers as they start their initial ramps, as well as MI250s with a number of customers who have now -- view that as a very good option for some of the workloads that are not necessarily the largest language models or the largest parameters, but let's call it more sort of the other AI workload. So, those are the components of the fourth quarter implied growth. Lots of pieces to it, but clearly a big piece of it is the MI300 ramp.
Toshiya Hari: That's helpful, thank you, Lisa. And then shifting gears a little bit and follow-up question on the Client side. You talked about the business returning to profitability in Q3, which is great. But you're still well below where you were in '21 and '22 from an operating margin perspective. Can you speak to the competitive landscape in the client business? Is there a path back to, call it, 20%, 30% operating margins there? And do you have any cost initiatives ongoing to get you back to that level of profitability in Client? Thank you.
Lisa Su: Yes, sure. Maybe let me start, and then maybe Jean can add some comment. So, look, I think the PC business has been fairly volatile over the last number of quarters, from the pandemic highs to some of the inventory digestion that we were all dealing with. I can say that I'm pleased to say that I think the growth that we're seeing -- that we saw during the second quarter and that we see in the second-half is the strength of our product portfolio. I think the Ryzen 7000 series is doing well, there's good customer pull. I think from a competitive dynamic standpoint, the business is always competitive, but we feel good about. The most important thing that was a little bit of a drag on operating margins was the revenue being low, as well as some of the -- we had a case where the sell-in was below consumption as we were normalizing inventory levels in the supply chain. As we get past that, what we see is I think the Client business continues to grow. We believe that Client will grow into 2024 as well. In terms of some of the cost initiatives, we have been, let's call it, optimizing sort of the overall R&D footprint, but maybe I'll let Jean comment some more.
Jean Hu: Yes, on the OpEx side, the team has done a great job during this process to really optimize the investment in Client, on the segment to be more efficient and effective. If you look at the overall common company level, our OpEx has been largely flattish. But we are investing in AI, Datacenter, and the strategic priorities we have which generate a much higher return on investment. So, we have optimized it. We feel pretty good about this level of operating expense to continue to invest in Client, the segment. As Lisa mentioned, it's really about revenue. The model we leverage to generate profitability, we should be able to get back to 20%.
Toshiya Hari: Thank you.
Operator: And the next question comes from the line of Harlan Sur with JP Morgan. Please proceed with your question.
Harlan Sur: Yes, good afternoon, and thank you for taking my question. Good to see the quarter-over-quarter inflexion in your EPYC business targeted at enterprise customers. I think you did mention a continued muted environment in enterprise. But the team continues to drive share gains with global corporations, you're ramping Genoa. Are you anticipating your enterprise segment to contribute to the strong second-half growth profile of your Datacenter business?
Lisa Su: Yes, thanks for the question, Harlan. Look, enterprise business is very strategic to us. We feel that we're underrepresented. It's a place that we're putting more resources because, again, when we look at the value proposition of Genoa and entire Zen 4 portfolio, we think it plays very well into the enterprise. So, pleased to see the growth in the second quarter. We do believe that we're on a path to continue to grow into the second-half of the year, and beyond. And the key here is also investments in some of the go-to-market activity, so investing in more business development folks that can call directly on these enterprise customers, together with our OEM partners, and ensure that our value proposition is very well understood.
Harlan Sur: Perfect, thank you. And then on the accelerated compute, general purpose compute demand might muted in China, but there is a significant amount of unmet demand for accelerated compute in this region. And I know there were performance thresholds put in place last year, and maybe U.S. government might lower that performance threshold again soon, I'm not sure. But let's say barring that, has the team looked into developing China-specific SKUs, if they are MI250 or your new MI300 platforms? It seems like the opportunity here is quite large.
Lisa Su: Yes, Harlan, look, China is a very important market for us, certainly across our portfolio, as we think about certainly the accelerator market. Our plan is to of course be fully compliant with U.S. Export controls, but we do believe there's an opportunity to develop product for our customer set in China that is looking for AI solutions, and we'll continue to work in that direction.
Harlan Sur: Thank you, Lisa.
Lisa Su: Thanks, Harlan.
Operator: And the next question comes from the line of Vivek Arya with Bank of America Securities. Please proceed with your questions.
Vivek Arya: Thank you for taking my question. The first one, just a clarification, would it be reasonable to assume that your GPU accelerator sales could be about, say, $500-ish million this year, so about 7%, 8% of datacenter sales? And if that is the right number, does it mean your server CPU sales are effectively flattish year-on-year this year?
Lisa Su: Yes, Vivek, I don't know that I would go into quite that granularity. What we will say is the GPU sales in the first-half of the year were very low as we were sort of in a product transition timing as we go into the second-half of the year. In particular, the fourth quarter, we'll have MI300 ramp. I think your number may be a little bit high in terms of the GPU sales, but overall in general, I think our expectation is that, as Jean said, the datacenter business, given all of the market dynamics, we see it up high single-digits year-on-year, we see much better second-half compared to first-half. And I think the product portfolio and the ramp of Genoa and Bergamo, as well as the ramp of MI300 are key components of the second-half ramp.
Vivek Arya: Thank you, Lisa. And for my follow-up, just kind of a broader question on AI accelerators in the commercial market, so I'm excluding the Supercomputing, the El Capitan projects, et cetera. What is AMD's specific edge in this market? You know there are already strong and established kind of merchant players, there are a number of ASIC options, a number of your traditional competitors, Intel and others, and several startups are also ramping. So my question is, what is AMD's specific niche in this market? What is your value proposition and how sustainable is it, because you're just starting to sample the product now. So, I'm trying to get some realistic sense of how big it can be and what the specific kind of niche and differentiation is for AMD in this market?
Lisa Su: Yes, sure Vivek. So, I think maybe let me take a step back and just talk about sort of our investments in AI. So, our investments in AI are very broad and I know there's a lot of interest around datacenter, but I don't want us to lose track of the investments on the edge as well as in the client. But to your question on what is our value proposition in the datacenter, I think what we have shown is that we have very strong capability with supercomputing, as you've mentioned. And then, as you look at AI, there are many different types of AI. If you look across training and inference, sort of the largest language models and what drives some of the performance in there when we look at MI300, MI300 is actually designed to be a highly flexible family of products that looks across all of these different segments. And in particular, where we've seen a lot of interest is in the sort of large language model inference. So, MI300X has the highest memory bandwidth, has the highest memory capacity. And if you look at that inference workload, it's actually a very, it's very dependent on those things. That being said, we also believe that we have a very strong value proposition and training as well. When you look across those workloads and the investments that we're making, not just today, but going forward with our next generation MI400 series and so on and so forth, we definitely believe that we have a very competitive and capable hardware roadmap. I think the discussion about AMD, frankly, has always been about the software roadmap, and we do see a bit of a change here on the software side. Number one, we've put a tremendous amount of resource on it. So, bringing together our former Xilinx software team, together with the AMD sort of based software team, we've dramatically increased the resources. And also the focus has now been on sort of optimizing at these higher level models. So, if you think about the frameworks around PyTorch and Triton and Onyx, I think many of the new AI centric companies are actually optimizing at a different level, and they're working very closely with us. So, in this place where AI is tremendously exciting, I think there will be multiple winners. And we will be first to say that there are multiple winners. But we think our portfolio is actually fairly unique in the sense that we do have CPUs, GPUs, we have the accelerator technology with Ryzen AI on the PC side as well as in the embedded side with our Xilinx portfolio. So, I think it's a pretty broad and capable portfolio.
Vivek Arya: Thank you, Lisa.
Operator: And the next question comes from the line of Stacy Rasgon with Bernstein Research. Please proceed with your question.
Stacy Rasgon: Hi, guys. Thanks for taking my questions. I wanted to first go back to the Q4 datacenter guide. So, if I do my math right, it's something like $700 million sequentially in datacenter from Q3 to Q4. So, how much of that is MI300 versus CPU? And given the lumpiness of the El Capitan piece, what does that imply for the potential seasonality into Q1 as most of it rolls off?
Lisa Su: Yes, sure. So, it is a large ramp, Stacy, into the fourth quarter. I think the largest piece of that is the MI300 ramp. But there is also a significant component that's just the EPYC processor ramp with, as I said, the Zen 4 portfolio. In terms of the lumpiness of the revenue and where it goes into 2024. Let me give you kind of a few pieces. So, I think there was a question earlier about how much of the MI300 revenue was AI centric versus let's call it supercomputing centric. The larger piece is supercomputing, but it's meaningful revenue contribution from AI. As we go into 2024, our expectation is again, let me go back to the customer interest on MI300X is very high. There are a number of customers that are looking to deploy as quickly as possible. So, we would expect early deployments as we go into the first-half of 2024, and then we would expect more volume in the second-half of '24 as those things fully qualify. So, it is going to be a little bit lumpy as we get through the next few quarters. But our visibility is such that there are multiple customers that are looking to deploy as soon as possible. And we're working very closely with them to do the co-engineering necessary to get them ramped.
Stacy Rasgon: But like, of the $700 million, it's like $400 million of it El Capitan or is it $500 million or $300 million like how big is the El Capitan piece?
Lisa Su: You can assume that the El Capitan is several hundred million.
Stacy Rasgon: Several hundred, okay. For my follow-up, just gross margins coming up in the back in the second. I mean, they still kind of missed in the quarter. I know they rounded up to 50%, but they were 49.7%. I know you're guiding 51 for Q3. Jean, where do you see gross margins sitting like in Q4 as we exit the year?
Jean Hu: Yes, I think the gross margin is, for us, the primary driver as we discussed in the past, it's really mixed. And if you look at our guidance or outlook of Q3, gross margin of 51%, it's more than one percentage point improvement sequentially despite of very significant headwind from embedded business declining in Q3. So, the datacenter and the client business are expect to grow double-digit sequentially and provide a positive impact on the gross margins, which actually more than offset the headwind from embedded business. So, going to Q4, again we're not guiding Q4 and it's going to depend on mix. I would say one thing is you will have a similar dynamics, right. Datacenters expect to grow very significantly. At the same time, we're going to have the same headwind from embedded business declining sequentially. So, overall, we do expect gross margin to improve from this level going forward.
Stacy Rasgon: Okay, thank you.
Operator: And the next question comes from the line of Joe Moore with Morgan Stanley. Please proceed with your question.
Joseph Moore: Great, thank you. You've talked about the embedded business declining as you move into the second-half. Can you give us a sense for how much? And is that decline a function of the common infrastructure market or are you seeing weakness beyond that part of the market?
Lisa Su: Yes, sure, Joe. Thanks for the question. So, look, when I look at the embedded business, I think we should start by remembering that we're coming off of six quarters of very strong growth. I mean, this business has performed extremely well and very pleased with the overall momentum in the business. To your exact question of what we're seeing in the markets, we're actually seeing the core markets hold up pretty well, so let's call it aerospace and defense strong; industrial vision and healthcare, strong; test and emulation strong. We are seeing communications weakness. So, that is the primary driver of the second-half commentary. And there's also some inventory optimization, as you might expect, since our league times have come down over the last several months. So, in terms of zip code, I would say think of it as double-digit down sequentially in the third quarter, and that's the current view that we have. But overall, the business has been extremely strong for us, so I think this is an expected decline as we come off the cycle.
Joseph Moore: Great. And any sense for beyond this quarter, since we've asked you so many Q4 questions already today, but any sense is that kind of the bottom level or do you expect there to be some continued contraction?
Lisa Su: As Jean would say, we're not guiding for the fourth quarter, but I think you should expect embedded sort of in that similar zip code. Yes, that's what I would say.
Joseph Moore: Okay, great. Thank you.
Operator: And the next question comes from the line of Timothy Arcuri with UBS. Please proceed with your question.
Timothy Arcuri: Thanks a lot. Jean, my first question is on inventory. You said it's going to come down a bit as you ramp into the Q4, obviously, you have a big Q4. Can you sort of shape that out for us? Before this, normalized inventory days were kind of 90 to 100 days. Where do you think you're going to exit Q4 in terms of inventory days?
Jean Hu: Yes, Tim, thanks for the question. I think as we ramp those product lines in Q3 and Q4, you will see inventory come down first in Q3 and Q4 again. I think the inventory days of inventory probably will be around 110 to 120 days. The key thing is, right, is if you look at a lot of our product, they are like advanced process technology, five nanometer, four nanometer, six. The manufacturing cycle tend to be long. So, in the longer term, you should expect us from days of inventory be more around 100 to 120 days versus traditionally like 80 days or 75 days. That will be too short for really most advanced process technologies.
Timothy Arcuri: Thanks a lot. And then, my follow-up is for you, Lisa. I mean if you kind of add up the units, the customer interest, you can easily get to several hundred thousand units, it seems to me, for the MI300X next year. So, the question really is on the supply chain, and particularly Cohost, do you think that's going to be a bottleneck for you? I know that they've been expanding capacity. I know you've been trying to procure more there. Can you sort of talk about that and sort of do you think that supply could become a limiting factor you next year? Thanks.
Lisa Su: Yes, absolutely. So, I'm not going to comment on the exact units, but what I will say is that we've been focused on the supply chain for MI300 for quite some time. It is tight. There's no question that it's tight in the industry. However, we have sort of commitments for significant capacity across the entire supply chain. So, co-host is one piece of it, high bandwidth memory is another piece of it and then just the general capacity requirements and look, our goal is to make this a significant growth driver for AMD, I think it's a great market opportunity. We love the engagements with customers, it's our responsibility to provide the supply for the demand and so that's what we've been working on.
Timothy Arcuri: Thanks a lot.
Lisa Su: Thanks, Tim.
Operator: And the next question comes from the line of Christopher Rolland with SIG. Please proceed with your question.
Christopher Rolland: Hey guys, thanks for the question and more on the MI300 opportunity that you guys called out as a multibillion dollar growth opportunity, I was wondering if perhaps you could put a time frame around that multibillion dollar opportunity but more specifically, have you guys ported over MI300 LLMs to MI300? Have you looked at the performance? How do they perform? Are you excited about that? And then, in terms of hyperscale uptake, is it the X version, the GPU only version, that you expect to be the biggest seller here? And have you had any semi-custom kind of configurations here that potentially might even include an FPGA or other kind of Lego movements on the MI300? Thank you.
Lisa Su: Sure. So, there were a lot of aspects to that question, Chris. So, let me try to give you some framework here. I don't think we're ready to talk about timing yet of revenue numbers. What we will say is we do believe it's a multibillion dollar opportunity. I think 2024 is a very important year for us. Ramping MI300 in multiple customers over the next several quarters is very important. I think I mentioned earlier in the Q&A that the customer interest is actually diverse, which is great. It includes sort of what you would expect in terms of the large Tier-1 hyperscalers. But I think these new class of sort of AI focused companies have been working very closely with us, and then some of the large enterprises are also looking at ramping up their efforts. The performance that we see is strong. I think the large language model work that we've done, we've done a lot of it on MI250, and we've seen very good results that's on both training as well as inference. I think as we go through MI300 again, the early results are strong. For AI applications, what we're seeing now is MI300X. So, let's call it the GPU only version is the one that is sort of most prevalent in the AI customer engagement. But the MI300A, actually, which is sort of where we have the CPU and the GPU more closely coupled together is also of interest. So, I think the key is I think we've built a platform that does allow people to kind of choose what is best for the models and for the workloads that they're trying to enable. And that's what we're working on.
Christopher Rolland: Great. And just as a quick follow-up, then, Siena Telco is a market kind of owned by your competitor there. They have a lot of software around Telco. What kind of share do you think you can take in the Telco market from them over the next few years?
Lisa Su: Yes, we're excited about Siena. I think Siena fits again. It's as you said, it's a niche that we haven't previously been focused on. I think our interactions with the Telco suppliers are they're anxious to have Siena be a part of their portfolio. Siena is also one that we'll use for other edge applications, or let's call it lower end applications that need the performance of Zen 4, but perhaps not the heavy platform that we have on the Genoa and Bergamo. So, we do think we're starting from a very low point. So, there's an opportunity to gain share over the next couple of years, and we'll focus on that.
Christopher Rolland: Thanks so much, Lisa.
Lisa Su: Thanks, Chris.
Operator: And the next question comes from the line of Chris Danely with Citi. Please proceed with your question.
Chris Danely: Hey, team. Thanks for squeezing me in. Lisa, so if the MI250, 300, et cetera ramp or the revenue is mostly GPU only, what kind of an impact would that have on AMD gross margin? Would that still be gross margin accretive or dilutive or net neutral to your corporate gross margin?
Lisa Su: Yes, thanks Chris. Let me just make sure I get the statement clear. So, both MI300A and MI300X will be part of the ramp, particularly in the fourth quarter. And as we go into next year for the AI specific applications, we are more heavily weighted towards MI300X, just given sort of where the software is written. And to your question about gross margins at the corporate level, so we would expect that our AI business will be accretive to gross margins at the corporate level. And obviously, as you start the ramp, there's a little bit of learning, but overall we expect it to be accretive to our corporate gross margins.
Chris Danely: Great. And then, from a follow-up I just had -- I guess clarification, so it sounds like most of the MI revenue you have in the hopper right now for at least the committed revenue is LCAP, is that true? And do you have other, I guess, confirmed or hard orders for that, or maybe just spend some time telling us how you're working with the customers or what it takes for them to go from, "Hey, we are interested," to, "Here is the purchase order?"
Lisa Su: Yes. So, may be if your question is do we have other customers who are committed to MI300 other than LCAP, the answer is yes. We have a number of customers who are actually committed. And the way these things go, actually it's not different, not very different than how a server ramp goes, right? I mean one starts with an initial deployment, ensures that the software works, ensures that we have all of the reliability and capability in the datacenter, and then they ramp from that. I will say the difference in AI deployments is, I think, customers are willing to go very quickly. There is sort of a desire and agility because we all want to accelerate the amount of AI compute that's out there. And so, the speed in which customers are engaged and customers are making decisions is actually faster than they would in sort of a normal, sort of regular environment. And that's great. I think that's helping us, as I said earlier, learn, perfect the software, get of all the capabilities in place for a significant ramp next year.
Chris Danely: Great. Thanks, Lisa.
Lisa Su: Thanks.
Mitch Haws: John, we have time for one more question.
Operator: Okay. And our final question comes from the line of Harsh Kumar with Piper Sandler. Please proceed with your question.
Harsh Kumar: Yes, hey, guys. Thanks for letting me ask the question. And Lisa, we are looking forward to an exciting second-half for your company. I had a quick question on the server share. Do you think that there is a theoretical limit to the share that AMD can get? Historically initially we heard 80/20 was a pervading rule, and then you busted through that. Now we are hearing customers say 70/30 is more like it. More importantly, are there anything large vendors for your server business, where you have significantly more than 30% share, let's say, 40% or even 50% share? And I have a follow-up.
Lisa Su: Yes, sure, Harsh. Thanks for the question. Look, in the server business, I think the most important thing for our customers is that we have a strong roadmap, and it's a roadmap that they can count on. And we've been building that sort of working model, that roadmap, and the trust over the past four or five years. So, I don't think there is any theoretical cap on AMD share. I would say, if we look today, there are multiple customers who have us deployed in their datacenters more than 50% share. And from our view, the place where we have perhaps been a bit more underrepresented is in the enterprise. And that's just a matter of sort of the breadth of enterprise customers and the breadth of enterprise software. So, we believe that we have leadership today, and we are very, very focused on ensuring that we continue leadership in the market. And with that, there is an opportunity to continue to gain share in the server market.
Harsh Kumar: Thank you, Lisa. On my second one, can you help us think a little bit about the generative AI spend, let's say, if you can side press some kind of metric, and so, how many dollars of spend today are you seeing from your customers on generative AI for let's say each dollar of regular server CPU spend? Is there a metric that we can think of? Is there a trend today? And where do you think it can be in a couple of years.
Lisa Su: Yes. I think, Harsh, the best way to answer that, and again, we are all sort of -- it's all on crystal ball as to what's going to happen over the next four or five years. There is no question that the demand for generative AI solutions is very high, and there is a lot of compute capacity that needs to put in. The way we size the market is, it can perhaps grow at a rate of, let's call it 50% CAGR plus or minus over the next three or four years. So, that would take us to $150 billion by the time we get to 2027. Now, that's all accelerators in the datacenters, so that includes GPUs, that includes other ASICs and other accelerators. But I think we have an opportunity to address a large portion of that market. So, that makes it very clear priority for us. It's our number one strategic priority, and we will continue to work closely with our customers as they optimize between CPU and GPU or spend.
Harsh Kumar: Thank you, Lisa.
Lisa Su: Thank you.
Mitch Haws: Great. John, that concludes today's call. Thank you to everyone for joining us today.
Operator: Ladies and gentlemen, you may now disconnect your lines at this time. Thank you for your participation.
