AMD: MI350 to compete with Blackwell (24Q2 conference call minutes)

AMD (AMD.O) released its second-quarter financial report for 2024 (ending in June 2024) after the US stock market on the morning of July 31, 2024 Beijing time. The key points of the conference call are as follows:

The summary of the AMD 2024 Q2 financial report conference call is as follows. For a detailed interpretation of the financial report, please refer to " AMD: Big Companies Stockpiling, AI Leading the Way Up Again "

I. $AMD(AMD.US) Financial Report Highlights:

II. Detailed Content of AMD Financial Report Conference Call

2.1. Key Points from Executive Statements

1) Operational Highlights:

① Business Segment Performance:

a. Data Center: Revenue increased by 115% year-on-year, reaching a record $2.8 billion. Instinct MI300 GPU shipments surged, and EPYC CPU sales remained strong.

Cloud Deployment: More fourth-generation EPYC CPUs were deployed in large-scale data centers, with a 34% year-on-year increase in cloud instances to over 900. Companies like Netflix and Uber chose fourth-generation EPYC public cloud instances as the primary solution for critical customer workloads.
Enterprise: Significant growth in enterprise sales, winning multiple large contracts. In the first half of this year, over one-third of new enterprise server customers deployed EPYC processors for the first time.
AI Business: Quarterly revenue for MI300 exceeded $1 billion for the first time, with Microsoft expanding the use of MI300X accelerators. Hugging Face adopted new Azure instances, allowing enterprise and AI customers to deploy tens of thousands of models on MI300X GPUs. The acquisition of Silo AI enhanced the ability to serve large enterprise customers, accelerating the optimization of AMD's inference and training solutions. Enhanced support and functionality of the software stack make it easier to deploy high-performance AI solutions on the AMD platform. Collaboration with the open-source community enables customers to implement the latest AI algorithms.

b. Client Segment: Revenue reached $15 billion, a 49% year-on-year growth, mainly driven by strong demand for the previous generation Ryzen processors and initial shipments of the next-generation Zen 5 processors. Zen 5's average instructions per clock increased by 16%, and the Ryzen 9000 series processors will continue to enhance productivity, gaming, and content creation workload performance and energy efficiency c. Gaming Division: Revenue decreased by 59% year-on-year to $648 million. The decline in semi-custom SoC sales was mainly due to the fifth year of the console cycle. Gaming graphics card revenue increased year-on-year, with improved sales of Radeon 6000 and 7000 series GPUs.

d. Embedded Division: Revenue decreased by 41% year-on-year to $861 million. Although second-quarter revenue remained flat , early signs of improved order patterns are emerging, and gradual recovery is expected in the second half of the year.

e. UALink: Collaborating with Broadcom, Cisco, HPE, Intel, Google, Meta, and Microsoft to establish a standardized approach for high-performance data center AI solutions.

② Outlook and Trends:

a. New Products and Technologies:

Turin Series: The next-generation Turin series (Zen 5 core) will be launched later this year, offering up to 192 cores and 384 threads.
MI325X Accelerator: To be launched later this year, further expanding the leading AI performance advantage.
MI350 Series: Planned for launch in 2025, with performance expected to be 35 times higher than CDNA 3.
MI400 Series: Planned for launch in 2026, based on the CDNA Next architecture.

b. Investment and Development: Significantly increasing investments in hardware, software, and solutions, focusing on launching leading data center GPU hardware annually and integrating industry-leading AI capabilities across the entire product portfolio.

2) Financial Highlights:

① Business Performance:

a. Data Center: Revenue of $2.8 billion, a year-on-year increase of 115% and a quarter-on-quarter increase of 21%. Mainly driven by significant growth in AMD Instinct GPU and strong growth in EPYC server revenue.

b. Client Division: Revenue of $1.5 billion, a year-on-year increase of 49% and a quarter-on-quarter increase of 9%. Mainly driven by strong sales of AMD Ryzen processors.

c. Gaming Division: Revenue of $648 million, a year-on-year decrease of 59% and a quarter-on-quarter decrease of 30%. Mainly due to semi-custom inventory digestion and declining end-market demand.

d. Embedded Division: Revenue of $861 million, a year-on-year decrease of 41% and a quarter-on-quarter increase of 2%. Mainly due to continued customer inventory adjustments.

② Shareholder Returns and Debt: Returned $352 million to shareholders, repurchased 2.3 million shares, with remaining authorized amount of $5.2 billion. Utilized existing cash to repay $750 million of maturing debt.

③ Third Quarter Guidance:

a. Revenue Guidance: Expected revenue of approximately $6.7 billion, with a variance of $300 million. Expected year-on-year growth of about 16% and quarter-on-quarter growth of about 15%, mainly driven by strong growth in the data center and client divisions. Expected increase in embedded division revenue, while gaming division revenue is expected to decline by a double-digit percentage.

b. Gross Margin and Expense Guidance: Expected non-GAAP gross margin of about 53.5%. Expected non-GAAP operating expenses of about $1.9 billion.

c. Tax Rate and Share Count Guidance: Expected non-GAAP effective tax rate of 13%. Expected diluted share count of approximately 1.64 billion shares

d. Acquisition: It is expected to complete the acquisition of Silo AI for approximately $665 million in cash in the third quarter.

2.2 Analyst Q&A

Q: I would like to understand the development trend of MI300 for the remaining time of this year. We still have $2.8 billion to achieve the annual target. How do you expect the growth in the fourth quarter?

A: We are very satisfied with the progress of MI300. At the beginning of the year, our main goal was to deploy the product to customer data centers, help them validate workloads, and accelerate applications in production environments, especially focusing on performance. Now, halfway through the year, we have made comprehensive progress.

Looking ahead to the second half of the year, we expect the revenue of MI300 to continue to grow in the third and fourth quarters. We are not only expanding deployments in existing customers but also actively collaborating with a large number of new customers to help them familiarize themselves with our architecture and software. Overall, we are very pleased with the current progress and are advancing product capabilities as expected.

Q: Regarding next year, do you foresee the possibility of rapid growth? Considering some discussions in the market, have you seen signs of significant growth next year based on our product roadmap?

A: Next year, we are confident in our product roadmap. In the expansion plan announced at Computex, we plan to launch MI325 later this year and introduce the MI350 series competing with Blackwell solutions next year; at the same time, our CDNA Next is also steadily progressing. We are very optimistic about the artificial intelligence market, as the demand for computing power continues to grow in the market, our hardware and software solutions are gaining market recognition, and we are constantly expanding our product line.

Q: Regarding the data center business, I would like to know your growth expectations for EPYC server CPUs for the full year. Also, can you share some latest views on the ability to continue to increase market share in the server market? Please briefly describe the expected trend of the server market in the coming quarters.

A: The fourth-generation EPYC, especially Bergamo, has been widely recognized in the market, especially excelling in cloud computing and enterprise applications. We have not only attracted new customers but also deepened cooperation with existing customers, and the adoption by third-party cloud services is increasing.

Looking ahead to the second half of the year, market trends are positive, with enterprise and cloud service spending rebounding, providing us with positive signals. Furthermore, we have started production of Turin and plan to fully launch it in the second half of the year, which is expected to bring us new sources of revenue. Overall, we are optimistic about the outlook for the server market and confident in continuing to expand market share.

Q: Regarding the future plans for data center GPUs, you mentioned that the 325 model will be released later this year. I would like to know if the expected revenue of over $4.5 billion has already taken into account the contribution of the 325 model? Also, could you provide a detailed introduction to the 350 model? We noticed that competitors' products are transitioning to rack-scale systems, will the 350 model have a similar design? Will it be equipped with liquid cooling technology and support rack-scale architecture?

A: We are on track to launch the 325X, with expectations of starting to contribute revenue on a small scale later this year, mainly as an extension of the MI300 series. The 325 model will kick off in the fourth quarter and ramp up quickly in early next year.

For the 350 series, we expect to introduce multiple SKUs covering solutions from air cooling to liquid cooling to meet different customer needs. We are enhancing system-level integration to provide rack-level solutions. At the same time, the 350 series is also compatible with the existing infrastructure of the MI300 series, providing convenient upgrades for customers already using our products. This is part of our roadmap expansion plan aimed at offering more choices.

Q: Can you share AMD's positioning in the AI PC field? With the rise of ARM architecture systems, are we facing a change in the competitive landscape? Please provide insights on how this trend will develop and its impact on the seasonal sales in the second half of the year.

A: We are very pleased with the results in the client business, especially with our strong product roadmap. In mid-year, we will launch laptops and desktops based on Zen 5, with very positive market feedback. For example, the laptop based on Strix that we launched last weekend received very positive reviews. We believe AI PC is an important new area in the PC category. As we enter the second half of the year, we expect better seasonal performance, especially benefiting from our strong product releases. Looking ahead to 2025, AI PC will cover a wider price range, bringing us broader market opportunities.

Overall, the PC market is huge, and although our share is still small, we are confident in our competitive position. The PC market is a significant revenue growth driver for us. With strong product competitiveness and good business operations, we are closely collaborating with ecosystem partners and OEM partners to ensure a series of strong product launches in the second half of the year.

Q: Looking back five or six years ago when EPYC's Naples series was launched, the market response was diverse, but most importantly, it laid the foundation for our long-term cooperation with customers. Now, does the MI300 series play a similar role? Can you share the level of customer involvement, intensity of participation, and the breadth of customer base for the 350 and 400 series?

A: We are very satisfied with the progress of the Instinct roadmap, which is clearly a long-term strategic focus. It has many similarities with the development trajectory of EPYC, and with each generation of products, we have the opportunity to expand more workloads and broader deployments. We have already achieved large-scale hardware deployments in multiple hyperscale and large tier-two customers and validated our software in various environments. ROCm has shown significant improvements in functionality and performance, and we are confident in this.

We continue to invest in system-level integration and networking, engaging in in-depth and long-term dialogues with customers on multi-generational products. While the MI300 has made good progress, there is still much work to be done. We believe that the future roadmap will help us further seize opportunities in the coming years.

Q: There is intense industry discussion about the profitability and investment return of AI. Customers currently have three options: purchasing GPUs from our main competitors, custom chips, or choosing AMD. Faced with such profitability considerations, how will customers choose next year? Will they concentrate more capital expenditure on the other two options? What are your expectations for next year in this industry debate? How will AMD find its market positioning?

A: AI investment is an inevitable choice for industry development, as the potential impact of AI on enterprise operations is significant, so investment trends are expected to remain strong. Among the many choices, I believe the market will accommodate multiple solutions, including GPUs, custom chips, and ASICs. AMD, with its strong product roadmap and close collaboration with customers, has demonstrated our strength, especially in hardware and software co-optimization. For large language models, GPUs remain the preferred choice. Therefore, our technical strength and key partnerships will further solidify our position in the market.

Q: I would like to understand the application of MI300 in training and inference. Although the initial focus was on inference, has there been any breakthrough in the training field? What are your expectations for the future changes in the proportion of these two?

A: MI300 leads the industry in memory bandwidth and capacity, which makes it perform well in inference applications. Although the initial focus was on inference, our customers are actively adopting it for training as well. We have optimized the ROCm software, simplifying the training process on the AMD platform. It is expected that over time, the use of training will gradually increase. While the market may lean more towards inference, for AMD, both inference and training are growth areas for us.

Q: I would like to know if the current shipment status of MI300 meets demand, or if the forecast of $4.5 billion for the year is constrained by supply, especially considering the HBM and CoWoS technologies mentioned in the previous quarter.

A: In terms of supply, we made significant progress in the second quarter, with supply exceeding $1 billion. The team performed exceptionally well, and we expect supply to continue to increase in the second half of the year. Although the overall supply chain remains tight and is expected to continue until 2025, we are confident in driving growth through close collaboration with supply chain partners and enhanced capacity. We will continue to optimize supply-demand balance, accelerate customer adoption, and proactively respond to market changes in the second half of the year.

Q: Regarding the profitability of MI300, you previously mentioned that the business would appreciate as issues are resolved. In the face of current competition and investment needs, has this expectation changed? Do you still have confidence in the growth of the business profit margin?

A: Our team has achieved great success with the MI300 product, despite its complexity. We have successfully increased production and have begun implementing operational optimization measures to improve gross margin. We expect that over time, the gross margin will exceed the company's average level.

In terms of profitability, AMD continues to invest in data platforms, including servers and data center GPUs, which are bringing significant revenue growth and expanding operating profit margins. We are confident in the market potential and will continue to invest. Currently, this is a profitable business.

Q: I would like to understand the details of the performance guidance for the third quarter. The decline in the gaming business requires revenue growth of approximately $1 billion in the data center, client, and embedded businesses. Can you disclose the revenue distribution ratio of these three business segments? How should we understand the distribution of this revenue?

A: The gaming business did experience a double-digit decline. In terms of revenue growth, the data center contributed the most, followed by the client business. The embedded business showed single-digit growth.

Q: Is the growth of the data center business mainly driven by Instinct or EPYC? If there is an increase of about $400 million to $600 million in the data center business, how is this growth allocated between the two?

A: Both Instinct GPUs and server businesses will experience growth.

Q: I understand that the difference in adoption rates between Instinct products and competitors mainly lies in rack-level performance, and there may be a lack of corresponding rack-level architecture. You mentioned UALink, could you further explain how it helps narrow this gap? Is this an important step in the industry?

A: Systems are becoming more complex, especially in large training clusters, where customers need our help to assemble these systems. This includes solutions like Infinity Fabric and rack-level system integration, which are the basis of UALink. We are satisfied with the lineup of partners for UALink, which we consider to be an important capability. AMD has all the necessary components, including Infinity Fabric and networking capabilities acquired through the Pensando acquisition. We will continue to invest in this area to help customers enter the market quickly, providing comprehensive support including CPU, GPU, network, and system-level solutions.

Q: I would like to understand the specific composition of MI300 customers. There were government customers last year, and I would like to know if there are still government customers currently. Additionally, given the investment in software assets, what are the main challenges faced in accelerating the adoption of new customers? While hardware issues such as memory challenges have been discussed, software investment is also a major challenge. What do you think is the biggest obstacle to acquiring new customers?

A: Regarding supercomputing, the focus was mainly on the fourth quarter of last year and the first quarter of this year. The revenue in the second quarter almost entirely came from AI, especially MI300X, serving large AI projects, hyperscale customers, and OEM customers targeting enterprises and secondary data centers In terms of memory, we have multiple suppliers of HBM3 and are certifying HBM3E for future products with multiple suppliers participating. We have good control over memory.

Our ROCm capability has significantly improved due to widespread customer usage. Especially companies using advanced frameworks like PyTorch can achieve fast plug-and-play performance, which helps us expand our product portfolio.

We will continue to invest in software, which is also the reason we acquired Silo AI. This acquisition brought in 300 experienced scientists and engineers who are skilled at helping customers quickly start and run AMD hardware. We see this as an opportunity to expand our customer base. Despite seeing leverage effects, we will continue to invest because the opportunity is huge, and we already have all the necessary components. Our goal is to scale up.

Q: You mentioned that client business in the second half of the year is expected to be higher than the seasonal average. Is this growth mainly due to ASP increases from new products, AI PC sales contributions, or are there other factors driving this expected growth?

A: We expect growth in the third quarter to exceed the usual seasonal levels, mainly because we will be launching the Zen 5 series desktops and laptops, with production volumes significantly increasing. The launch of AI PCs is also part of the growth, but more importantly, it is the overall refresh of our product line. Typically, the release of desktops in the third quarter is a good timing for us, and we are very confident in the market positioning of our products.

Q: You mentioned a slight decrease in incremental gross margin in the third quarter, can you explain the reasons? Additionally, has there been any adjustment to the gross margin expectations for MI300, and has the profit growth point been delayed?

A: This year, we have seen a significant increase in our gross margin, with the third quarter expected to reach 53.5%, thanks to the strong growth in the data center business, which now accounts for nearly 50% of our revenue. In the second half of the year, the data center business will continue to drive our revenue and gross margin growth. Nevertheless, the PC business will also perform better in the second half, especially with the seasonal growth in the consumer market. The embedded business will also grow quarter by quarter, albeit at a more gradual pace. These factors collectively impact the growth rate of the gross margin.

For MI300, we remain confident in its long-term positive contribution to the company's gross margin, with the data center business continuing to be the main driver of gross margin growth.

End of the document

Risk Disclosure and Statement of this article: Dolphin Research Disclaimer and General Disclosure