AMD’s AI Accelerator Business Is on the Rise and Should Take a Piece of the AI Pie

Strategist

Brian Colello

Revenue segments as reported by the company.

Computing and Graphics

Enterprise, Embedded and Semi-Custom

Data Center

Client

Gaming

Embedded

Revenue Segments

Gross Profit, Operating Income, and Net Income as a percentage of Revenue

Gross Margin

Operating Margin

Profit Margin

Margins

An operating expense is an expense a business incurs through its normal business operations. Often abbreviated as OPEX.

SG&A

Amortization of acquisition-related intangibles

Licensing

Operating Expenses

Data Center Operating Income

Client Operating Income

Gaming Operating Income

Embedded Operating Income

All Other Operating Income

Computing and Graphics Operating Income

Enterprise, Embedded and Semi-Custom Operating Income

Operating Income by Segment

Year over year growth in Revenue and Operating Expenses.

Revenue Y/Y

OpEx Y/Y

Growth

Shares outstanding refer to a company's stock currently held by all its shareholders, including share blocks held by institutional investors and restricted shares owned by the company's officers and insiders.

Outstanding Shares

Market Cap

Also known as Price/Sales Ratio. Calculated using Average market cap for the quarter / Annualized quarterly revenue.

Revenue Multiple

Advanced Micro Devices has a wealth of digital semiconductor expertise and is well positioned to prosper from favorable trends in data centers, artificial intelligence, and gaming. We consider AMD to be one of two notable firms in graphics processing units, which are especially well suited for AI. The company may play second fiddle to Nvidia in AI GPUs, but its GPU expertise should become increasingly valuable, and lucrative, in the years ahead. 

AMD’s primary products include processors and GPUs tailored to PCs, game consoles, and servers. In our view, AMD’s PC and server success stems from the rare x86 architecture license that it possesses from Intel, which allows AMD and Intel to build x86 CPUs for Microsoft Windows PCs. We view it as a heavy lift for Windows to rewrite its x86 software to work with other processors, but Apple made this move in recent years to support its internal ARM-based processors. ARM will likely gain share in the PC market, but we still expect x86-based chips from AMD and Intel to retain leadership in the Windows PC market for quite some time.

AMD has benefited from its outsourced manufacturing model, as its tight relationship with industry leader Taiwan Semiconductor enabled AMD to grab a technological lead as its rival, Intel, stumbled with its internal manufacturing roadmap. We anticipate that AMD will continue to gain market share over the next three years as Intel strives to turn it around, but AMD's gains could be longer lasting if Intel were to stumble further.

We think AMD's data center business should boom over the next few years. Its server CPUs should be in high demand, as should its GPUs suited for AI workloads. AMD pegs the total addressable market for AI accelerators, such as GPUs, at $150 billion by 2027. While we foresee Nvidia capturing the bulk of this value over the next several years, we think that all AI vendors and customers will seek alternatives to keep Nvidia's dominance at bay, and AMD might be the best positioned company to emerge as a second source in AI.

Business Strategy & Outlook

We assign AMD a narrow economic moat, based on intangible assets around a variety of chip designs (including those from Xilinx, which we rated as narrow moat prior to the business being acquired in 2022). We think it is more likely than not that AMD will generate excess returns on capital over the next 10 years, and perhaps even longer.

AMD is perhaps best known for its CPUs for PC desktops and notebooks within its Client business segment. We think that AMD has a moat in this business, not only because of chip design expertise honed over decades, but also because AMD is one of two prominent firms to hold an x86 instruction set architecture license. 

Intel initially developed x86 in the 1970s and licensed the architecture to AMD, to satisfy conditions from IBM that required a second source of chips. Intel and AMD are the primary licensees of x86 and owners of x86 intellectual property, as they have a cross-licensing agreement that covers the x86 patent portfolio and would be terminated if either firm was acquired or had a change of control. The exclusivity of the x86 license serves as a significant barrier to entry for other chipmakers (such as Qualcomm, Apple, or others) that cannot make an x86 processor without a license.

For many years, effectively all PC software (including Microsoft’s Windows, Apple’s Mac operating systems, and any desktop software applications running on both operating systems) was designed for the x86 architecture. Software makers could not easily deviate from building x86-optimized applications, while Microsoft and Apple would have a heavy lift to convert their operating systems to alternate instruction sets (such as those by ARM, which is the dominant instruction set for mobile devices, used in processors from Qualcomm, Samsung, Apple’s iPhone and others). In turn, Apple (for the Mac) and PC vendors (for Windows machines) had little choice than to buy x86 processors from (mostly) Intel or (on occasion) AMD to run their PCs.

Today, however, Apple has already done the heavy lifting of converting its Mac software to ARM and has now built excellent in-house processors for its Mac lineup. Further, an increasing portion of software is hosted in the cloud and can run on multiple operating systems and isn’t tied to only x86 devices. Microsoft has dabbled in ARM-based versions of Windows and may increase these investments over time. We still think it would be a heavy lift for any on-device software makers to port their software to be compatible with ARM-based PCs, but we think the writing is on the wall here to move such software to the cloud or on to ARM-based devices like the Mac. Thus, we don’t think the x86 architecture is as moaty as it once was, although we still foresee x86 based processors from Intel and AMD as making up a significant portion of the PC market for the next several years.

In years past, Intel dominated the PC market, not only because of its x86 know-how, but also its internal manufacturing advantage over virtually any other chipmaker, which allowed Intel to come out with best-of-breed products that made it nearly impossible for AMD to gain meaningful market share. These dynamics have flipped in recent years, however, as Intel has stumbled in internal manufacturing. Meanwhile, AMD has partnered closely with Taiwan Semi (TSMC), the world’s largest outsourced chip manufacturer or foundry. TSMC now has a market lead in making cutting edge chips, and AMD is bringing more competitive products to the PC market. Further, AMD has focused on a chiplet strategy with TSMC that is improving yields and flexibility, boosting AMD’s portfolio even further. 

In turn, AMD has gained market share over Intel in PCs in recent years. Intel is investing heavily in new process nodes to strive for manufacturing parity, if not leadership, with TSMC. If Intel is successful, it could regain some lost x86 PC market share. However, we no longer view AMD as an also-ran in PC CPUs, and even if Intel were to improve, we still think AMD could retain enough customers to maintain a healthy PC business.

In server CPUs within AMD’s Data Center segment, we see similar dynamics where most of the software (particularly in the enterprise) is based on x86, and we think AMD has a moat in this segment too. Again, Intel’s manufacturing stumbles and AMD’s partnership with TSMC has enabled AMD to gain market share and profitability, and we don’t think Intel will easily recapture such share even if its internal manufacturing were to improve.

We also think that any cloud-based software vendors face high switching costs to move their software from x86 servers and on to ARM-based servers, although the switching costs are likely far lower between hosting software on an Intel x86 server versus an AMD one. Still, like the PC market, we see ARM-based processors as a threat, as Amazon, Nvidia and others are developing ARM-based processors to run traditional server workloads, as well as AI applications. Optimizing data center architectures and workloads for ARM-based servers, rather than x86 ones, is again a heavy lift, from what we can gather. Still, we recognize that mega-cap tech firms and cloud leaders (Microsoft, Apple, Amazon) may have the incentive to do so if they can use ARM-based processors to elevate their data centers and devices. Ultimately, we don’t foresee a mass exodus away from x86 architectures in the data center for the next several years.

Looking ahead, we think AMD’s graphics processing unit (GPU) expertise is becoming increasingly valuable in AI applications. GPUs perform parallel processing, in contrast to the serial processing performed by CPUs to run the software and applications on PCs, and so on In the past decade, the parallel processing of GPUs was found to more efficiently run the matrix multiplication algorithms needed to power artificial intelligence models. AMD is working diligently to deliver AI-centric GPUs to market. 

However, Nvidia has a clear lead in GPU hardware and has captured the vast majority of AI training workloads today. Further, Nvidia has spent a decade or more developing its CUDA software platform, creating and hosting a variety of libraries, compilers, frameworks, and development tools that allowed AI professionals to build their models. CUDA is proprietary to Nvidia and only runs on Nvidia GPUs, so we think it will take some effort for AI customers to port any AI models based on CUDA over to AMD’s hardware. 

That said, since Nvidia and CUDA are clearly dominant in AI GPUs today, we think that leading cloud vendors will continue to seek second sources for AI and have the necessary incentives to port AI models from Nvidia to other vendors, at least enough to keep Nvidia honest as a vendor. In-house chip development by the hyperscalers will continue, but partnerships with AMD may emerge as well, assuming AMD can improve upon its software chops to allow new customers to use AMD’s GPUs for AI training.

We view AMD’s embedded business as moaty, as this business is mostly inorganic from the acquisition of Xilinx. We long believed that Xilinx warranted a narrow moat, based on its expertise in the field-programmable gate array niche of the broader chip industry (with Intel-Altera being the other major FPGA supplier). Although FPGAs are standard components that can be sold to many different users for a multitude of applications, converting a complex algorithm to an efficient custom hardware architecture can be an arduous task. Customers are generally reluctant to switch FPGA vendors, which would require their engineers to learn new software and design tools. Some of Xilinx's FPGAs are used extensively in communication equipment that may have operational lifetimes of at least a decade if not longer. Thus, it can not only be difficult for new entrants to gain market share but also less likely that existing customers defect from Altera (owned by Intel) to Xilinx or vice versa.

We generally do not view AMD’s discrete GPU business for gaming, nor its semicustom chip business for gaming consoles as moaty, although we acknowledge that the chip design expertise for both likely aided AMD in its ability to deliver profitable GPUs to the data center in the future. In gaming, we think AMD has intangible assets related to GPU patents and chip design expertise. However, it did not lead to excess returns on capital for AMD when its CPU business was lagging, and we believe AMD’s GPU operating margins lag its CPU business.

Narrow

Economic Moat

Advanced Micro Devices

Advanced Micro Devices, Inc. engages in the provision of semiconductor businesses. It operates through the following segments: Computing & Graphics, and Enterprise, Embedded and Semi-Custom. The Computing and Graphics segment includes desktop and notebook processors and chipsets, discrete and integrated graphics processing units, data center and professional GPUs and development services. The Enterprise, Embedded and Semi-Custom segment includes server and embedded processors, semi-custom System-on-Chip products, development services and technology for game consoles. The company was founded by W. J. Sanders III on May 1, 1969 and is headquartered in Santa Clara, CA.

Earnings data Q2 2021

Earnings data Q3 2021

Earnings data Q4 2021

Earnings data Q1 2022

Earnings data Q2 2022

Earnings data Q3 2022

Earnings data Q4 2022

Earnings data Q1 2023

Earnings data Q2 2023

Earnings data Q3 2023

Earnings data Q4 2023

Earnings data Q1 2024

Earnings data Q2 2024

AMD reports better-than-expected Q1 revenue, eyes growth in AI

AMD's Q1 2023 performance reflects strength in data center and embedded segments, while facing challenges in the client segment. The company's focus on expanding its portfolio with Xilinx and pursuing semi-custom or full-blown ASIC-based hyperscale programs presents promising opportunities for growth. AMD's competitive advantages in the AI market, coupled with expectations of growth in the second half, position it for future success.

AMD anticipates revenue growth in Q3 2023 despite Q2 decline

AMD reported a decline in Q2 2023 revenue, but anticipates a sequential and year-over-year revenue increase in Q3 2023. The company's strengths lie in expanding EPYC CPU adoption, strong performance in enterprise and cloud applications, and its position in the AI accelerators market. Challenges include declines in client and gaming segment revenues, as well as supply chain constraints for MI300. AMD sees opportunities in revenue synergies, generative AI solutions, and gaining market share in the Siena Telco market.

AMD reports strong Q3 performance, anticipates growth in Data Center

AMD's Q3 performance showcased strong revenue growth, particularly in the Data Center and Client segments. While facing challenges in the Gaming and Embedded segments, the company is focused on seizing opportunities in the Data Center GPU business and expanding its AI capabilities. Overall, AMD anticipates returning to growth in the second half of the year after inventory correction.

AMD reports strong Q4 2023 performance, anticipates growth in data center and CPU markets

AMD's Q4 2023 performance was marked by strong revenue growth in the data center and client segments, while facing challenges in the gaming and embedded segments. The company is well-positioned to capitalize on opportunities in the data center GPU market and expects growth in the CPU market in 2024.

Revenue

Cost of revenue

Gross profit

Operating expenses

Other

Operating income

EBITDA

Net income

Earnings per share

Total assets

Current assets

Cash & equivalents

Receivables

Inventory

Non-current assets

Net PPE

Intangibles

Total liabilities

Current liabilities

Short term debt

Income tax payable

Non-current liabilities

Long-term debt

Total equity

Stockholders equity

Minority interest

Total liabilities & equity

Total debt

Net debt

Book value per share

Cash from operations

Funds from operations

Change in working capital

Cash from investing

Capital expenditure

Purchase/sale of investments

Purchase/sale of business

Other sources

Cash from financing

Change in stock

Change in debt

Cash dividends paid

Cash from other sources

Net change in cash

Free cash flow

Advanced Micro Devices, Inc. engages in the provision of semiconductor businesses. It operates through the following segments: Data Center, Client, Gaming, and Embedded. The Data Center segment includes server-class CPUs, GPUs, AI accelerators, DPUs, FPGAs, SmartNICs, and Adaptive SoC products. The Client segment refers to the computing platforms, which are a collection of technologies that are designed to work together to provide a more complete computing solution. The Gaming segment is a fundamental component across many products and can be found in APU, GPU, SoC or a combination of a discrete GPU with another product working in tandem. The Embedded segment focuses on the embedded CPUs, GPUs, APUs, FPGAs, and Adaptive SoC products. The company was founded by W. J. Sanders III on May 1, 1969 and is headquartered in Santa Clara, CA.