AMD's Q3 data center revenue reaches $8.5 billion, nearly doubling, with revenue growth expected to exceed 80% in 2027

Article body EM_StockImg_Start

EM_StockImg_End

On November 3, Eastern Time, AMD released its financial results for the third quarter of fiscal year 2026, delivering the highest quarterly revenue and profit scale in the company's history.

On the earnings call, the company provided fourth-quarter guidance stronger than market expectations and projected that revenue growth in 2027 will exceed 80%, with server CPU revenue expected to achieve double-digit... [truncated]

Dr. Lisa Su, AMD Chair and CEO: Thank you, Matt, and good afternoon, everyone. The third quarter was the largest and most profitable quarter in our history, as we began volume production of our next-generation data center products. Revenue grew 45% year-over-year to $13.4 billion, driven by record EPYC and Instinct sales and strong growth in our embedded business. Earnings per share grew even faster, up 74% to $2.09. We executed very well this quarter, with Helios ramping volume production on schedule, and EPYC continuing to expand its footprint across cloud, enterprise, and ultra-intelligent workloads. Data center revenue accelerated sequentially every quarter this year, and the segment contributed 64% of total revenue in the third quarter, up from just 47% a year ago. We expect this acceleration to continue in the fourth quarter, driven by growing demand for inference and agentic workloads and volume production of our latest products.

Looking at our segments, data center revenue grew 97% year-over-year to a record $8.5 billion. In servers, both cloud and enterprise sales more than doubled year-over-year. We significantly exceeded our prior outlook of "growth of more than 80%" and gained revenue share. Customers are deploying EPYC in more applications, with agentic workloads contributing significant growth. In the cloud, hyperscalers deployed a large number of EPYC servers this quarter to run more internal services and expand their public cloud offerings. Turin is now the most widely deployed EPYC generation in public cloud, with the world's largest providers offering more than 540 instance types. Cloud providers also significantly increased EPYC capacity this quarter for personal and enterprise agents, which are rapidly becoming the next killer app for cloud CPUs.

Meta's Muse and OpenAI's Dots illustrate how fast personal agents are growing, with more than 10 million people activating the Muse agent in the first six weeks. Enterprises are also moving quickly, applying agents to areas such as software development, supply chains, and IT operations. These agents work continuously and in parallel across applications and enterprise systems, increasing both accelerator demand and significantly boosting CPU demand. Adoption is just beginning, and we expect demand to grow sharply as agents are deployed more broadly and take on more complex work. In head nodes, we won major deployments this quarter with several of the largest model developers and cloud providers, which will ramp over the coming quarters. EPYC's leading performance and mature ecosystem have been proven at scale, making it the host CPU of choice for custom accelerators.

For example, OpenAI selected Turin EPYC processors as the host CPU for its first custom accelerator platform. In the enterprise market, EPYC server sell-through set a record, with strong demand from new and existing customers. This quarter we added dozens of first-time EPYC customers, including large financial services, telecommunications, retail, and healthcare companies. Existing customers are also making EPYC their primary compute platform, expanding their applications from specific workloads to a larger share of their overall infrastructure. We began volume shipments of EPYC Venice processors in the third quarter, with the first OEM systems expected to launch this quarter. Early demand is the strongest we have seen for any new EPYC generation. Our OEM and ODM partners have committed to more than 240 Venice platforms, 20% more than Turin, and cloud providers are deploying Venice this quarter, ahead of instances planned for early 2027.

We recently introduced Helion, a Venice reference design that enables turnkey EPYC rack-scale systems for agentic infrastructure. Customers can configure Helion for their workloads, from all-CPU racks for hyperscalers to enterprise systems combining Venice and Instinct GPUs that can run both agents and models. In an all-CPU configuration, a single Helion rack can hold up to 144 Venice processors with nearly 74,000 threads, delivering the industry's highest agent density. All major server OEMs and ODMs plan to offer Helion, with the first systems shipping this quarter.

More broadly, demand for server compute across cloud, enterprise, agentic, and ultra-intelligent infrastructure is accelerating. EPYC is the industry's broadest portfolio, covering all these markets, with per-socket leadership in cloud and agentic, per-core leadership in enterprise and head nodes, and memory and I/O options tuned for every workload. Venice builds on this leadership with significantly better performance per watt, per core, and per dollar than any other x86 or ARM server CPU, positioning us to gain share and significantly expand our server business as the TAM grows to approximately $220 billion by 2030.

Turning to the data center AI business, Frontier Labs, hyperscalers, and cloud providers expanded their Instinct deployments for inference and training, driving record revenue. We began volume shipments of MI455X this quarter, and leading customers including Anthropic, Meta, Microsoft, OpenAI, and Oracle quickly brought their first Helios racks online. Customers are now running their latest models and workloads on Helios, and we are seeing very strong performance in initial deployments. We expect Helios deployments to increase significantly in the fourth quarter as production continues to ramp and HPE, Dell, Lenovo, Supermicro, and other OEM and ODM partners begin shipping systems.

We are very pleased with the overall performance at the rack level. MI455X leads in compute FLOPS, memory capacity, and scale-out bandwidth. With Helios now on the market, its lead over competitors in memory bandwidth and higher scale-up bandwidth is even greater. We are working closely with manufacturing and OEM partners to scale production as customers expand deployments. In supercomputing, EuroHPC selected the Instinct MI430X GPU and sixth-generation EPYC processors for LUMI-AI, Europe's next flagship supercomputer, planned to come online in the second half of 2027. MI430X delivers nearly nine times the double-precision performance of competitors, which is critical for scientific simulation. As a result, MI430X has been selected to power a new generation of systems around the world, and as governments and research centers plan their next major investments, we are well positioned to win more designs.

On the software side, we released ROCm 10, bringing major enhancements for production deployments at scale, including faster startup for large GPU clusters and greater resilience for long-running workloads. ROCm 10 also adds agentic performance tuning through ROCm.AI, which profiles workloads, identifies bottlenecks, applies and validates improvements, compressing weeks of engineering optimization into hours. Customers are also using their own agents to optimize models on our open-source software stack. For example, Anthropic highlighted that one engineer used Claude to bring its latest model online on an Instinct MI355X rack over a single weekend and achieved strong performance. Our collaboration with Frontier Labs already enables their models to handle complex software development and Instinct performance tuning. As these capabilities continue to improve, everyone in the AMD ecosystem will benefit. This quarter we announced two acquisitions to strengthen our ultra-intelligent product portfolio.

Taalas brings an experienced engineering team and purpose-built chips for low-latency, efficient inference. We plan to integrate its technology across our solution portfolio to meet growing demand for low-latency ultra-intelligent applications. World Labs, led by co-founder and CEO Dr. Fei-Fei Li, brings a world-class research team with expertise in spatial intelligence and models that understand the physical world, giving us deeper insight into how frontier models and workloads will evolve. Fei-Fei Li will join AMD as Chief Scientist, leading our model research and helping shape our future hardware, software, and systems roadmap. We expect both acquisitions to close by the end of the year.

Looking ahead, our focus is on execution at scale. Volume production of Venice, the MI450 series, and Helios is still in its early stages, and all three are expected to grow significantly.

We now expect data center segment revenue growth to accelerate in the fourth quarter, with full-year segment revenue more than doubling from 2025. Our visibility into 2027 also continues to strengthen. Last quarter, we said we expected server annual revenue growth of more than 70% and data center segment revenue to more than double for the full year. Since then, customers have expanded their EPYC and Instinct deployment plans, and we have secured additional supply to support that growth. We now expect server revenue to double year over year in 2027, and data center AI revenue to grow faster than we forecast 90 days ago. As a result, we expect data center to account for more than 80% of total company revenue in 2027 and to drive the majority of revenue and earnings growth.

Looking further out, customer communications on MI500 are progressing very well. Market interest is high because it represents another major leap in performance. We are working closely with multiple Frontier labs to optimize MI500 for their planned multi-gigawatt deployments. Turning to the client and gaming businesses, segment revenue declined 8% year over year to $3.7 billion. Client revenue grew 7% year over year to $2.9 billion, driven by strong sales of our Ryzen mobile processors. Commercial is becoming a larger part of the client business as large enterprises adopt Ryzen PRO PCs across their workforces. Eight of the world's 10 largest banks and seven of the 10 largest healthcare and automotive companies now use Ryzen PRO PCs. OEMs continue to expand their AMD-based commercial product lines, from everyday business PCs to high-end workstations.

HP, Lenovo and others launched the first commercial laptops and small form-factor desktops featuring our latest Ryzen AI Max PRO 400 series processors. These systems deliver workstation-class performance and can run super-intelligent models with up to 300 billion parameters locally. Looking at the market, PC sell-through slowed at the end of the quarter as higher memory and component costs weighed on demand. We are managing the business accordingly, aligning sell-in with sell-through and keeping customer and channel inventory healthy. In 2027, we expect PC market unit shipments to decline by a double-digit percentage. In this environment, as we deliver on our Ryzen roadmap and expand our commercial footprint, we expect to outperform the market.

In gaming, revenue declined 40% year over year to $773 million, primarily due to lower semi-custom sales, which is in line with expectations for the current stage of the console cycle. Gaming graphics revenue declined year over year, while Radeon GPU channel sell-through grew sequentially. Beyond gaming, Radeon AI PRO revenue grew significantly quarter over quarter as more developers choose AMD to run and fine-tune models locally. We expect Radeon AI PRO to maintain strong growth, driven by our leading performance per dollar and an open software stack optimized for the latest open-source models. Turning to the embedded segment, revenue grew 29% year over year to $1.1 billion, the fourth consecutive quarter of accelerating growth. Demand was very strong across all end markets, led by strong growth from aerospace and defense, automotive, communications and industrial customers. Data center buildouts also drove higher demand for our embedded CPUs and adaptive SoCs in control plane and optical networking applications.

We remain on track to close more than $18 billion in design wins in 2026, which would be our third consecutive record year. Since our acquisition of Xilinx, adaptive design wins have increased every year, with embedded x86 and semi-custom wins growing even faster. As these design wins ramp over the coming years, we expect the embedded business to deliver strong double-digit annual revenue growth and continue to outperform the market.

In summary, we delivered an outstanding third quarter and expect an even stronger fourth quarter as growth in the data center and embedded segments outpaces the decline in client CPU sales. We enter 2027 with the strongest product portfolio in the company's history and significant customer momentum. We have coordinated across the entire supply chain to secure the wafers, substrates, advanced packaging, memory and other key components needed to support higher growth next year.

Based on current visibility, we now expect revenue growth of more than 80% in 2027, more than double the long-term financial model we presented at our Financial Analyst Day last November. Data center will be the largest contributor, driven by expanding Helios deployments and continued EPYC share gains. We also expect embedded growth to accelerate next year, driven by adoption of our adaptive and x86 solutions. Looking beyond 2027, we are increasing R&D investment in our silicon, software and systems roadmaps and working closely with our largest customers to support their ambitious infrastructure plans over the next three to five years. We are also deepening collaboration and co-investing with our supply chain partners to secure the long-term capacity needed for our growth.

Taken together, these give us strong confidence in delivering outstanding revenue growth over multiple years, with earnings growing even faster. I will now hand the call over to Hu Jin for further remarks on our third quarter results. Hu Jin?

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: Thank you, and good afternoon everyone. I will first review the financial results and then provide our outlook for the fourth quarter of fiscal 2026. We delivered an outstanding third quarter. Revenue grew 45% year over year to a record $13.4 billion, above the high end of guidance, driven by our data center segment, which nearly doubled year over year. Revenue grew 16% sequentially, driven by strong growth in the data center and embedded segments, partially offset by declines in the client and gaming segments. The data center segment accounted for approximately 64% of total revenue, up from 58% last quarter, highlighting the continued shift in our business mix and its growing contribution to AMD's growth. Gross margin was 56%, up 190 basis points year over year, reflecting a richer product mix.

Operating expenses were $3.6 billion, up 29% year over year, as we continued to invest in R&D to expand our data center silicon, systems and software capabilities. Operating income grew 75% year over year to a record $3.9 billion, significantly outpacing revenue growth. Operating margin was 29%, up more than 5 percentage points year over year, highlighting the operating leverage of the business as it scales. In the third quarter, with shipments of our MI450 series GPUs, we began recognizing the accounting impact of warrants issued under previously announced customer agreements. Our GAAP diluted earnings per share was $1.46, including an $1.1 billion non-cash charge in other income and expense, primarily reflecting an increase in our stock price and changes in the estimated probability of satisfying the warrant conditions. Looking ahead, as we and our customers make progress under these agreements, these quarterly non-cash fair value adjustments are expected to increase.

The quarterly amounts will vary based on AMD's stock price and other valuation assumptions. These non-cash adjustments will not affect our non-GAAP results or free cash flow. Our non-GAAP diluted earnings per share was $2.09. Now looking at the reporting segments, starting with Data Center. Data Center segment revenue was a record $8.5 billion, up 97% year over year, driven by record server revenue driven by strong demand for our fifth-generation EPYC processors, and record data center AI revenue highlighted by the launch of the AMD Instinct MI450 series GPUs. Sequentially, Data Center revenue grew 27%, driven primarily by the first MI450 series sales and strong server revenue growth. Data Center segment operating income was $2.9 billion, up 168% year over year, with operating margin rising to 34% from 25% a year ago, demonstrating the strong operating leverage of the Data Center business.

Client and Gaming segment revenue was $3.7 billion, down 8% year over year, as semi-custom and graphics sales declined, partially offset by higher demand for our latest-generation Ryzen mobile processors. Client business revenue was $2.9 billion, up 7% year over year, driven by higher sales of Ryzen mobile processors. Sequentially, Client revenue declined 4% as higher memory costs suppressed PC unit demand, partially offset by a richer product mix of higher-end processors. Gaming revenue was $773 million, down 40% year over year, due to lower semi-custom revenue and lower graphics demand. Sequentially, Gaming revenue was roughly flat. Client and Gaming segment operating income was $573 million, or 15% of revenue, compared with $867 million, or 21%, a year ago.

Embedded segment revenue was $1.1 billion, up 29% year over year and up 14% sequentially, as demand strengthened across multiple end markets, led by growth in the aerospace and defense, automotive, communications and industrial end markets. Embedded segment operating income was $453 million, or 41% of revenue, compared with $283 million, or 33%, a year ago. Now looking at the balance sheet and cash flow. Inventory increased by approximately $3 billion sequentially to $11.5 billion to meet our strong data center demand. Cash from operating activities from continuing operations was $1.2 billion. Free cash flow was $554 million, reflecting increased working capital investments and ongoing capital expenditures to secure the equipment and capacity needed for the rapid growth of our data center business, including dedicated back-end capacity to support data center demand. We expect to continue investing in inventory and capacity in 2027 to support our strong growth.

At quarter end, cash and equivalents and short-term investments were $15.8 billion. Total debt was $7.1 billion, including the $4.75 billion senior notes offering we completed in August. Now looking at our fourth quarter 2026 outlook. We expect revenue of approximately $17 billion, plus or minus 3%. The midpoint of guidance represents approximately 66% year-over-year revenue growth, driven primarily by acceleration in the Data Center segment. Sequentially, we expect revenue to grow approximately 27%, driven by rapid growth of approximately 50% in the Data Center segment, a decline of more than 20% in the Client and Gaming segment, and double-digit growth in the Embedded segment.

In addition, we expect fourth quarter non-GAAP gross margin of approximately 55%, and we expect non-GAAP operating expenses of approximately $3.75 billion. We expect non-GAAP net interest and other expense to be a gain of approximately $4 million. We expect a non-GAAP effective tax rate of 13%, and diluted share count is expected to be approximately 1.67 billion shares, including the estimated impact of completing the acquisitions of Taalas and World Labs in the fourth quarter. We do not expect the fourth quarter share count to be affected by customer warrants. In summary, the third quarter was an outstanding quarter for AMD, with record revenue, significant operating margin expansion and earnings growth, reflecting disciplined execution across the business.

As the Data Center business continues to scale, we are very well positioned to drive improved profitability and long-term shareholder value. Next, I'll turn the call back to Matt for the Q&A session.

Matt Ramsay, Vice President of Financial Strategy and Investor Relations at AMD: Thank you very much.

Operator: Thank you. We will now begin the question-and-answer session. Our first question comes from Tom O'Malley of Barclays. Please go ahead.

Tom O'Malley (Barclays Analyst): Hi, good evening, and thanks for taking my question, and congratulations on the outstanding results. I guess, Lisa, let's start with the point where you gave a bit more information on 2027. It looks like revenue growth of 80%, but the Data Center portion of that is growing much faster, with Data Center accounting for more than 80% of your business next year. Maybe you could talk about, first, whether that is supported by supply, whether it's wafers, packaging, substrates or memory? And then looking at the demand behind it, if you could get more supply, would demand exceed supply? Thank you.

Dr. Lisa Su, Chair and CEO of AMD: Okay, sure, Tom, thanks for your question. We have strong momentum entering the fourth quarter and 2027. We are indeed seeing very strong demand, with demand across the data center business, the CPU business, and the data center AI business. Looking at the rate and pace of growth, we have been planning for this over the past several years. We have definitely spent a great deal of time working with customers to plan their needs in terms of deployments and deployment timelines, and also with supply chain partners to ensure we have wafers, substrates, advanced packaging, memory, and all the other components that come with that.

Demand is indeed greater than supply, so we continue to work with supply chain partners to increase supply, and I believe we have a great opportunity there because we have built strong relationships across the entire ecosystem. But in terms of where we are now, we can say that the entire company is growing more than 80%, especially with the server business roughly doubling in 2026 on an already quite high base, and then the data center AI business, where we continue to maintain strong visibility, see strong demand from customers, and also see good follow-through progress across other aspects of the required infrastructure. So overall, we are very optimistic about data center growth in 2027.

Tom O'Malley (Barclays Analyst): Very helpful. And then on client, as a follow-up, you said fourth quarter is down more than 20%. That is consistent with what we are hearing from others in the market. But looking at 2027, I think the difficult point is that typically you see a seasonal decline in Q1, and then recovery only in Q2 and Q3. So maybe talk about how the duration of this correction looks, why you have been able to slightly outperform so far, and how you see coming out of the trough into next year. Thank you, Lisa.

Dr. Lisa Su, Chair and CEO of AMD: Okay, sure, Tom. The PC market remains an important market for us. I think we have been signaling over the past few earnings calls that we expected the second half of this year to be below seasonal, and that is indeed what happened. Our view is that this is the kind of period when component price increases affect overall system pricing. It is reasonable to expect the consumer market to be more sensitive to that, so we are seeing a slight slowdown in end sales. We are simply managing the business very prudently to ensure that channel and customer inventories remain healthy. Our planning assumption for 2027 is that the market will decline by a double-digit percentage.

We will of course focus on the segments where we can create the most value, such as the enterprise market—where we are still underrepresented—high-end consumer, and the desktop channel. These are very good markets for us. We are planning for a down year in 2027 under the current environment, and we will get through it. The good news is that our supply chain has considerable flexibility between the PC business and the data center business. That really allows us to allocate supply flexibly according to market signals. Overall, as I said earlier, our demand is greater than supply, so we continue to work over the next several quarters to increase supply from all sources.

Operator: Thank you. The next question comes from Vivek Arya of BofA Securities. Please go ahead.

Vivek Arya (BofA Securities Analyst): Thank you for taking my question. Lisa, there is a lot of debate in the market around data center financing, land, power, and the availability of shell buildings. So this is a two-part question. Is AMD confident that its anchor customers have the financing and physical infrastructure to support—I think it is roughly 3 gigawatts of AMD's planned deployments in 2027? Looking further out, many of your competitors are investing in the ecosystem, providing residual value guarantees, and so on, and saying their products have greater interchangeability. Does that create a competitive dynamic for AMD? I would love to hear your view on readiness for 2027 and the competitive landscape in 2028.

Dr. Lisa Su, Chair and CEO of AMD: Okay, Vivek, thank you for your question. As I said earlier, we are seeing very strong demand from leading data center AI customers. I think you can see strong demand for more compute, and MI455 is very well positioned. When we look at Venice, MI455, and Helios, we feel very good about our competitiveness. We have been planning this ramp together with customers, so we have good visibility into 2027, and as you said, land, power, shell buildings, all of that data center infrastructure—we are working closely with customers. So we know which data centers and which sites the GPUs will go to, and so on. Looking ahead to 2028 and beyond, I would put it this way: I think we see very strong growth overall.

I think we have a very strong business and a very strong balance sheet, and from that perspective, our goal is first to direct capital into the business that supports growth, which means investing in R&D, investing in the supply chain and capacity, and co-investing with partners to ensure we have the capabilities we talked about earlier that are necessary for this kind of ramp. Then we make selective investments with leading partners in places where we believe it helps identify data center infrastructure many years—several years—in advance. I think that is the right move to ensure that the right infrastructure is in place together when our new chips come to market.

But we are selective and strategic in how we do this, and the first principle is to invest in our growth.

Vivek Arya (BofA Securities analyst): Got it. As a follow-up for me, Lisa, one more question on the competitive landscape. Many of your Frontier Lab customers have been developing their custom ASICs, and I'm curious how they conceptually make the choice between using ASICs versus AMD products? What determines their allocation? Is it cost per token, energy efficiency, software flexibility, or deployment time? Just wondering about the trade-offs in that decision, and whether there are certain workloads where AMD GPUs work better than the alternatives they have? Thank you.

Dr. Lisa Su, AMD Chair and CEO: Okay, Vivek. I think we feel very good about our positioning on our own performance and on the total cost of ownership (TCO) we deliver through Helios. That is fundamentally the most important factor when we think about deployment scale and future deployments. I think we have made significant investments in rack-scale systems to ensure Helios ramps really smoothly. Q3 was our first revenue shipment of MI455 and Helios racks. I think that is a very important milestone. We have a big step-up in Q4. We will continue to ramp throughout 2027. The key is to continue demonstrating TCO advantages across a variety of workloads. Indeed, from a general-purpose GPU perspective, we have more interchangeability when you consider the various workloads that occur. There is also very strong co-design.

Beyond the MI450 series, we are also developing MI500 and MI600 to ensure our technology is even more closely aligned with the needs of Frontier Labs. I think these are key factors in deployment choices.

Operator: Thank you. The next question comes from Joshua Buchalter of TD Cowen. Please go ahead.

Joshua Buchalter (TD Cowen analyst): Hi everyone, thanks for taking my question and congratulations on the progress. I was wondering if you could give us some insight into how HBM costs affect your model as the Instinct business ramps. I think investors have been asking about your revenue contribution per gigawatt over the past year or so, and your outlook has been stable in the $15 billion to $20 billion per gigawatt range, given the changes happening in HBM pricing. Is that because the early buildouts are fairly project-specific and pricing was agreed well in advance? Looking further out at this business, do you have the ability to pass through these higher input costs to your customers, including your three largest customers? Thank you.

Dr. Lisa Su, AMD Chair and CEO: Okay, Josh. Think of it this way: as we ramp, we have long-term agreements with some suppliers and some customers. But given the current memory market, and not just the memory market—by the way, other component costs are also facing some inflationary pressure because more capital expenditure is being invested across the entire supply chain environment. We are also working with customers to ensure these costs are shared as we go forward. So, from my perspective, going back to the question Vivek Arya asked, it is all about TCO, ensuring that from a cost-per-token perspective we continue to offer a very strong advantage for choosing AMD Instinct and our overall Helios system. Even with some inflationary pressure, we certainly believe we are delivering leading TCO for AI workloads.

Joshua Buchalter (TD Cowen analyst): Thank you, Lisa. And then maybe a question for Jean Hu: it is good to see operating expenses staying modest in Q3 and Q4. This has been a pain point for investors for a while. But I guess given how steep the revenue growth is expected to be in 2027, we are somewhat flying blind because you are roughly more than double the target model. How should we think about operating expense growth in 2027? And while I am asking, do you think you can maintain the target gross margin range, that 55% floor, as the year progresses and Helios continues to ramp? Thank you.

Jean Hu, AMD Executive Vice President, Chief Financial Officer and Treasurer: Okay, Josh, thanks for your question. On operating expenses, as you said, we have been investing aggressively. What you are seeing is that these investments are translating into the acceleration in revenue growth we see in 2026. We are actually quite pleased with 2026; this is our operating leverage. Our revenue is growing much faster than operating expenses, which is driving operating margin expansion. If you look at our Q4 guidance, we are guiding revenue growth much greater than operating expenses. We believe this dynamic will continue into 2027. Lisa talked about it: we expect company-wide growth of more than 80%. We should expect operating expense growth to be lower than revenue growth, less than half of revenue growth. This will drive very significant operating margin expansion in 2027.

Back to gross margin: we are actually very pleased with the current gross margin. In Q3 we are ramping MI450. The data center AI business is doing very well, in fact better than we expected when we entered Q3. For Q4 we are guiding data center revenue up 50% sequentially. This will actually be the quarter in which the data center AI business surpasses the server business. We all know this is a bit dilutive to the company's overall average, but the gross margin dollars are growing very significantly. Looking ahead to 2027, we actually believe the data center business will account for more than 80% of the company's total revenue. The data center business mix will significantly drive the company's gross margin. Think of it this way: data center AI will continue to ramp in 2027, while the server business is expected to double.

There may be fluctuations quarter by quarter, because the mix of these businesses will differ slightly. In addition, the embedded business strengthens in 2027. I think this will also help our gross margin. Overall, when you put the whole business model together, you should expect us to achieve significant revenue growth, significant gross margin expansion, and operating expense growth lower than revenue growth, which will drive very significant operating margin expansion and earnings expansion.

Operator: Thank you. The next question comes from Timothy Arcuri of UBS. Please go ahead.

Timothy Arcuri (UBS Analyst): Thank you very much. Lisa, I was wondering if you could break down a little what data center will look like next year. You did raise server CPU from more than 70% growth to more than 100% growth, but you were not particularly specific on data center AI. Is the right way to think about it that the GPU business will still be larger than the server CPU business in dollars, but not by much? Is that the right way to break it down? And then how should we think about the shape of the year? I know the GPU business is volatile, but I guess it could be fairly flat for the full year. Is that reasonable to think? Thank you.

Dr. Lisa Su, AMD Chair and CEO: Tim, let me maybe break down a few things. First, we believe 2027 will be a year of very strong momentum for both the CPU business and the data center AI business. Some of the metrics we gave, you should treat them as metrics. We said the whole company will grow more than 80%, the data center CPU business will roughly double, and the data center AI business will significantly more than double. From that perspective, you should expect the data center AI business to continue, and as Jean just said, Q4 is the first quarter in which the data center AI business surpasses the CPU business in scale. We expect both businesses to grow steadily throughout the year. Hopefully that gives you a rough sense of the shape of the year.

Timothy Arcuri (UBS Analyst): Thank you very much. Jean, I was wondering if you could help us understand the warrants. How will they layer in this year? Can you give us some guardrails on how to model it through the year and into 2027? Thank you.

Jean Hu, AMD Executive Vice President, Chief Financial Officer and Treasurer: Okay, thank you for your question. I think the simple way to think about it is that there are three parts. The first part is that our reported net revenue, Q3 revenue, Q4 guidance, whenever we talk about revenue, is always net revenue, and it already includes the warrant revenue deduction, which is determined at fair value on the grant date. You do not need to worry about that. Whenever we talk about revenue, it is always included in revenue. The second part, which you may be asking about, is the share count. Think of it this way: the share count will dilute shares only when all vesting conditions and exercise conditions are satisfied by our customer. That means shipments need to be completed, and certain commercial conditions need to be met.

We will provide guidance in 2027 to clearly tell you when we believe the share vesting and exercise conditions will be met. That is something we will provide transparency on as we enter 2027. The last part is that it is a non-cash expense, a quarterly non-cash expense, which is a remeasurement of fair value. This actually does not affect our non-GAAP financial results and free cash flow. So on the share count, please wait for us to provide clear guidance when we get there.

Operator: Thank you. The next question comes from CJ Muse of Cantor Fitzgerald. Please go ahead.

CJ Muse (Cantor Fitzgerald Analyst): Yes, good afternoon, and thank you for taking the question. I guess for the first question, on the server CPU side, you obviously raised the outlook. I am curious how we should think about units versus mix, including higher core counts, and higher average selling prices (ASP) associated with Venice. Is there any way to break that down to help us understand that 100% growth?

Dr. Lisa Su, AMD Chair and CEO: Okay, CJ. On the server CPU side, when we talk about 100% growth, you should imagine that we are indeed shipping significantly more units. Of course, some of those units have higher core counts as we enter Venice. But I would say the revenue growth comes primarily from units and core counts, and less from average selling prices, although there is indeed an ASP uplift as core counts increase.

CJ Muse (Cantor Fitzgerald analyst): Great, thanks. As a follow-up, just to confirm the Q1 seasonality, based on what you said tonight, my interpretation is that we should see Q1 client and gaming down seasonally, and data center roughly flattish above the December level. As part of that, what's the view on embedded?

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: For the embedded business, there is typically seasonality. When we look at the whole of 2027, we do see the embedded business growing significantly and strengthening gradually throughout the year. But typically in Q1, actually the embedded business does have some seasonality.

Operator: Thank you. The next question comes from Stacy Rasgon of Bernstein Research. Please go ahead.

Stacy Rasgon (Bernstein Research analyst): Hi everyone, thanks for taking my question. I wanted to know, I know you said that in the context of that 50% guidance, GPU and CPU cross over in Q4. I was wondering if you could help us understand the magnitude of that crossover. Is it just barely, or is it significantly higher? I'm just trying to size it in my head going into next year.

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: Okay, hi Stacy. Maybe I'll give you some color. Think of it this way: if you look at the MI450 ramp, from Q3 to Q4 is a very significant sequential jump. Compared with servers, servers are growing very well and very strongly, but data center AI ramps past servers in Q4, and that's a very significant jump. I think the server business is of course very strong, but data center AI will be well above the server business in Q4.

Stacy Rasgon (Bernstein Research analyst): In dollar terms?

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: Yes, in dollar terms. Yes.

Stacy Rasgon (Bernstein Research analyst): Okay, great, thanks. As a follow-up, Hu Jin, there was an earlier question about next year's gross margin. I'm not quite sure how to interpret your answer on gross margin next year relative to the Q4 level. Do you think you could give us some color on a good entry rate and a good exit rate for gross margin in 2027?

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: I think that's—

Stacy Rasgon (Bernstein Research analyst): Given the ramps of different products and everything that's going on.

Hu Jin, AMD Executive Vice President, Chief Financial Officer and Treasurer: Yes, Stacy. The reason we can't provide quarterly gross margin is indeed because when you think about 2027, the data center mix is the main driver. When you have the data center business, revenue is more than 80% of the company's total revenue. Then within data center, interestingly, you have the data center AI business continuing to ramp significantly. On the server side, we mentioned, Lisa mentioned, that we will double in 2027. Of course, we also know it's a supply-constrained environment, and Lisa also mentioned we are working with suppliers to secure more supply. So the mix between these two businesses really determines the company's gross margin. Right now, I actually don't have that visibility.

I think we want both businesses to do very well, and we know they will both do very well. Q4 is actually a good baseline to think about, because in Q4 the data center AI business is well above the CPU business. So I think that gives you some sense of that baseline. Then in 2027, different quarters may differ. There could be volatility there.

Operator: Thank you. The next question comes from Joe Moore of Morgan Stanley. Please go ahead.

Joe Moore (Morgan Stanley analyst): Great, thanks. We're hearing a lot of positive feedback about the Helion rack, and I was wondering if you could talk a little bit about that. In the pure CPU version, how do these agentic workloads differ from traditional racks? And then, do you expect GPU blades to be attached in many of these racks?

Dr. Lisa Su, AMD Chair and CEO: Okay, Joe. Yes, there is indeed quite a bit of positive excitement about the Helion rack, and I think the main point is that these agentic racks really want to be optimized for the environment they're in. So if they're in a hyperscaler or a large cloud rack, you'll see them in an all-CPU configuration. It will be quite dense. Venice is the industry's best in terms of per-socket performance, so if you fill it completely with CPUs, it will translate into a very powerful agentic rack. But in the enterprise space, you'll also see opportunities for mixed CPU and GPU, which lets people really optimize for the workload they're in. We contributed to OCP, so it's a fully open rack, and I think that's seen as a positive. From that perspective, we've received very positive feedback from OEMs, ODMs, and customers.

Joe Moore (Morgan Stanley analyst): Okay, thanks. If I can follow up, on the 100% server growth, do you think you're meeting demand at that level, or do you think you'll still be in a supply shortage throughout next year?

Dr. Lisa Su, AMD Chair and CEO: Okay, Joe, to grow that much, we're meeting a lot of demand. When we talk about TAM growth, like roughly 50% CAGR over the next three to four years, we see growth in three areas, right? Traditional cloud and enterprise, agentic CPU and agentic workloads, and leading-edge nodes. We believe that when we give our roughly 100% forecast, we're gaining share in each of these segments. I do believe there's more demand, and we're actively working with supply chain partners to meet additional demand. Obviously, we also have to make sure the kits are all there, DDR5 and all the other components. My view is that we're taking a big step forward on customer demand forecasts.

What excites me is that our demand is broad, and almost every customer is expanding their footprint with AMD, so that's a positive. We'll continue working to increase supply to meet the demand as it ramps.

Matt Ramsay, AMD VP of Financial Strategy and Investor Relations: John, I'd like to ask you to bring on one more caller, thank you. I think we have time for one more question.

Operator: Thank you. Our final question comes from Harsh Kumar of BMO Capital Markets. Please go ahead.

Harsh Kumar (BMO Capital Markets analyst): Yes, hi, Lisa. The question is about Helios. You're one of two companies shipping complete racks. Can I ask what your share ambition is? The reason I ask is you've done extremely well in servers, gaining share from a basically very difficult competitor. The second part of the question is, do you have ambitions to enter the enterprise market through Helios racks and their accompanying solutions? Then I have a follow-up.

Dr. Lisa Su, AMD Chair and CEO: Okay, Harsh. We're very pleased with the progress on Helios. I think it's an important step forward. Rack-level solutions are a major R&D effort. I think the Helios ramp is going very smoothly. We saw a good first quarter in Q3. We're on track to ramp into Q4 and 2027. Our ambition is to be a very important compute supplier to our largest customers, whether Frontier Labs or hyperscalers. From that perspective, for Helios and MI450 volume production

(Source: Cailian Press) Mid-article embedded ad