Data center business explosive growth drives AMD results comprehensively above expectations, with the company raising next year's outlook, expecting server revenue to double and AI chip revenue to grow even faster.
AMD reported after the U.S. market close on Tuesday that its third-quarter 2026 revenue rose 45% year-over-year to $13.377 billion, setting another quarterly record and exceeding analyst expectations by approximately 2.3%. On a non-GAAP basis, earnings per share rose 74% year-over-year to $2.09, beating analyst estimates by more than 6%.
In the subsequent earnings call, AMD CEO Lisa Su stated that data center demand is accelerating and that data center revenue in 2027 is expected to exceed previous forecasts, with server revenue set to double by then, accounting for up to 80% of total revenue.
She said the company has "strong confidence in its ability to sustain exceptional revenue growth over the coming years, with earnings growth outpacing revenue growth." In response to analyst questions, Su acknowledged that demand currently exceeds supply, and the company is working closely with wafer, substrate, advanced packaging, memory and other key component suppliers to expand capacity, expressing "confidence" that it can further scale supply through deep partnerships across the ecosystem.
On the product strategy front, AMD is simultaneously advancing upgrades across both CPU and AI accelerator product lines. Su revealed that the EPYC Venice processor began shipping in the third quarter, with the first OEM systems expected to launch officially in the fourth quarter.
On the AI accelerator side, AMD has begun volume shipments of the MI455X, with leading customers including Anthropic, Meta, Microsoft, OpenAI and Oracle completing initial deployments of the first Helios racks in the third quarter. Su said customers are running their latest models on Helios with satisfactory overall performance, and Helios deployment scale will increase substantially in the fourth quarter.
Additionally, Su explicitly warned that the PC market is expected to see a double-digit percentage decline in 2027.
Citing Emarketer analyst Jacob Bourne, Bloomberg reported that the results "will encourage investors, who have recently been scrutinizing AI spending and its returns more rigorously," especially amid sustained corrections in tech stocks and the semiconductor sector.
Record Q3 Results, 2027 Total Revenue Expected to Surge Over 80%
AMD delivered what could be called a flawless report card in the third quarter of this year. Quarterly total revenue reached $13.4 billion, up 45% year-over-year; non-GAAP earnings per share (EPS) growth was even more striking, surging 74% year-over-year to $2.09. Gross margin also expanded to 56%, up 190 basis points year-over-year.
What truly ignited market sentiment was the forward guidance from management. AMD expects fourth-quarter revenue of approximately $17 billion (up about 66% year-over-year). More importantly, the company gave an extremely optimistic outlook for 2027.
Su said during the call:
Based on current visibility, we now expect revenue growth of more than 80% in 2027, which is more than double the long-term financial model we outlined at our Financial Analyst Day last November.
She explicitly stated that the data center business will account for more than 80% of the company's total revenue in 2027, becoming the absolute primary driver of revenue and earnings growth. Additionally, server revenue is expected to double year-over-year in 2027.
Addressing the supply-demand question that the market cares most about, Su said bluntly:
Demand is indeed higher than supply, so we continue to work with our supply chain partners to increase supply. Given the strong relationships we have across the ecosystem, I believe we have a great opportunity to do so.
Historic Crossover: Q4 AI Chip Revenue to Surpass Server CPU
The data center business was AMD's core engine this quarter, with Q3 revenue surging 97% year-over-year to a record $8.5 billion, rising from 47% of total revenue a year ago to 64%.
Behind this growth is the full-scale rollout of next-generation products. The MI455X began volume shipments in the third quarter, with leading customers including Anthropic, Meta, Microsoft, OpenAI and Oracle quickly bringing their first Helios racks online.
For the upcoming fourth quarter, CFO Jean Hu revealed a key milestone figure:
Data center business will surge approximately 50% quarter-over-quarter in the fourth quarter. In fact, this will be the first quarter in which data center AI business (GPUs, etc.) revenue surpasses server business (CPU) revenue. Compared to the server side, the data center AI business will see a very significant step-function increase in Q4.
To further ramp up the AI arms race, AMD announced two key acquisitions in the third quarter (Taalas and World Labs). Among them, World Labs co-founder and CEO Fei-Fei Li will join AMD as Chief Scientist, leading model research and helping shape the future software and hardware system roadmap.
"AI Agents Are Becoming the Next Killer App for Cloud CPUs"
During the lengthy earnings call, Su repeatedly mentioned a core term — Agentic Workloads — and views it as a core force reshaping computing infrastructure.
Sharing industry insights, Su noted:
Cloud providers added substantial EPYC capacity this quarter for personal and enterprise agents, which are rapidly becoming the next killer app for cloud CPUs. These agents work continuously and in parallel across applications and enterprise systems, creating a sharp increase in CPU demand while also increasing accelerator demand. Adoption is just beginning, and we expect demand to grow dramatically as agents are deployed more broadly and take on more complex work.
Leveraging advantages in performance and total cost of ownership (TCO), EPYC processors are rapidly gaining market share. Su specifically mentioned:
OpenAI selected the Turin EPYC processor as the host CPU for its first custom accelerator platform.
Additionally, AMD's recently launched Helion received unprecedented early demand, with OEM and ODM partners committing to supply 20% more Venice platforms than the previous generation.
Hidden Concerns and Countermeasures: PC Market Faces Double-Digit Decline Next Year
While the AI business surges ahead, AMD's client and gaming businesses face macro headwinds.
In the third quarter, client revenue rose 7% year-over-year to $290 million but declined 4% quarter-over-quarter. Su gave a blunt warning about the PC market:
PC end sales slowed at the end of the quarter as higher memory and component costs pressured demand. Our planning assumption for 2027 is that PC market unit shipments will decline by a double-digit percentage.
Facing a PC market winter, AMD chose to avoid price wars in the low-end market. Su stated:
We will focus on segments where we can add the most value, such as enterprise, premium consumer and desktop channels. We recognize there is significant volatility in the low-end market, and frankly, that is also where we have the least exposure.
However, PC market weakness has not affected the company's overall profit expansion logic. CFO Jean Hu emphasized that with revenue surging over 80% in 2027, the company's operating expense growth will be less than half of revenue growth, which will bring "very significant operating margin expansion and profitability improvement."
AMD Q3 Earnings Call Full Transcript (AI-Assisted Translation):
Operator: Welcome to the AMD Third Quarter 2026 Earnings Conference Call. All participants are currently in listen-only mode, and a question-and-answer session will follow the formal presentation. Please note that this conference is being recorded.
I will now turn the call over to Matt Ramsey, Vice President of Financial Strategy and Investor Relations. Matt, please go ahead.
Matt Ramsay, Corporate Vice President, Financial Strategy and Investor Relations:
Thank you, operator, and welcome to AMD's Third Quarter 2026 Financial Results Conference Call.
I trust everyone has had the opportunity to review our earnings press release and accompanying slides. If you have not, the materials are available on the Investor Relations page of AMD.com.
During today's call, we will primarily refer to non-GAAP financial measures. A complete reconciliation of non-GAAP to GAAP measures is included in today's press release and the slides posted on our website.
Joining us today are Dr. Lisa Su, Chairman and Chief Executive Officer, and Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer.
This is a live broadcast and will be replayed via webcast on our website. Before we begin, please allow me to note that AMD will participate in the following financial community events this quarter: the U.S. Technology and AI Conference on Tuesday, December 1, and the Barclays 24th Annual Global Technology Conference on Wednesday, December 9.
The forward-looking statements in today's discussion are based on our current beliefs, assumptions and expectations, including forward-looking statements about financial projections, business and industry trends, represent our views only as of today, and involve risks and uncertainties that could cause actual results to differ materially from current expectations.
Please refer to the disclaimer in the press release for a description of factors that could cause actual results to differ materially.
I will now turn the call over to Lisa.
Lisa Su, Chairman and Chief Executive Officer:
Thank you, Matt, and greetings to everyone listening today. The third quarter was our largest and most profitable quarter ever, driven by our beginning to ramp production of next-generation data center products. Revenue grew 45% year-over-year to a record $13.4 billion, driven primarily by record EPYC and Instinct sales as well as strong growth in our embedded business. EPS grew even faster, up 74% to $2.09.
We executed well this quarter: Helios production ramp is on track, and EPYC adoption continues to expand across cloud, enterprise and superintelligence workloads. Data center revenue growth has 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 primarily by increasing demand for inference and agentic workloads as well as accelerating ramp of our latest products.
Data Center Business
Data center revenue grew 97% year-over-year to a record $8.5 billion. In the server segment, both cloud and enterprise sales more than doubled year-over-year. We significantly exceeded our previous "over 80% growth" guidance, achieving notable revenue share gains.
Customers are deploying EPYC in more use cases, with agentic workloads contributing significant growth. In the cloud, hyperscalers deployed substantial 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 more than 500 instance types from all major providers available globally. Cloud providers also added substantial EPYC capacity this quarter to support personal and enterprise agents — which is rapidly becoming the next "killer app" for cloud CPUs.
Meta's Muse and OpenAI's Dots demonstrate the rapid growth momentum of personal agents — the Muse agent surpassed 10 million activated users within the first six weeks of launch. Enterprise customers are also moving quickly, actively deploying agents in areas such as software development, supply chain and IT operations. These agents run continuously and in parallel across applications and enterprise systems, driving both accelerator demand growth and significantly boosting CPU demand. Deployments are still in their early stages, and we expect demand to grow substantially as agents are deployed more broadly and take on more complex tasks.
In the head node segment, we won several important deployment contracts this quarter from top model developers and cloud service providers, which will land over the coming quarters. EPYC's leading performance and mature ecosystem have been thoroughly validated in large-scale production environments, making it the preferred host CPU for custom accelerator platforms. For example, OpenAI selected the Turin EPYC processor as the host CPU for its first custom accelerator platform.
In the enterprise market, EPYC server channel sales hit a record high, with strong demand from both new and existing customers. This quarter, we added dozens of first-time EPYC customers spanning large financial services, telecommunications, retail and healthcare enterprises. Existing customers are also establishing EPYC as their primary computing platform, expanding from workload-specific deployments to covering a larger proportion of their overall infrastructure.
In the third quarter, we began shipping EPYC Venice processors to customers, with the first OEM systems also launching officially this quarter. Based on initial demand, Venice has seen the strongest response of any new EPYC generation to date. OEM and ODM partners have confirmed more than 240 Venice platforms, 20% more than Turin; cloud providers also began deploying Venice this quarter, with instances planned for official launch in early 2027.
We recently launched Helion — a Venice-based reference design that provides turnkey EPYC rack-level systems for agentic infrastructure. Customers can customize Helion for their specific workloads: from all-CPU racks for hyperscalers to enterprise-grade hybrid Venice and Instinct GPU systems — the latter capable of running both agents and models simultaneously. In an all-CPU configuration, a single Helion rack can house up to 144 Venice processors with nearly 74,000 threads, achieving the industry's highest agentic compute density. All major server OEMs and ODMs plan to offer Helion products, with the first systems shipping this quarter.
Looking at the broader market, server compute demand is accelerating across cloud, enterprise, agentic and superintelligence infrastructure. EPYC addresses all of these markets with the industry's broadest portfolio — leading single-socket performance for cloud and agentic, leading single-core performance for enterprise and head nodes, and memory and IO configuration options optimized for different workloads.
EPYC further leads all other x86 and ARM server CPUs with significantly better performance-per-watt, performance-per-core and performance-per-dollar. With TAM (total addressable market) projected to grow to approximately $220 billion by 2030, this positions us to continue gaining share and significantly growing our server business.
Data Center AI Business
Frontier labs, hyperscalers and cloud service providers expanded Instinct deployments across inference and training, driving record revenue.
This quarter we began volume shipments of the MI455X, with core customers including Anthropic, Meta, Microsoft, OpenAI and Oracle quickly completing initial deployment of their first Helios racks. Customers are now running their latest models and workloads on Helios, with initial deployments performing excellently.
As capacity continues to ramp and OEM partners including HPE, Dell, Lenovo and Supermicro begin shipping systems, we expect Helios deployments to increase substantially in the fourth quarter. The MI455X leads the industry in compute performance, memory capacity and scale-out bandwidth, and with Helios launching, its advantages in memory bandwidth and scale-up bandwidth relative to competitors will further widen.
We are working closely with manufacturing and OEM partners to expand capacity in step with customer deployment scaling.
In supercomputing, EuroHPC selected Instinct MI430X GPUs and 6th Gen EPYC processors to power LUMI-AI — Europe's next-generation flagship supercomputer planned to come online in the second half of 2027. The MI430X's double-precision compute performance is nearly nine times that of competitors, which is critical for scientific simulation. The MI430X has already been selected for multiple new supercomputing systems worldwide, and we are well-positioned to win more orders as governments and research institutions plan new major investments.
On the software side, we released ROCm 10, bringing major upgrades for large-scale production deployments, including accelerated startup of large GPU clusters and higher resilience for long-running workloads. ROCm 10 also adds ROCm.ai agentic performance tuning capabilities that can analyze workloads, identify bottlenecks and automatically apply and validate optimizations, compressing engineering optimization that previously took weeks into hours. Customers are also using their own AI agents to optimize models on our open software stack. For example, Anthropic shared a case where an engineer used Claude to complete deployment of its latest model on an Instinct MI355X rack over a single weekend with excellent performance.
Our collaboration with frontier labs has enabled their models to assist with complex software development and performance tuning for Instinct, and as these capabilities continue to improve, the benefits will extend to all participants across the AMD ecosystem.
This quarter, we announced two acquisitions to strengthen our superintelligence product portfolio. Taalas has an experienced engineering team and specialized chips designed for low-latency, high-efficiency inference; we plan to integrate its technology into our solution portfolio to meet growing demand for low-latency superintelligence applications. World Labs, led by co-founder and CEO Dr. Fei-Fei Li, has a world-class research team focused on spatial intelligence and models that build understanding of the physical world, which will help us gain deeper insight into the evolution of frontier models and workloads. Fei-Fei Li will join AMD as Chief Scientist, leading our model research efforts and helping shape the future hardware, software and systems roadmap. Both acquisitions are expected to close by year-end.
Outlook and Forward Guidance
Looking ahead, our core task is execution at scale. Venice, the MI450 series and Helios are all in early stages of production ramp, and all three product lines are expected to continue growing significantly.
Currently, we expect data center revenue growth to accelerate further in the fourth quarter, with full-year segment revenue more than doubling compared to 2025. Our confidence in the 2027 outlook continues to strengthen.
Last quarter, we indicated we expected annual server revenue growth of more than 70% and data center full-year revenue to double. Since then, customers have expanded their EPYC and Instinct deployment plans, and we have further secured additional supply resources to support growth. We now expect server revenue to double year-over-year in 2027, and data center AI revenue growth to exceed expectations from 90 days ago. As a result, we expect the data center business to account for more than 80% of total revenue in 2027 and to become the primary driver of revenue and earnings growth.
Looking further out, MI500 customer engagement continues to progress well with extremely strong market interest, as it represents another major leap in performance. We are working deeply with multiple frontier labs to optimize MI500 for their workloads, and these customers are planning multi-gigawatt scale deployments.
Client and Gaming Business
Client and gaming revenue declined 8% year-over-year to $3.7 billion. Client revenue grew 7% year-over-year to $2.9 billion, driven primarily by strong Ryzen mobile processor sales. Commercial PCs are becoming an increasingly important part of our client business, with large enterprises rolling out Ryzen Pro commercial PCs across their workforces. Eight of the world's top ten banks and seven of the world's top ten healthcare and automotive companies have now adopted Ryzen Pro commercial PCs. OEMs continue to expand their AMD-based commercial product lines, from everyday business PCs to premium workstations; HP, Lenovo and others have launched commercial laptops and small form factor desktops powered by the latest Ryzen AI Max Pro 400 series processors, delivering workstation-class performance and capable of running superintelligence models with up to 300 billion parameters locally.
From an overall market perspective, PC sales slowed toward the end of the quarter due to higher memory and component costs weighing on demand. We are adjusting our business pace accordingly to align sell-in with sell-through to maintain healthy customer and channel inventory levels. In our 2027 planning, we expect PC market shipments to decline by a double-digit percentage. In this environment, we will outperform the market through our continuously evolving product roadmap and expanding commercial market share.
Gaming revenue declined 40% year-over-year to $773 million, primarily due to lower semi-custom chip sales at the current stage of the console cycle, in line with expectations. Gaming graphics revenue declined year-over-year, but Radeon GPU channel sales grew sequentially. Beyond gaming, Radeon AI Pro revenue grew significantly from the prior quarter, with more developers choosing AMD to run and fine-tune models locally. Driven by leading performance-per-dollar advantages and an open software stack optimized for the latest open-source models, we expect Radeon AI Pro to continue its strong growth momentum.
Embedded Business
Embedded revenue grew 29% year-over-year to $1.1 billion, marking the fourth consecutive quarter of accelerating growth. Demand was very strong across end markets, with the most notable growth from aerospace and defense, automotive, communications and industrial customers. The ongoing data center buildout is also driving more demand for embedded CPUs and adaptive SoCs in control plane and optical networking applications.
We continue to progress toward our target of more than $18 billion in design wins for 2026, which would be our third consecutive record year. Since the Xilinx acquisition, adaptive design wins have grown every year, with embedded x86 and semi-custom orders growing even faster. As these wins ramp into production over the coming years, we expect the embedded business to achieve strong double-digit annual growth and continue to outperform the market.
Summary and Outlook
In summary, we delivered an excellent third-quarter report card and expect the fourth quarter to be even stronger as data center and embedded growth more than offset soft client CPU sales.
We enter 2027 with the strongest product portfolio in our history and significant customer momentum. We have mobilized our entire supply chain to secure the wafer, substrate, advanced packaging, memory and other key component resources needed to support higher growth next year.
Based on current business visibility, we expect revenue growth of more than 80% in 2027, more than double the long-term financial model target we disclosed at last November's Analyst Day. The data center business will be the largest contributor, driven by continued Helios deployment expansion and continued EPYC share gains. We also expect the embedded business to accelerate next year, benefiting from broad adoption of adaptive architectures and x86 solutions.
Looking beyond 2027, we are increasing R&D investment in our silicon, software and systems roadmap and working closely with our largest customers to support their macro infrastructure plans for the next three to five years. We are also deepening strategic collaboration with supply chain partners and co-investing with them to secure the capacity needed to support our long-term growth.
Taken together, all of this gives us strong confidence in our goal of sustaining exceptional revenue growth with even faster earnings growth over the coming years.
I will now turn the call over to Jean for a more detailed review of third-quarter financials.
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer:
Thank you, Lisa, and greetings to everyone.
I will first review our quarterly financial results and then provide our outlook for the fourth quarter of fiscal 2026.
Third Quarter Financial Results Review
Third quarter results were excellent, with revenue reaching a record $13.4 billion, above the upper end of our guidance range, driven primarily by the data center segment — which nearly doubled year-over-year.
Revenue grew 16% quarter-over-quarter, driven primarily by strong growth in data center and embedded segments, partially offset by declines in client and gaming. The data center segment accounted for approximately 64% of total revenue, up from 58% last quarter, highlighting continued optimization of our business mix and its growing contribution to AMD's growth.
Gross margin was 56%, up 190 basis points year-over-year, reflecting a more favorable product mix. Operating expenses were $3.6 billion, up 29% year-over-year, primarily due to continued R&D investment to expand 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, demonstrating the operating leverage from business scale expansion.
In the third quarter, we began recognizing the accounting impact of warrants issued under the previously announced customer agreement — triggered when we began shipping MI450 series GPUs. Our GAAP diluted EPS was $1.46, which includes approximately $1.1 billion in non-cash charges (recorded in other income and expense), primarily reflecting the impact of the company's stock price increase and changes in the estimated probability of warrant vesting conditions being met.
Going forward, as we and our customer continue to progress under this agreement, these quarterly non-cash fair value adjustments are expected to increase, with the specific amount each quarter varying based on changes in AMD's stock price and other valuation assumptions. These non-cash adjustments do not affect non-GAAP results or free cash flow.
Our non-GAAP diluted EPS was $2.09.
Business Segment Details
Data Center segment: Revenue reached a record $8.5 billion, up 97% year-over-year, driven primarily by record server revenue from strong demand for 5th Gen EPYC processors and record data center AI revenue from the launch of AMD Instinct MI450 series GPUs. Sequentially, data center revenue grew 27%, driven primarily by initial 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 expanding from 25% a year ago to 34%, demonstrating strong operating leverage in the data center business.
Client and Gaming segment: Revenue was $3.7 billion, down 8% year-over-year, primarily due to lower semi-custom and graphics sales, partially offset by increased demand for the latest generation Ryzen mobile processors. Client revenue was $2.9 billion, up 7% year-over-year, driven by higher Ryzen mobile processor sales; sequentially, client revenue declined 4% due to higher memory costs weighing on PC shipments, partially offset by improved premium processor product mix. Gaming revenue was $773 million, down 40% year-over-year, primarily due to lower semi-custom revenue and weak graphics demand; sequentially, gaming revenue was roughly flat. Client and Gaming segment operating income was $573 million, or 15% of segment revenue; compared to $867 million, or 21%, in the year-ago period.
Embedded segment: Revenue was $1.1 billion, up 29% year-over-year and 14% quarter-over-quarter, with strengthening demand across multiple end markets, most notably in aerospace and defense, automotive, communications and industrial end markets. Embedded segment operating income was $453 million, or 41% of segment revenue; compared to $283 million, or 33%, in the year-ago period.
Balance Sheet and Cash Flow
This quarter, inventory increased by approximately $3 billion sequentially to $11.5 billion to meet strong data center demand. Cash flow from operations was $1.2 billion and free cash flow was $554 million, reflecting increased working capital investment and ongoing capital expenditures to support rapid data center business growth, including backend capacity investments dedicated to data center demand. We expect to continue increasing inventory and capacity investment to support strong growth in 2027. Cash, cash equivalents and short-term investments totaled $15.8 billion at quarter-end, with total debt of $7.1 billion, including the $4.75 billion senior notes issuance completed in August.
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 accelerating growth in the data center segment. Sequentially, revenue is expected to grow approximately 27%, with the data center segment growing approximately 50% quarter-over-quarter, the client and gaming segment declining more than 20% quarter-over-quarter, and the embedded segment growing at a double-digit rate.
Additionally, fourth-quarter non-GAAP gross margin is expected to be approximately 55%, non-GAAP operating expenses are expected to be approximately $3.75 billion, non-GAAP net interest and other expense is expected to be a net benefit of approximately $4 million, the non-GAAP effective tax rate is expected to be 13%, and diluted share count is expected to be approximately 1.67 billion shares (including the estimated impact of the Taalas and World Labs acquisitions expected to close in the fourth quarter); fourth-quarter share count is not expected to be affected by the customer warrants.
Closing Remarks
The third quarter was an excellent quarter for AMD, with record revenue, significantly improved operating margin and substantial earnings growth, demonstrating disciplined execution across the company. As the data center business continues to scale, our profitability and long-term shareholder value will be fully enhanced.
I will now turn the call back to Matt for the Q&A session.
Operator: Thank you all. We will now begin the question-and-answer session. Please go ahead with your questions.
First Question: Tom O'Malley (Barclays)
Tom O'Malley: Good evening, thanks for taking my question and congratulations on the excellent results. Lisa, I'd like to start with your forward guidance for 2027. Revenue growth of 80%, but data center growing even faster, with data center accounting for more than 80% of total revenue next year. Could you elaborate on the supply side — including wafers, packaging, substrates, memory, etc. — and from the demand side, if more supply were available, would demand exceed supply?
Lisa Su, Chairman and Chief Executive Officer: Of course, Tom, thanks for your question. Entering the fourth quarter and 2027, we have strong momentum, with robust demand across both our data center CPU business and data center AI business. To address this growth rate, we have been planning for years. We have had extensive and in-depth communication with customers to fully understand their deployment needs and planned pace; we are also working closely with supply chain partners to ensure wafers, substrates, advanced packaging, memory and all other key components are in place.
Demand is indeed higher than supply, so we continue to work with supply chain partners on capacity expansion, and given the deep relationships we have across the ecosystem, I believe this is entirely achievable. From where we stand, being able to say that the overall company will grow more than 80%, with server business set to double in 2026 off an already high base, and data center AI business visibility and demand remaining strong, makes us very optimistic about the data center business in 2027.
Tom O'Malley: Very helpful. One follow-up on the client business: you mentioned a quarter-over-quarter decline of more than 20% in the fourth quarter, which is broadly consistent with what other players in the market have reported. But 2027 is worth watching — typically Q1 has a seasonal downturn, with recovery often not coming until Q2 or Q3. How long do you expect this adjustment to last? Why has AMD been able to outperform previously? And how do you emerge from the trough in 2027?
Lisa Su, Chairman and Chief Executive Officer: The PC market remains an important market for us. We have been signaling over the past several earnings calls that we expected a seasonally slow second half, and that has indeed been the case. When component prices rise, system pricing also increases, and the consumer market is naturally more sensitive to this, so volumes have slowed. We are managing the business prudently, aligning sell-in with sell-through to keep channel and customer inventory healthy.
Our planning assumption for 2027 is a double-digit percentage decline in PC market shipments. We will focus on segments where we can deliver the most value — such as the enterprise market (where we still have room to grow), premium consumer and desktop channels, which are our strongholds.
Fortunately, our supply chain has strong flexibility between the PC business and data center business, allowing us to adjust supply based on market signals. Overall, we remain in a demand-exceeds-supply situation, so we will continue working to increase supply over the coming quarters.
Second Question: Vivek Arya (Bank of America Securities)
Vivek Arya: Thanks for taking my question. Lisa, there's a lot of discussion in the market about data center financing, land, power, campus supply and other issues. I'd like to ask two things: First, how confident are you that AMD's core customers have the capital and physical infrastructure to support approximately 3GW of planned AMD deployments in 2027? Second, many competitors are claiming greater versatility through ecosystem investments, residual value guarantees and other means — does this create competitive pressure?
Lisa Su, Chairman and Chief Executive Officer: Vivek, thanks for your question. As mentioned earlier, demand from our core data center AI customers is very strong. From Venice to MI455 to Helios, we are very confident in our competitiveness and have been jointly planning this capacity ramp with customers. For data center infrastructure such as land, power and campuses, we have good visibility into which data centers and specific locations the GPUs will be deployed in.
For 2028 and beyond, my core view is: we are on a strong overall growth trajectory with a solid business foundation and strong balance sheet. On this basis, our first priority is investing in growth, including R&D, supply chain and capacity building, as well as co-investing with partners to ensure we have adequate infrastructure in place when new products launch. At the same time, we will selectively pursue forward-looking data center infrastructure arrangements with core partners, identifying and securing them years in advance. Our approach is: under the guiding principle of investing in growth, proceed strategically and selectively with such collaborations.
Vivek Arya: Okay. As a follow-up on the competitive landscape: many of your frontier lab customers are simultaneously developing their own custom accelerators — how do they decide between using EPYC or AMD products? Are the key decision factors cost per token, energy efficiency, software flexibility or deployment timeline? Are there specific workloads where AMD GPUs outperform alternatives?
Lisa Su, Chairman and Chief Executive Officer: We are very satisfied with Helios's competitiveness in performance and total cost of ownership, which are the most critical considerations in large-scale deployments. The third quarter was our first time delivering MI455 and Helios racks to customers and recognizing revenue — a major milestone. In the fourth quarter we will see substantial ramp, continuing throughout 2027.
The key is continuously demonstrating our TCO advantage across workloads. As a general-purpose GPU, we inherently have strong versatility in handling various workloads. At the same time, co-design work is advancing — beyond the MI450 series, we are developing MI500 and MI600 to ensure our technology further aligns with frontier lab needs. These are the core factors influencing deployment decisions.
Third Question: Joshua Buchalter (TD Cowen)
Joshua Buchalter: Hello, thanks for taking my question and congratulations on the progress. I'd like you to discuss the impact of HBM costs on the model — as the Instinct business scales, investors have been watching the revenue contribution per GW, and your guidance of $15.5 billion to $20 billion per GW has been fairly stable over the past year. Is this because procurement and pricing for early products were locked in through long-term contracts? Looking ahead, does the company have the ability to pass on higher raw material costs to the three core customers?
Lisa Su, Chairman and Chief Executive Officer: Josh, in short, during the capacity ramp, we have long-term agreements with some suppliers and customers. Given inflationary pressure on memory and other component costs — as more capital expenditure flows into the entire supply chain, various costs have risen — we are also communicating with customers to share these cost increases.
Ultimately, it all comes down to total cost of ownership and tokens per dollar. We need to ensure that AMD Instinct and Helios systems continue to deliver strong value advantages for customers. Even in the face of some inflationary pressure, we remain confident in delivering industry-leading TCO for AI workloads.
Joshua Buchalter: Thanks, Lisa. One follow-up for Jean: Operating expenses improved in both Q3 and Q4, which is a focus for investors. But given how fast revenue is growing in 2027 — more than double the target model — we have limited visibility into operating expenses. How should we think about opex growth in 2027? And on gross margin, as Helios continues to ramp, can the 55% floor be maintained?
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: Josh, thanks for the question. On operating expenses, as you said, we have been actively investing. Currently, these investments are translating into accelerating revenue growth. Our operating leverage performance in 2026 has been quite satisfactory — revenue growth significantly outpacing opex growth, driving operating margin improvement. In our Q4 guidance, revenue growth is also significantly higher than opex growth, and we believe this trend will continue into 2027.
As Lisa mentioned, we expect overall company growth of more than 80%, with opex growth expected to be lower than revenue growth and no more than half of revenue growth, which will drive significant operating margin expansion in 2027.
On gross margin, we are quite satisfied with the overall situation. In the third quarter, we were ramping MI450, and data center AI business performance actually exceeded our expectations entering the quarter. In the fourth quarter, the data center business will grow 50% quarter-over-quarter, and this is the quarter in which data center AI business surpasses server business — as is well known, the AI business has a dilutive effect on overall gross margin, but total gross profit is growing substantially.
Looking at 2027, the data center business will account for more than 80% of total company revenue, and the business mix shift will significantly impact gross margin. Data center AI business will continue to ramp throughout the year, server business is expected to double, and business mix differences between quarters will cause some gross margin fluctuation. The strengthening of the embedded business in 2027 will also help improve overall gross margin.
Overall, our goal is to achieve strong revenue growth and gross profit growth while significantly expanding operating margin and overall profitability, driven by opex growth being lower than revenue growth.
Fourth Question: Timothy Arcuri (UBS)
Timothy Arcuri: Thanks. Lisa, can you further break down the data center business mix for next year? Server CPU guidance was raised from 70% growth to more than 100% growth, but you didn't give a specific number for data center AI. Is the correct way to think about it: GPU business in dollar terms will still be larger than server CPU business, but not by much? Also, how should we think about the cadence through the year — will GPU business be relatively evenly distributed across the year?
Lisa Su, Chairman and Chief Executive Officer: Tim, let me break it down along several dimensions. First, we believe 2027 will be a strong momentum year for both the CPU business and data center AI business. For the reference numbers we provided, please understand them as milestone indicators: overall company growth of more than 80%; data center CPU business roughly doubling; data center AI business growth far exceeding doubling. As Jean said, data center AI business will surpass CPU business in size for the first time in the fourth quarter, and both segments are expected to grow solidly through 2027. I hope this gives everyone a basic reference for the cadence through the year.
Timothy Arcuri: Thanks. One more question for Jean: regarding the warrants, can you give us a framework for modeling through the year and into 2027?
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: Thanks for the question. Simply put, the warrants have three components.
The first is the net revenue impact. Our reported Q3 actual revenue, Q4 guidance and all subsequent revenue are already net revenue, having deducted the warrant revenue reduction based on grant date fair value — no separate adjustment needed.
The second is share count dilution. Warrant shares are only included in diluted share count after all vesting and exercise conditions are met by the customer — meaning specified shipment volumes must be completed and certain commercial conditions satisfied. Once we enter 2027, we will provide clear guidance on the expected timing for meeting vesting and exercise conditions, at which point there will be full transparency.
The third is the quarterly non-cash fair value remeasurement expense, which does not affect non-GAAP results or free cash flow. On share count, please wait for our clear guidance once we enter 2027.
Fifth Question: CJ Muse (Cantor Fitzgerald)
CJ Muse: Good afternoon, thanks for taking my question. On server CPUs, the outlook has been significantly raised. I'd like to understand, within the 100% growth, roughly how much comes from unit volume increases versus product mix (including higher core counts) and ASP increases from Venice?
Lisa Su, Chairman and Chief Executive Officer: CJ, for server CPU achieving 100% growth, significant unit volume increase is the primary driver, with some units carrying higher core counts as we transition to Venice. What I would say is that revenue growth is primarily from unit volume and core count increases, with ASP increases contributing relatively less — although there is some ASP uplift as core counts increase.
CJ Muse: Okay, thanks. One more question on Q1 seasonality: based on what you said tonight, my understanding is that Q1 client and gaming will have normal seasonal decline, data center will be roughly evenly distributed at a level above December. What about embedded?
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: The embedded business typically has seasonal patterns, and Q1 does have some seasonal decline. But for full-year 2027, the embedded business will achieve significant growth and show a gradually strengthening trend through the year.
Sixth Question: Stacy Rasgon (Bernstein Research)
Stacy Rasgon: Hello, thanks for taking my question. You mentioned that GPUs will surpass CPUs in Q4 against the backdrop of 50% guidance — how significant is this crossover? Is it slightly above, or significantly ahead? I'd like to get a basic sense of the order of magnitude entering next year.
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: Stacy, let me give everyone some reference. The MI450 production ramp is a substantial jump from Q3 to Q4, and the server business is also growing very strongly; but data center AI business surpassing server business is a very significant step-function increase. In the fourth quarter, data center AI revenue (in dollar terms) will be far higher than the server business.
Stacy Rasgon: Is that in dollar terms?
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: Yes, in dollar terms.
Stacy Rasgon: Okay. One follow-up, Jean: On next year's gross margin, I'm not sure how to interpret your earlier answer. Can you give a rough directional guide for 2027 gross margin relative to Q4 levels — including reasonable reference ranges for the beginning and end of the year?
Jean Hu, Executive Vice President, Chief Financial Officer and Treasurer: Stacy, we cannot provide quarterly guidance at this time, primarily because the gross margin trajectory in 2027 will be dominated by the internal mix of the data center business — server will double, AI business will grow even faster, and the difference in their revenue proportions will cause some gross margin fluctuation between quarters.
For now, Q4 is a good reference baseline — at which point data center AI revenue will be far higher than CPU revenue. Entering 2027, the business mix may differ across quarters. We are also constrained by current supply visibility and are working to secure more supply with suppliers. Both segments will achieve excellent growth — that is certain — but the specific quarterly mix distribution is not yet clear.
Seventh Question: Joseph Moore (Morgan Stanley)
Joseph Moore: Thanks. We've heard a lot of positive feedback about the Helion rack. In an all-CPU configuration, how does this product differ from traditional racks for agentic workloads? And what attach rate do you expect for GPU blades?
Lisa Su, Chairman and Chief Executive Officer: Joe, the Helion rack has indeed generated quite enthusiastic positive feedback. The core premise is that these agentic racks need to be deeply optimized for their environment. In hyperscale cloud scenarios, all-CPU configurations with high-density deployment are mainstream; Venice, as the best single-socket performance processor, can achieve extremely high agentic compute density in a fully populated all-CPU configuration. In enterprise scenarios, there will be demand for hybrid CPU-GPU deployments, allowing users to truly optimize configurations flexibly based on workloads. Helion has contributed a full set of open specifications to the Open Compute Project (OCP), and the fully open rack design has won positive market reception. Feedback from OEMs, ODMs and end customers has all been very positive.
Joseph Moore: Okay, thanks. One more question: With server achieving 100% growth, do you believe you are essentially meeting market demand, or will supply remain tight throughout 2027?
Lisa Su, Chairman and Chief Executive Officer: Joe, achieving this level of growth itself means we have satisfied a great deal of demand. Server TAM is growing at roughly 50% annually over the next three to four years, with growth coming from three areas: traditional cloud and enterprise, agentic CPU workloads and head nodes. We believe that in our approximately 100% growth forecast, every sub-segment is gaining share.
I do think there is more demand in the market yet to be met, and we are actively working with supply chain partners to address this, while also ensuring that chip sets, DDR5 and other supporting components are ready. Demand sources are highly diversified, covering nearly every customer that is expanding AMD deployment scale — which is an encouraging positive signal. We will continue working to increase supply to meet these continually materializing incremental demands.
Eighth Question: Harsh Kumar (BMO Capital)
Harsh Kumar: Lisa, I'd like to ask a question about Helios. Your company is one of only two globally shipping complete rack systems. What is your market share target? You have achieved excellent results competing for share against strong competitors in the server market — what are your ambitions in data center GPUs and complete racks? Also, does the company intend to enter the enterprise market?
Lisa Su, Chairman and Chief Executive Officer: Harsh, we are very pleased with Helios progress. Rack-level solutions represent significant R&D investment, and Helios capacity ramp is going very smoothly: the third quarter was a good initial quarter, with continued ramp in Q4 and 2027. Our goal is to become one of the most important compute suppliers for our largest customers — whether frontier labs or hyperscalers. On the MI450 production ramp, we are highly focused in Q4 and throughout 2027; we are also in deep discussions with our largest customers about MI500 and MI600, aiming to win a larger share of their compute needs. From everything we see, demand continues to rise.
Regarding the enterprise market, although enterprise customer deployment pace is indeed slower than large frontier labs and hyperscalers, we believe the enterprise market is very important. We have made great progress in enterprise CPUs, and many enterprise customers are proactively consulting us about AI deployment efficiency, including both on-premises and cloud deployment. We are expanding our enterprise-focused product portfolio, having launched some enterprise-specific GPUs, and hope to continue expanding share in this area to become a strategic core supplier for large enterprise customers.
Harsh Kumar: Okay, understood. One last question on the client business: Historically, your ability to manage this business in similar situations has been evident. If major competitors choose to cut prices, what is your response strategy? Will you follow, or decide based on circumstances?
Lisa Su, Chairman and Chief Executive Officer: Harsh, the PC market is an important market for us — I have always believed in the value of end-to-end AI deployment, from cloud to edge to client, and we have a uniquely complete product portfolio in this dimension. On this basis, our strategy in the PC business has always been to focus highly on the highest-value, highest-margin segments — enterprise, premium consumer and desktop channels, which are our strong lanes. We are keenly aware that the low-end market has significant volatility, and frankly, that is also where we have the least exposure in the PC market.
Operator Closing Remarks
Matt Ramsay, Corporate Vice President, Financial Strategy and Investor Relations: Thank you very much to John for organizing this Q&A, and thank you all for participating in our conference call and for your questions and attention. John will close the meeting. Thank you everyone.
Operator: Thank you all for participating. Today's conference call is now concluded. Please disconnect your lines and have a great day.
Comments