Advanced Micro Devices - Earnings Call - Q2 2025

Executive Summary

• Record revenue of $7.685B (+32% y/y, +3% q/q) with non-GAAP EPS $0.48; revenue beat Street ($7.43B*) while EPS was essentially in line ($0.483*), as $800M export-control inventory charges cut gross margin to 40% GAAP and 43% non-GAAP; excluding charges, non-GAAP GM ~54%.
• Data Center revenue rose 14% y/y to $3.24B, but segment posted a $(155)M operating loss due to export-control charges; Client & Gaming revenue surged 69% y/y to $3.62B with operating income $767M; Embedded revenue fell 4% y/y to $824M.
• Guidance: Q3’25 revenue ~$8.7B ±$300M, non-GAAP GM ~54%, OpEx ~$2.55B, OI&E net ~$10M, tax ~13%, diluted shares ~1.63B; outlook excludes any MI308 China revenue pending license review—key upside catalyst if approvals arrive.
• Stock narrative catalysts: rapid MI350 accelerator ramp with broad hyperscaler adoption, continued EPYC CPU share gains, plus potential China license approvals; near-term margin puts/takes from MI gross margin mix versus high-margin Client/Gaming strength.

What Went Well and What Went Wrong

What Went Well

• Record server and PC processor sales drove 32% y/y revenue growth and record free cash flow ($1.18B), despite export-control headwinds; “We delivered strong revenue growth… robust demand across our computing and AI product portfolio” — Dr. Lisa Su.
• Client & Gaming strength: Client revenue hit a record $2.5B (+67% y/y) on “Zen 5” Ryzen and richer mix; Gaming revenue rose 73% y/y to $1.1B on semi-custom and Radeon demand.
• MI350 ramp and AI roadmap: launched MI350 with leadership memory bandwidth/capacity; ROCm 7 delivers >3x perf vs prior gen; Oracle building 27k+ node cluster combining MI355X, 5th Gen EPYC and Polara 400.

What Went Wrong

• Export-control impact: ~$800M inventory and related charges tied to MI308 restrictions reduced non-GAAP EPS by ~$0.43 and drove GAAP gross margin down to 40% and Data Center operating loss.
• Non-GAAP profitability compression: non-GAAP GM fell to 43% (–10ppts y/y), non-GAAP operating income fell to $897M (–29% y/y), reflecting charges and higher investment spend.
• Embedded softness: revenue down 4% y/y with operating margin down to 33% (from 40%) on mixed end-market demand and product mix.

Transcript

Operator (participant)

Greetings, and welcome to the AMD second quarter 2025 conference call. At this time, all participants are in a listen-only mode. A question and answer session will follow the formal presentation. If anyone should require operator assistance, please press star zero on your telephone keypad. As a reminder, this conference is being recorded. It is now my pleasure to introduce to you Matthew Ramsay, VP of Investor Relations and Financial Strategy. Thank you, sir. Please go ahead.

Matthew Ramsay (VP of Investor Relations and Financial Strategy)

Thank you, and welcome to AMD's 2025 second quarter financial results conference call. By now, you should have had the opportunity to review a copy of our earnings press release and the accompanying slides. If you have not had the chance to review these materials, they can be found on the Investor Relations page of amd.com. We will refer primarily to non-GAAP financial measures during today's call. The full non-GAAP to GAAP reconciliations are available in today's press release and slides posted on our website. Participants in today's conference call are Dr. Lisa Su, our Chair and Chief Executive Officer, and Jean Hu, our Executive Vice President, Chief Financial Officer, and Treasurer. This is a live call and will be replayed via webcast on our website.

Before we begin, I would like to note that Jean Hu, Executive Vice President, Chief Financial Officer, and Treasurer will present at Cities 2025 Global TMT Conference on Wednesday, September 3, and Forrest Norrod, Executive Vice President and General Manager of Data Center Solutions Business Unit, will present at the Goldman Sachs CommuniCopia and Technology Conference on Monday, September 8. Today's discussion contains forward-looking statements based on our current beliefs, assumptions, and expectations, speaks only as of today, and as such, involves risks and uncertainties that could cause actual results to differ materially from our current expectations. Please refer to the cautionary statement in our press release for more information on factors that could cause actual results to differ materially. With that, I will hand the call over to Lisa.

Lisa Su (CEO and Chair)

Thank you, Matt, and good afternoon to all those listening today. We delivered very strong second quarter results, with revenue exceeding the midpoint of guidance as higher EPYC and Ryzen processor sales more than offset headwinds from export controls that impacted Instinct sales. We set records for both EPYC and Ryzen CPU sales, reflecting the broad-based demand for our differentiated high-performance data center, PC, and embedded processors. Second quarter revenue increased 32% year over year to a record $7.7 billion, and we delivered over $1 billion in free cash flow. Excluding the $800 million inventory write-down related to data center AI export controls, gross margin was 54%, marking our sixth consecutive quarter of year-over-year margin expansion led by a richer product mix. Turning to the segments, data center segment revenue increased 14% year over year to $3.2 billion.

We saw robust demand across our EPYC portfolio to power cloud and enterprise workloads, and increasingly for emerging AI use cases. In particular, adoption of agentic AI is creating additional demand for general-purpose compute infrastructure, as customers quickly realize that each token generated by a GPU triggers multiple CPU-intensive tasks. Against this backdrop, fifth-gen EPYC turn shipments ramped significantly, and we had sustained demand for our prior generation EPYC processors. As a result, we set records for both cloud and enterprise CPU sales and delivered our 33rd consecutive quarter of year-over-year share gains. In cloud, adoption expanded with the largest hyperscalers as they deployed EPYC to power more of their mission-critical infrastructure, services, and public cloud products.

More than 100 new AMD-powered cloud instances launched in the quarter, including multiple turn instances from Google and Oracle Cloud that deliver up to twice the performance of our previous generation, which were already the industry's highest performing offerings. There are now nearly 1,200 EPYC cloud instances available globally as providers continue expanding both the breadth and regional availability of their AMD offerings. This continued expansion is accelerating enterprise adoption of EPYC in the cloud, with deployments growing significantly from the prior quarter as we close large wins with dozens of large aerospace, streaming, financial services, retail, and energy companies. EPYC adoption also grew with telecom customers as providers modernize their infrastructure for next-generation networks. For example, KDDI announced plans to deploy EPYC processors to power its 5G virtualized network, and Nokia selected EPYC for its cloud platform used by service providers to build, deploy, and manage core network functions.

Turning to enterprise on-prem adoption, HPE, Dell, Lenovo, and Super Micro launched 28 new turn platforms in the quarter that deliver leadership performance, efficiency, and TCO across a wide range of enterprise workloads. EPYC enterprise deployments grew significantly from the prior quarter, supported by new wins with large technology, automotive, manufacturing, financial services, and public sector customers. To extend our momentum with SMB and hosted IT service customers, we launched the EPYC 4005 series that combined enterprise-grade performance and features in cost-optimized platforms purpose-built for smaller-scale deployments. Turning to HPC, AMD now powers more than one-third of the world's fastest supercomputers, including El Capitan and Frontier, which retain the number one and number two spots on the latest top 500 list. We also power 12 of the top 20 systems on the Green 500, highlighting the performance per watt advantages of EPYC and Instinct for large-scale deployments.

Looking ahead, we remain bullish on our service CPU business, driven by durable tailwinds, including growing demand for cloud and on-prem compute, sustained share gains, and the growing investments in general-purpose infrastructure required to enable AI. Turning to our data center AI business, revenue declined year over year as U.S. export restrictions effectively eliminated MI-308 sales to China, and we began transitioning to our next-generation MI-350 series accelerators. We made solid progress with MI-300 and MI-325 in the quarter, closing new wins and expanding adoption with tier-one customers, next-generation AI cloud providers, and end users. Today, seven of the top 10 model builders and AI companies use Instinct, underscoring the performance and TCO advantages of our data center AI solutions. We launched our Instinct MI-350 series with industry-leading memory bandwidth and capacity and broad adoption across hyperscalers, AI companies, and OEMs.

From a competitive standpoint, MI-350 series matches or exceeds B200 in critical training and inference workloads and delivers comparable performance to GB200 for key workloads at significantly lower cost and complexity. For at-scale inferencing, MI-350 series delivers up to 40% more tokens per dollar, providing leadership performance and clear TCO advantages. With the MI-350 series, we're also expanding our system-level capabilities to support deployments powered by AMD CPUs, GPUs, and NICs. As one example, Oracle is building a 27,000-plus node AI cluster, combining MI-350 series accelerators, fifth-gen EPYC turn CPUs, and Polara 400 Smart NICs. We began volume production of the MI-350 series ahead of schedule in June and expect a steep production ramp in the second half of the year to support large-scale production deployments with multiple customers.

Our sovereign AI engagements accelerated in the quarter as governments around the world adopt AMD technology to build secure AI infrastructure and advance their economies. As one example, we announced a multi-billion dollar collaboration with a leading company to build AI infrastructure powered entirely on AMD CPUs, GPUs, and software. Initial deployments are underway in key regions, with quarterly expansions planned over the coming years. In addition, we have more than 40 active engagements globally and see significant opportunities to power an increasingly larger portion of national computing centers and sovereign AI initiatives. On the AI software front, we made significant progress this quarter, increasing the performance, improving the usability, and expanding the adoption of ROCm.

We announced ROCm7 with major upgrades across every layer of the stack, delivering more than 3x higher inferencing and training performance compared to our prior generation and adding support for large-scale training, distributed inference, and lower precision data types. To deepen developer engagement, we introduced nightly ROCm builds and expanded access to Instinct compute infrastructure, including launching our first developer cloud that provides pre-configured containers for instant access to AMD GPUs. We also expanded native support for ROCm across key frameworks, including vLLM and SGLang, enabling Frontier models like Llama 4, Gemma 3, and DeepSeek R1 to launch with day zero AMD support. To accelerate enterprise adoption, we introduced ROCm Enterprise AI, a full-stack platform that integrates seamlessly with existing IT infrastructure and includes everything needed for an enterprise to deploy, manage, and scale AI across their business. Looking ahead, the development of our next-generation MI-400 series is progressing rapidly.

These are the most advanced GPUs we have ever built, with up to 40 petaflops of FP4 AI performance and 50% more memory, memory bandwidth, and scale-out throughput than the competition. With the MI-400 series, we're bringing together everything we've learned across silicon, software, and systems to deliver Helios, a full-stack rack-scale AI platform. Helios is purpose-built for the most demanding AI workloads, with each rack connecting up to 72 GPUs that can operate as a single massive AI accelerator. Helios is expected to deliver up to a 10x generational performance increase for the most advanced Frontier models, and we believe it will be the highest performance AI system in the world when it launches. MI-400 series development is progressing well towards our planned launch in 2026, with significant interest in large-scale deployments from multiple high-profile customers.

To accelerate our development, we have invested significantly to expand our AI software and hardware capabilities, both organically and inorganically, with a number of acquisitions and strategic investments. We strengthened our software stack last quarter with the addition of the BREEAM and Lemony teams, building on our acquisitions of Node.ai, MIPSology, and SiloAI. On the hardware side, we added a world-class rack and data center scale design team in the second quarter with our acquisition of ZT Systems. The ZT team has integrated seamlessly, and they are actively engaging with multiple customers to accelerate deployments of our Helios solutions at scale. We also announced last quarter that Sanmina intends to acquire ZT Systems' U.S.-based manufacturing business, becoming our lead partner for AI rack manufacturing.

Turning to the AI regulatory environment, earlier this quarter, we were notified by the Department of Commerce that it is moving forward with the review of our license applications to export Instinct MI-308 GPU to China. We appreciate the focus the Trump administration is placing on assuring that U.S. technology remains central to global AI infrastructure, and we expect to resume Instinct MI-308 shipments as licenses are approved, subject to end customer demand and supply chain readiness. As our licenses are still under review, we are not including any Instinct MI-308 revenue in our third quarter guidance. Despite that, we expect Instinct revenue to grow year over year in the third quarter, driven by the ramp of Instinct MI-350 at multiple customers.

In client and gaming, segment revenue increased 69% year over year to $3.6 billion, driven by record client CPU sales and strong demand for our semi-custom game console SoCs and Radeon GPUs. Client revenue increased 67% year over year to $2.5 billion, led by record desktop CPU sales. Demand for our latest generation Ryzen 9000 series was strong, especially for our differentiated X3D processors. We delivered record desktop channel CPU sales as Ryzen processors consistently topped the best-selling CPU lists at major global e-tailers throughout the quarter. We also expanded our Zen 5 desktop portfolio with the launch of our latest Threadripper processors that feature up to 96 cores and deliver up to double the performance of the competition in many popular content creation and design workloads. In mobile, demand for AMD-powered notebooks was strong, with sell-out growing by a large double-digit percentage year over year.

We drove a richer mix of higher ASP mobile parts year over year as we expanded our share in the premium notebook segment, where our Ryzen AI 300 CPUs deliver leadership performance and value for both general purpose and AI workloads. In commercial PCs, Ryzen adoption accelerated as OEM consumption increased more than 25% year over year. We saw strong sell-through for AMD commercial notebooks with Lenovo and HP, and a significant uptick in Dell sales as they ramp availability of their AMD commercial portfolio. We also closed new enterprise wins with Forbes 2000, pharma tech, automotive, financial services, aerospace, and healthcare companies. We expect to continue growing our commercial client share based on the strength of our product portfolio and expanded breadth of OEM offerings.

Looking more broadly, we remain confident we can continue growing client processor revenue ahead of the market over the coming quarters, driven by increased adoption of our desktop and notebook products, growing commercial momentum, and a richer product mix. In gaming, revenue increased 73% year over year to $1.1 billion. Semi-custom revenue increased by a large double-digit % year over year as console inventories normalized and our customers began preparing for the holiday season. We announced a new multi-year collaboration with Microsoft for custom chips that will power the next generation of Xbox devices, including consoles, PCs, and handhelds. We also deepened our collaboration with Sony through Project Amethyst, a co-engineering program that will use machine learning to power the next wave of immersive gaming experiences.

In PC gaming, demand for our latest generation Radeon 9000 series GPUs was very strong, with desktop GPU sell-through accelerating in the quarter as demand outpaced supply. We launched the Radeon 9600 XT, extending the performance advantages of RDNA 4 to mainstream gamers and delivering a significant uplift in gaming performance, including more than double the ray tracing of our prior generation. As part of our end-to-end AI strategy, we introduced the Radeon AI Pro R9700 GPU for local inferencing, model fine-tuning, and other data-intensive workloads. The R9700 features more memory, full ROCm support, and multi-GPU scalability, enabling advanced AI development and deployment directly on the desktop. Turning to our embedded segment, revenue decreased 4% year over year to $824 million.

Demand continues recovering gradually, with sell-through in the second quarter picking up as strength in most markets was offset by a few pockets of softness and inventory reduction actions, largely with industrial customers. We expanded our embedded portfolio with the first production shipments of Spartan UltraScale+ FPGA that deliver leadership performance and advanced security for cost-sensitive low-power applications. Adoption of our Versal adaptive SoCs continues expanding in high-end applications, including next-generation robotaxi platforms developed by Bosch in Europe, where Versal serves as a high-performance controller, enabling real-time processing, security, and encryption in fully electric automated vehicles. Looking ahead, we expect improving demand in the test and measurement, communications, and aerospace markets will drive a return to sequential growth in the second half of 2025.

Longer-term, design win momentum continues to build, tracking ahead of this point last year and putting us on pace to surpass the record $14 billion in design wins we achieved in 2024. In summary, demand is very strong across our product portfolio, and we are well positioned to deliver significant growth in the second half of the year, led by the steep ramp of MI-350 series accelerators and ongoing EPYC and Ryzen share gains. Our server and PC CPU businesses are accelerating, driven by growing demand for high-performance compute, sustained share gains, the strength of our product portfolio, and expanded go-to-market investments. Our embedded and gaming businesses are returning to growth and are well positioned for long-term success, supported by strong design win momentum.

In AI, we are seeing strong adoption of our MI-350 series and ROCm7 as we deliver leadership performance and TCO advantages across a broader range of workloads and ramp deployments with an expanded set of cloud and enterprise customers. Looking ahead, we see a clear path to scaling our AI business to tens of billions of dollars in annual revenue. We are very excited about our next-generation MI-400 series, which is another giant step forward on our roadmap and has been designed to deliver leadership performance at the chip, server, and rack levels. Customer interest for the MI-400 series is very strong, and we are actively engaging with an expanding set of customers to support large-scale deployments in 2026.

We are in the early stages of an industry-wide AI transformation that will drive a step function increase in compute demand across all of our markets, positioning us for significant revenue and earnings growth over the coming years. Now, I'd like to turn the call over to Jean to provide some additional color on our second quarter results. Jean?

Jean Hu (CFO, EVP, and Treasurer)

Thank you, Lisa, and good afternoon, everyone. I'll start with a review of our financial results and then provide our outlook for the third quarter of fiscal 2025. We are pleased with our strong second quarter financial results. We delivered record revenue of $7.7 billion, exceeding the midpoint of our guidance, up 32% year over year, reflecting strong momentum across our business. Record sales of Ryzen and EPYC processors and higher semi-custom shipments more than offset the impact of the U.S. export controls, restricting Instinct MI-308 sales to China. Revenue increased 3% sequentially due to strong growth in the client and gaming segment, partially offset by the data center revenue decrease due to export controls. Gross margin was 43%, down 10 points from 53% a year ago. The decrease was due to the $800 million inventory and related charges associated with export restrictions.

Excluding this charge, non-GAAP gross margin would have been approximately 54%. Operating expenses were approximately $2.4 billion, and the increase of 32% year over year as we continue to invest aggressively in go-to-market activities for revenue growth and in R&D to capitalize on significant future AI expansion opportunities. Operating income was $897 million, representing a 12% operating margin compared to $1.3 billion, up 22% a year ago. The decline was primarily due to the inventory and related charges. Taxes, interest expense, and other totaled $126 million. For the second quarter of 2025, diluted earnings per share were $0.48 compared to $0.69 a year ago. The inventory and related charges reduced earnings per share by approximately $0.43. Now, turning to our reportable segment, starting with the data center.

Data center segment revenue was $3.2 billion, up 14% year over year, driven by strong EPYC CPU revenue and share gains across both cloud and enterprise customers. On a sequential basis, data center revenue decreased 12% due to the impact of the export controls on Instinct MI-308 GPU. The data center segment operating loss was $155 million compared to operating income of $743 million a year ago, up 26% of revenue. The loss was primarily due to the inventory and related charges. Client and gaming segment revenue was $3.6 billion, up 69% year over year and 20% sequentially, driven by record client CPU sales and strong demand for our PC and console gaming products. In the client business, revenue was a record of $2.5 billion, up 67% year over year, driven by record sales of our Ryzen desktop CPUs and the richer product mix.

Gaming revenue rose to $1.1 billion, up 73% year over year, reflecting strong demand for our newly launched gaming GPUs and higher semi-customer revenue as inventory has now normalized and customers prepare for the holiday season. Client and gaming segment operating income was $767 million, up 21% of revenue compared to $166 million, up 8% a year ago, driven by richer client product mix and operating leverage on higher revenue. Embedded segment revenue was $824 million, down 4% year over year and flat sequentially, as embedded end market demand remains mixed. Embedded segment operating income was $275 million, up 33% of revenue compared to $345 million, up 40% a year ago. The decline in operating income was primarily due to product mix. Before I review the balance sheet and the cash flow, as a reminder, we closed the acquisition of ZT Systems early in the second quarter.

As we had announced our intent to divest the ZT manufacturing business, the financial results of the ZT manufacturing business are reported separately in our financial statement as discontinued operations and are excluded from our non-GAAP financials. Subsequently, in May, we entered into agreement withSanmina Corporation to sell the ZT manufacturing business for $3 billion in cash and stock, inclusive of contingent payment. The transaction is expected to close near the end of 2025, subject to regulatory approvals and customary closing conditions. Turning to the balance sheet and the cash flow, during the quarter, we generated $1.5 billion in cash from operating activities of continuing operations and free cash flow was a record of $1.2 billion. We returned $478 million to shareholders through share repurchase, resulting in $1.2 billion in share repurchases for the first half of 2025.

In May, our board of directors approved an additional $6 billion authorization, and at the end of the quarter we have $9.5 billion remaining in our share repurchase program. At the end of the quarter, cash, cash equivalents, and short-term investment were $5.9 billion. Our long-term debt was $3.2 billion. During the quarter, we paid down $950 million of commercial paper used to finance the ZT Systems acquisition close. Now, turning to our third quarter 2025 outlook, please note that our third quarter outlook does not include any revenue from AMD Instinct MI-308 shipment to China, as our license applications are currently under review by the U.S. government. For the third quarter of 2025, we expect revenue to be approximately $8.7 billion, plus or minus $300 million.

The midpoint of our guidance represents approximately 28% year-over-year revenue growth, driven by strong double-digit growth in our client and gaming and the data center segments. Sequentially, we expect revenue to grow by approximately 13%, driven by strong double-digit growth in the data center segment with the ramp of our AMD Instinct MI-350 series GPU products, modest growth in our client gaming segment with client revenue increasing and gaming revenue to be flattish, and our embedded segment revenue to return to growth. In addition, we expect third quarter non-GAAP gross margin to be approximately 54%, and we expect non-GAAP operating expenses to be approximately $2.55 billion. We expect net interest and other expenses to be a gain of approximately $10 million. We expect our non-GAAP effective tax rate to be 13%, and diluted share count is expected to be approximately 1.63 billion shares.

In closing, we executed very well in the first half of the year, delivering record revenue and building strong momentum for growth in the second half. The strategic investment we're making positions us to capitalize on the expanding AI opportunities across all our end markets, driving sustainable long-term revenue growth and earnings expansion for compelling value creation. With that, I'll turn it back to Matt for the Q&A session.

Matthew Ramsay (VP of Investor Relations and Financial Strategy)

Operator, will you please poll the audience for questions? Thank you.

Operator (participant)

Yes, thank you, Matt. We will now be conducting a question and answer session. If you would like to ask a question, please press star one on your telephone keypad. A confirmation tone will indicate that your line is in the queue. You may press star two to remove yourself from the queue. For participants using speaker equipment, it may be necessary to pick up the handset before pressing the star keys. We ask that you please limit yourself to one question and one follow-up. Thank you. One moment, please, while we poll for questions. The first question comes from the line of Thomas O'Malley with Barclays. Please proceed with your question.

Thomas O'Malley (Analyst)

Hey, thank you for taking my questions, and I appreciate it. Lisa, I wanted to start on the client business. You had previously laid out a second half outlook that was roughly flattish with the first half as you were kind of protecting against some pull forwards. First, do you think that your Q2 results included some pull forwards and the second half should still be flattish? Longer term, after the Intel commentary regarding 18A, maybe what that means is a knee-jerk reaction just right away for AMD longer term in terms of share and ASPs.

Lisa Su (CEO and Chair)

Sure, Tom, thanks for the question. First of all, you know our client and gaming segment, and particularly our client business, just performed very well in the first half of the year. I think if you look at the entire first half, it was up 68% year over year. If you look underneath that, what we're seeing is strength in every part of our client business. We saw very strong sales in our desktop channel area. We have a leadership product there, best gaming GPUs with our X3D GPUs. We've had strong Ryzen AI adoption as well in the first half of the year. We see that in sell-through. In addition, we've had strong enterprise sell-through as we brought that forward. To your question of how much is pull forward, we don't think a whole lot of that is.

When we look at the sell-through patterns, the end user consumption is actually quite strong for client in terms of going into the second half of the year. As we said in our Q3 guide, the primary driver of our Q3 guide is a very strong data center driven by MI-350 ramping. We are expecting some growth in the client business. I wouldn't say it'll be flat to the first half, but we're planning for it to be a little bit less than seasonal, just given some of the uncertainties out there. The client business is performing extremely well for us. We believe we are gaining share in all the right places. If you look at the numbers in the first quarter, and it'll show through in the second quarter as well, a lot of the uplift in revenue is in ASPs.

That is basically, we're selling up the stack on the strength of our portfolio. I think we're still quite underrepresented in the enterprise portion of the business. That is where we have increased our go-to-market resources and focus. We're seeing nice traction there, especially with the portfolios that we have from HP and Lenovo in enterprise PCs. Now we're adding Dell as well, as it's ramping here, started in the second quarter, will ramp more in the second half of the year. All of those are tailwinds for our client business beyond the second half of 2025, but really into the next number of quarters as we think about the portfolio and the opportunities for us.

Thomas O'Malley (Analyst)

Super helpful. Secondly, I was hopeful you could provide us a little more color on China. The guide doesn't include MI-308, but perhaps you could comment on when you get approval, if the supply chain's ready, what's currently in inventory, and maybe compare what you think the contribution will look like versus the $700 million in Q2 and the $800 million for the second half you spoke about in April.

Lisa Su (CEO and Chair)

Sure, Tom. Let me answer some of the questions on China. I'm sure that there are some questions. Look, we're very pleased with the progress that's been made with the administration over the last couple of months. We've been working very closely with the administration. I think the focus here on ensuring that U.S. technology gets utilized throughout the world is something that we certainly support and very much want to contribute to. China is an important market for us. Given the timing of licenses, we have a number of licenses that are under review now. We are working with the Department of Commerce to get those reviewed. We do expect that once those licenses are approved, we will start MI-308 shipments. In terms of the supply chain, most of our inventory was not in finished goods, so it was work in process.

It'll take us a couple of quarters to run through that. The exact timing of revenue and contribution will depend a bit on when the licenses are actually granted. Overall, I think this is a better position than we were 90 days ago. We certainly view China as a market that we would like to service with MI-308. We're working closely with the administration to do that.

Operator (participant)

The next question comes from the line of Vivek Arya with Bank of America Securities. Please proceed with your question.

Vivek Arya (Analyst)

Thank you for taking my question. Lisa, if we look into 2026, right, that's when I think the sovereign opportunity could get quite meaningful for AMD. What is the right way to size that? What does this JV structure mean with some of the contracts that you have signed? Would you consider this incremental to the kind of growth rate that you're seeing with your current MI business, or would this be instead of? If you could give us a way to size, what is that incremental opportunity from sovereign customers when it comes to 2026? Is it dependent on MI-400, in which case it might be more backup data, etc.? Just some ways to think about sovereign for AMD next year.

Lisa Su (CEO and Chair)

Yeah, absolutely, Vivek. Thanks for the question. We're really excited about the overall AI opportunity for us with MI-355 and the MI-400 series as we go through the back half of this year and into 2026. I think there's a very large opportunity with, let's call it hyperscalers, some of the leading AI companies, as well as Sovereign. I think Sovereign is additive to that for sure. From the standpoint of what to expect, there are also some regulatory things that need to be worked through on the Sovereign side. We're working closely with the administration as they go through the various regulatory decisions that need to be made. From my perspective, I think the fact that countries want their own sovereign computing capability is very, very clear. I think we see that all over the world.

The humane opportunity that you're referring to that we announced with the Kingdom of Saudi Arabia, I think, is a great example of where, together with their ambitions, our technology, I think you heard from Tarek that he was at our event saying that that would start with MI-355, that we would expect that that would continue on. I think what's attractive about our offering is our open ecosystem. I think that really resonates with the Sovereign community. To your original question, I think it's an additive opportunity. It's one that we believe will continue to be very important for us going forward with both MI-355 as well as the MI-400 series.

Vivek Arya (Analyst)

Got it. For my follow-up, I wanted to ask about gross margins for your MI product. I understand in the early days, right, it has been dilutive. What kind of sales level is required for it to start becoming additive to margins? If I fast forward to Q4 and assume that your Q4 sales are growing year over year at roughly the same rate as Q3 sales, should we expect gross margins to kind of stay at these Q3 levels, or are there other plus-minus drivers we should think about in terms of gross margins as you go into Q4? Thank you.

Jean Hu (CFO, EVP, and Treasurer)

Yeah, Vivek, thank you for the question. The gross margin of our MI product, we said it's a little bit below corporate average. At this point, our priority is really to address the larger, faster-growing revenue opportunities we have and provide customers better TCO to really expand our presence in the marketplace. The way to think about our gross margin, there are different dynamics, right, different customers, different generations. Also, our operation team has been continuing to really drive operational efficiency to improve our MI family's gross margin. That has been ongoing. It's not necessarily really tied to, OK, the revenue level each quarter, but you should think about it as a trend in the longer term. It should improve continually going forward. Overall, the way I think about it is gross margin dollars, right? This is one of the fastest-growing market opportunities for any financial metrics.

Gross margin dollars is what we try to grab as much as we can. Hopefully, that helped your question.

Operator (participant)

The next question comes from the line of Timothy Arcuri with UBS. Please proceed with your question.

Timothy Arcuri (Analyst)

Thanks a lot. Lisa, my question is on data center GPU. You did say that June is up year over year, so it sounds like it's maybe a little more than $1 billion. You use words like strong ramp into the back half of the year. Can you give us any color on what that means? Can you get to, say, $7 billion for the year? Can you give us some mile post on what you're assuming for Q3? That would be great.

Lisa Su (CEO and Chair)

Yeah, thanks, Tim, for the question. I think what we said in the prepared remarks is that we are seeing a strong ramp from Q2 into Q3. MI-355, we actually started production in June. We had some shipments in the month of June, but it really is ramping as we go through this quarter and the third quarter. In terms of guideposts, we said it would grow year over year from last year. That, I think, is a strong ramp. We would expect it to grow into the fourth quarter as well. The demand, I should say, what we're seeing from customers is, I think, really positive around MI-355. The way I would contrast it with maybe the MI-300 ramp, I think MI-300 started with perhaps some smaller deployments. What we're seeing with MI-355 is very competitive versus the B200, GB200 family of products.

I think there's a strong desire to really use us at scale. MI-355 is very strong for inferencing. We're also working with a number of customers on training. This is also an opportunity for us to build into the MI-400 series as we go into 2026. We're bullish on MI-355 and where the AI opportunity is for us. I think we're right on track to what we expected to be as we were going through the development of the roadmap.

Timothy Arcuri (Analyst)

Thanks a lot. Lisa, just on that point also, you did talk about a new developer cloud. Obviously, you're beginning to lease back some of the capacity that you're selling into the clouds and the Neo clouds. Is that going to be a material portion of the revenue you're going to recognize for MI-355 in the back half of the year? Can you just talk about that and maybe how to think about how much demand that's going to stimulate and what the ultimate goal is for that cloud? Thanks.

Lisa Su (CEO and Chair)

There are a couple of things in that question. Let me answer. The developer cloud is simply, we want to make it super easy for developers to get on AMD Instinct GPUs. One could say, again, if we look back at the MI-300 family, we were very focused on the largest hyperscalers and the largest customers. There is a lot of interest in our GPUs across a number of customers who just wanted easier access. By ensuring that a developer cloud is there, that it has ready to deploy containers, you can run training and inference easily. You don't necessarily have to make longer-term commitments. I think that's the purpose of the developer cloud. I don't think it adds meaningfully to revenue in the second half of the year, but it certainly adds to customers getting experience with AMD.

The larger revenue opportunities for us are really with large customer adoption as they ramp to larger deployments. We're very actively trying to get those deployments up and running as soon as possible. One of the things, just as a reminder, that the MI-355 is, given that it's a similar infrastructure to MI-300, we actually think it's going to ramp very quickly and very well for customers. I think that's one of the attractive portions of it as well.

Operator (participant)

The next question comes from the line of Ross Seymore with Deutsche Bank. Please proceed with your question.

Ross Seymore (Analyst)

Hi, thanks for letting me ask a question. Lisa, I want to go back to the Instinct side of things and the MI-350 ramp. Looks like the second half is going to ramp really significantly. You said it's going to be up year over year in the third quarter. I believe a quarter ago, you said roughly the same thing. The MI-308 is out of both numbers, so that shouldn't really matter. I guess it would be upside. I've just wondered, how have things changed from a quarter ago as far as the MI-350 family adoption, especially because you launched it a little bit early? Is the growth a little bit more than you would have expected a quarter ago, about the same, or a little worse? Any sort of color on that would be helpful.

Lisa Su (CEO and Chair)

Yeah, Ross, thanks for the question. I think the main thing I would say is I think the adoption is a bit faster than we might have expected. Whenever you launch a product, we want to make sure that we go through the full validation and all of that with our customers. I think there's a lot of interest, broad-based interest in MI-350 series. I feel like over the last 90 days, we've had significant sort of new customer interest, and that's certainly positive. I will say.

Ross Seymore (Analyst)

Great, I guess.

Lisa Su (CEO and Chair)

I'm sorry, Ross. I was just going to add, our engagements are, I think the other piece is I think there's also a lot of excitement around MI-400 and what we can do with the Helios rack. There are a number of customers who, based on sort of the strong roadmap that we're showing, want to get familiar and really work with us earlier in the lifecycle, which I think is, again, positive.

Ross Seymore (Analyst)

Great, thank you for that. I guess as my follow-up, an earlier question, you talked about a little bit below seasonality in the second half of the year for your client business. It seems like there's just, I don't even know if seasonality is a framework that matters. How are you thinking about that for the second half as a whole for client? Gaming was just up a huge amount sequentially in the second quarter. You described a little bit of what you're expecting there. How do you think about seasonality for the second half in its entirety on the gaming side as well?

Lisa Su (CEO and Chair)

Yeah, let me try. Jean might add if you want a little bit more color. The way to think about it is we do expect some growth, sequential growth in clients as we go into the third quarter. I would say sort of single-digit type growth. We continue to see good traction for our products and that portfolio. On the gaming side, I would call it flattish to Q2. We're coming off of such a strong Q2 that I think flattish is actually to be expected. As we go into the fourth quarter, the dynamics that we would see is we would see that the console business would actually be down substantially. Think about it as down strong double digits. The customers usually build for the holiday season sort of before that, and that will be completed by the fourth quarter.

We would expect, as the client and gaming segment, that the segment would probably be down in the fourth quarter. Hopefully that helps. Jean, did you want to add to that?

Jean Hu (CFO, EVP, and Treasurer)

No, I think you covered it.

Lisa Su (CEO and Chair)

OK.

Operator (participant)

The next question comes from the line of Joshua Buchalter with TD Cowen. Please proceed with your question.

Joshua Buchalter (Analyst)

Hey, guys, thank you for taking my question. I wanted to ask about lead times on the Instinct family, both for the MI-350 family and MI-400. I mean, as you move into larger cluster sizes, which it sounds like you're doing at least with Oracle on 350 and then endeavor to do more so on 400, how much visibility and lead time do you need from your customers? I would imagine the lead time for your parts is measured in months. On the infrastructure side, in particular for the larger scale deployments, those things are measured in years at this point. Maybe you could speak to the visibility specifically on the 400. Thank you.

Lisa Su (CEO and Chair)

Yes, sure, Josh. Our lead times are long, given all of the processing steps that we have to go through. Think about it as somewhere between eight, nine months, that type of thing. We have a very, very strong supply chain. We've been preparing for these ramps of both MI-350 series and MI-400 series, and that preparation is ongoing. We feel like we have a very strong supply chain there. In terms of visibility with customers, we're absolutely working with customers very closely on near-term MI-350 series deployments, getting those deployments up as quickly as possible. One of the things about the MI-350 series that is good is that it can go into existing data centers, just given the platform that it is in. We have been certainly working with our customers there.

For the MI-400 series, there are lots and lots of details in sort of full rack scale design implementation. We're actively working with the largest customers right now on just ensuring that our Helios rack is fully compatible with their data center build-outs as we go into 2026. That visibility is important. I think that co-development, co-engineering is important as we get into the rack scale architecture. The ZT team that we've brought in has been extremely, extremely helpful in terms of both internal platform build-out, as well as ensuring that we're working closely with our customers on their data center needs.

Joshua Buchalter (Analyst)

Thank you for that, Lisa. Maybe for Jean, I wanted to follow up on Vivek's question earlier on gross margins. If we add back the charges in Q2, then your gross margin implied in the guidance is roughly flat sequentially. That's despite what's implied to be data center GPUs up meaningfully sequentially. It doesn't seem like consoles falling off in the third quarter. Can you maybe talk to the underlying drivers of how you're able to keep the flat gross margins despite what sounds like still margin-dilutive data center GPUs up significantly within the mix? Thank you.

Jean Hu (CFO, EVP, and Treasurer)

Yeah, Josh, thanks for the question. Yeah, we are guiding our Q3 gross margin around 54%. The Q2, you're right, excluding the $800 million charge, it was close to 54%. I think the gaming business actually is quite high. The mix actually is unfavorable. We do have some tailwinds we have been really driving. First is we have been expanding our server business, which has a really nice gross margin. On the client business side, we are expanding the commercial PC business. That really helps us to drive the gross margin up. In addition, we do have a really strong operational team. They are driving the gross margin improvement just from an operational efficiency perspective across the board. Overall, our objective is to continue to improve gross margin. Despite MI-350's very strong ramp in Q3, we are able to continue to drive the margin up.

Operator (participant)

The next question comes from the line of Joe Moore with Morgan Stanley. Please proceed with your question.

Joe Moore (Analyst)

Great, thank you. You've used this language before of the kind of tens of billions opportunity around MI-400. Can you talk about the time frame when that might occur? Not to pin you down too much, what would help you get to that level sooner rather than later? Should we think of that as a 2027 realistic outcome that you could be looking at $20 billion plus? Just a little bit more color on that tens of billions comment.

Lisa Su (CEO and Chair)

Yeah, I mean, maybe without being specific, Joe, I can give you sort of the way I look at it. Back to this notion of are we incrementally more confident, I think we're seeing a lot of positive signs in our AI customer adoption. I think the strength of the MI-350 series, the very positive feedback that we're getting on MI-400 from customers, the work that we're doing in terms of ensuring that we are fully ready for large-scale deployments of not just inference, but training. I think when we get to tens of billions of dollars, we're talking about significant gigawatt scale type deployments. Those would be important for us to get there. We're certainly, I think, engaged with all of the right customers that could enable that type of ramp.

I won't necessarily speculate on the exact time, other than to say certainly that would be our set of aspirations.

Joe Moore (Analyst)

Great, that's helpful. Thank you. As these workloads evolve, you've sort of talked about inference and training as sort of different opportunities for AMD. Are you seeing those start to come together? It seems like with inference, the reasoning models are requiring much higher complexity. Is rack scale more important to the inference market than you thought it might be? Just any color around how that complexity of inference is impacting you guys.

Lisa Su (CEO and Chair)

Yeah, I think that's absolutely true, Joe. I think with the proliferation of models, what we're seeing is GPUs continue to be very much the computing of choice as you think about all the models that are out there. As you go into distributed inference and some of the newer techniques, we are seeing the importance of the scale-up and scale-out architecture, which we are investing in. I think the overarching thing is we have a very competitive roadmap across the next couple of generations. That has now gotten strong customer validation. We're getting a lot of feedback from customers on where they would like to see us continue to add more resources and add more focus, so that is very helpful. The key is to be a full-scale solution provider for these large customer deployments. That's what we're working on.

Operator (participant)

The next question comes from the line of Aaron Rakers with Wells Fargo. Please proceed with your question.

Aaron Rakers (Analyst)

Yeah, thanks for taking the question. I do have a follow-up as well. I guess the first question is, when we look at the data center guide, Jean, you alluded to double-digit sequential growth. Obviously, the MI-350 series is kind of ramping. I'm curious, how could we conceptualize what you're expecting in the server side? Where do you think your market share is today in traditional enterprise servers outside of cloud?

Jean Hu (CFO, EVP, and Treasurer)

I think when you look at our data center business, we do have a strong double-digit growth. Both server and MI, both sides are growing sequentially. Of course, MI ramp is the most significant one. As far as the server market share, we do think we continue to drive the market share up compared to Q1. It's not a third party has not published a report yet, but we feel really good about the Q2 market share increase versus Q1.

Lisa Su (CEO and Chair)

Aaron, if I just add to that, one of the things that it's important for people to understand is in some of the cloud CapEx numbers that have come out that have been quite positive, that is not only a GPU statement, but there's actually significant CPU CapEx in there as well. We've started to see more robust forecasts going out a number of quarters on the server CPU side because all of that AI content really requires traditional CPUs as well. We are very bullish on the opportunity in servers. I think the team has really executed extremely well. If you look at our portfolio now, Turin and Genoa are very well adopted, broadening workloads. Enterprise adoption is also increasing. I think all of those are positive for the server opportunity in the second half of 2025, as well as going into 2026 and beyond.

Aaron Rakers (Analyst)

Yeah, and then as a follow-up, I'm kind of thinking about the China, the MI-308 opportunity. When we do see a license, I think you alluded to this earlier, it's going to take a little bit of time to kind of ramp and get the supply chain to satisfy the demand. I'm curious, the $800 million write-down that you had taken, is there no kind of finished inventory there? Does that come back? Do you have any reversal aspects of that once a license gets approved?

Jean Hu (CFO, EVP, and Treasurer)

Let me start first, then Lisa can add. First, on the $800 million, majority of them are WIPs. We really don't have the on-the-shelf finished good we can ship immediately. We do need to take time if we get a license.

Lisa Su (CEO and Chair)

Yeah, I think so. [cross talk]

Operator (participant)

The next question comes from the line of C.J. Muse with Cantor Fitzgerald. Please proceed with your question.

C.J. Muse (Analyst)

Yeah, good afternoon. Thank you for taking the question. I guess, Lisa, was hoping you could level set us on the Instinct ramp into 2026. How are you thinking about the timing of the handoff 350 to 400? How are you thinking about Helios contributions? I guess, very importantly, from a customer contribution perspective, how you might see that evolve from traditional hyperscalers to perhaps more sovereign and Neo cloud within the mix?

Lisa Su (CEO and Chair)

Sure, C.J. Certainly, as the second half this year, it's all about MI-350 series ramp into the first half of next year. I think the MI-400 series development is right on track. The development of the Helios platform is also right on track. We would expect significant revenue contribution from Helios in 2026. Relative to the contribution of the various things, hyperscalers, sort of some of the versus Neo clouds versus sovereign, I think it's a little early to really talk about the different pieces, other than to say you would expect that hyperscalers and let's call it Neo clouds that would be working for other large AI natives may be significant pieces of the initial ramp. Sovereign may come a little bit later in time, just given sort of the timing of when different build-outs would happen. Hopefully, that gives you some color, C J.

C J Muse (Analyst)

Yeah, very helpful. Jean, I guess a question for you with the sale of ZT Systems for $3 billion in cash and stock, and you only have $3 billion in debt outstanding. How are you thinking about the use of proceeds? Is there saving for a rainy day or bolt-on acquisitions, perhaps more aggressive share buyback? How are you thinking about it today?

Jean Hu (CFO, EVP, and Treasurer)

Yeah, thanks for the question. Our business model actually generates a lot of free cash flow. As you see in Q2, free cash flow generation was $1.2 billion. If we close the ZT Systems sale, we'll get more cash. Overall, our capital allocation principle continues to be the first thing is investing, especially with the tremendous AI opportunities ahead of us. We will continue to return cash to shareholders. We did a $1.2 billion repurchase in the first half of the year. We are committed to continue to return cash to shareholders through share repurchase.

Matthew Ramsay (VP of Investor Relations and Financial Strategy)

Operator, I think we have time for one more caller, please.

Operator (participant)

OK, the final question comes from the line of Ben Reitzes with Melius Research. Please proceed with your question.

Ben Reitzes (Analyst)

Hey, thanks for squeezing me in here. I wanted to clarify a little bit on the $1 billion increase in sequential sales. It would seem like it's coming from GPUs primarily. I was wondering if you could back that. That's with nothing in China. If the answer to the prior question that GPUs are over $1 billion, that kind of puts you at a $2 billion run rate. I was just wondering if that was accurate in terms of thinking. I have just a very quick follow-up. Thanks.

Jean Hu (CFO, EVP, and Treasurer)

Hi, Ben. Thanks for the question. If you look at the sequential revenue increase, as I mentioned during the prepared remarks, we see data center strong double-digit increase, which includes both GPU and CPU. GPU definitely drives the largest incremental amount increase. We also mentioned the client actually is going to increase sequentially. In addition, the embedded business will return to sequential growth. Multiple businesses contribute to sequential increase, but the majority of increase is really driven by MI350 series, the strong ramp.

Ben Reitzes (Analyst)

OK, great. If indeed that gets you pretty close to a couple billion dollars, if the MI300 comes in, do you see it at the same run rate that you exited? You have the ability to get at that $800 million run rate right away? Do you think it'll take several quarters to ramp when you get the license? Thanks.

Lisa Su (CEO and Chair)

Yeah, Ben, it will take some time to ramp, just given particularly today. I mean, we're sitting already in early August. I don't think you would see a lot of it in Q3. Certainly, as licenses would be approved, we would schedule that, and it would take a little while to ramp.

Matthew Ramsay (VP of Investor Relations and Financial Strategy)

All right, Operator, thank you very much. We appreciate everybody that joined the call today. We'd just like to end the call now. Thank you.

Operator (participant)

Ladies and gentlemen, that does conclude today's teleconference. We thank you for your participation. You may disconnect your lines at this time.