AMD (Minutes): Even excluding MI308, next quarter's AI GPU will still see year-over-year growth.

The following are the minutes of AMD's Q2 2025 earnings call organized by Dolphin Research. For earnings interpretation, please refer to AMD: CPU Dominates Intel, When Will AI GPU Compete with NVIDIA?

I.$AMD(AMD.US) Key Financial Highlights

1. Overall Performance:

a. Revenue: $7.7 billion, exceeding guidance midpoint, up 32% year-over-year; up 3% quarter-over-quarter (growth in client and gaming segments offsetting some decline in data center revenue). Record sales of Ryzen and EPYC processors, increased shipments of semi-custom products, offset the impact of U.S. export controls on MI308 sales to China.

b. Profitability (non-GAAP): Gross margin 43% (down 10 percentage points year-over-year) due to $800 million inventory and expenses related to export controls; excluding this, non-GAAP gross margin is 54%. Operating expenses $2.4 billion (up 32% year-over-year) for market expansion and AI R&D investment. Operating income $897 million (margin 12%), down year-over-year mainly due to inventory expenses; diluted EPS $0.48 (inventory expenses dragged down $0.43 per share).

2. Business Segments:

a. Data Center: Revenue $3.2 billion (up 14% year-over-year), benefiting from EPYC CPU and cloud/enterprise market share growth; down 12% quarter-over-quarter (affected by MI308 export controls); operating loss $155 million (compared to $743 million profit in the same period last year).

b. Client and Gaming: Revenue $3.6 billion (up 69% year-over-year, up 20% quarter-over-quarter). Operating income for this segment $767 million (margin 21%), significantly improved year-over-year.

- Client: Revenue $2.5 billion (up 67% year-over-year), a record, due to Ryzen desktop CPU sales and product mix optimization.

- Gaming: Revenue $1.1 billion (up 73% year-over-year), driven by new GPU demand and semi-custom business (console inventory recovery + holiday season stocking).

c. Embedded: Revenue $824 million (down 4% year-over-year, flat quarter-over-quarter), with demand differentiation; operating income $275 million (margin 33%), down year-over-year due to product mix.

3. Financial Status:

a. Operating cash flow $1.5 billion, free cash flow $1.2 billion; $478 million this quarter, $1.2 billion cumulative in the first half; board authorized an additional $6 billion repurchase, remaining authorization $9.5 billion; cash and short-term investments $5.9 billion, long-term debt $3.2 billion; repaid $950 million commercial paper for ZT Systems acquisition.

b. ZT Systems Acquisition: Acquisition completed, plans to sell its manufacturing business to Sanmina for $3 billion (expected completion by end of 2025), this business is classified as discontinued operations.

4. Outlook:

a. Q3: Revenue expected to be approximately $8.7 billion (±$300 million), up 28% year-over-year, up 13% quarter-over-quarter. Excludes MI308 shipments to China (license pending U.S. government review).

- Data center segment (MI350 series GPU mass production) to grow double digits quarter-over-quarter; client and gaming business to grow moderately (client business growth, gaming roughly flat quarter-over-quarter); embedded to resume growth.

b. Non-GAAP gross margin 54%, operating expenses $2.55 billion. Net interest and other expenses approximately $10 million, effective tax rate 13%, diluted shares 1.63 billion.

II. Detailed Content of AMD Earnings Call

2.1 Key Information from Executive Statements

1. Q2 Business Performance:

a. Record revenue of $7.7 billion, up 32% year-over-year, exceeding the midpoint of guidance.

b. Excluding the $800 million inventory write-down related to data center AI export controls, gross margin was 54%, achieving the sixth consecutive quarter of year-over-year gross margin expansion, thanks to a better product mix.

c. Growth in EPYC and Ryzen processor sales offset the negative impact of export controls on Instinct sales.

2. Data Center Segment: Revenue grew 14% year-over-year to $3.2 billion.

a. EPYC Processor Performance: Strong demand in cloud services, enterprise workloads, and emerging AI scenarios, with AI applications driving additional demand for general computing infrastructure. Fifth-generation shipments increased significantly, with stable demand for previous generations; record sales of cloud and enterprise CPUs, with market share growing year-over-year for 33 consecutive quarters.

- Cloud Services: Increased adoption by hyperscale enterprises for mission-critical tasks; over 100 new AMD-powered cloud instances added (e.g., new products from Google Cloud, Oracle Cloud, with doubled performance), nearly 1,200 globally; significant growth in enterprise cloud adoption, with partnerships with dozens of large enterprises; increased adoption by telecom customers (e.g., deployments by KDDI, Nokia).

- Enterprise On-Premises: 28 new platforms launched by HPE, Dell, with standout advantages; significant growth in enterprise deployments, with new partnerships in multiple fields; launched EPYC 4005 series for SMBs and managed IT service customers.

- HPC Field: AMD supports over 1/3 of the world's fastest supercomputers, including Frontier, El Capitan; holds 12 spots in the top 20 of the Green 500, highlighting EPYC and Instinct's per-watt performance advantage in large-scale deployments.

b. Data Center AI Business: Revenue declined year-over-year due to MI308 export restrictions and transition to MI350 series. Solid progress with MI300, MI325, with expanded collaboration and adoption; seven of the top ten global model builders and AI companies use Instinct.

- Instinct MI350 Series: Industry-leading memory bandwidth and capacity, widely adopted by hyperscale enterprises; MI355 performance benchmarked against B200/GB200, with significant cost advantage in large-scale inference (40% more tokens per dollar); supports coordinated deployment of CPU, GPU, and NIC (e.g., Oracle clusters); early production in June, with significant production increase in the second half of the year.

- Sovereign AI Collaboration: Accelerating with multiple governments, such as a multi-billion-dollar collaboration with HUMAIN, with deployments initiated in key regions, planning quarterly expansion; over 40 active projects globally, with significant potential.

- AI Software: Rackham 7 performance improved by over 3x, with expanded features; launched nightly build versions and developer cloud, expanding framework support; released Rackham Enterprise AI full-stack platform.

Next-Generation Planning: MI400 series development progressing smoothly, with FP4 performance reaching 40 petaflops, leading competitors by 50% in memory, etc.; launched Helios rack platform, expected to improve efficiency by 10x, launching in 2026.

- Acquisitions and Investments: Acquired Brium, Lemini to strengthen software, acquired ZT Systems for rack design capabilities, with Sanmina becoming a major manufacturing partner.

- Regulatory Environment: U.S. Department of Commerce advancing MI308 export license review to China, resumption of shipments depends on demand and supply chain; Q3 guidance excludes MI308 revenue, as MI350 procurement increases, Instinct revenue expected to grow year-over-year.

3. Client and Gaming Segment: Driven by record client CPU sales and strong demand for semi-custom gaming console SoCs and Radeon GPUs, segment revenue grew 69% year-over-year to $3.6 billion.

a. Client Business: Revenue $2.5 billion (up 67% year-over-year), mainly due to record desktop CPU sales. Strong demand for Ryzen 9000 series (especially X3D processors), with record desktop channel sales, consistently topping major global e-commerce bestseller lists; launched 96-core Zen 5 Threadripper processors, with performance for content creation and other workloads twice that of competitors. Future revenue growth is expected to continue leading the market, benefiting from increased product adoption, commercial market efforts, and a premium product mix.

- Mobile: AMD-driven notebook sales grew significantly year-over-year, with expanded high-end market share, leading performance in general and AI workloads with Ryzen AI300.

- Commercial: Accelerated Ryzen adoption, with OEM procurement up 25% year-over-year, strong sales from Lenovo, HP, Dell, with new partnerships with several Forbes 2000 companies.

b. Gaming Business Revenue $1.1 billion (up 73% year-over-year). Console inventory recovery + holiday season stocking drove significant year-over-year revenue growth, with next-gen Xbox collaboration with Microsoft, deepening machine learning collaboration with Sony through the "Amethyst Project." Strong demand for Radeon 9000 series GPUs, with desktop sales outstripping supply; launched 9600 XT (double ray tracing performance) and AI Pro R9700 (supports local inference, larger memory, more scalability).

4. Embedded Segment: Revenue declined 4% year-over-year to $824 million. Demand continues to gradually recover, with sales rebounding in Q2, but strong performance in most markets was offset by weakness in a few areas and inventory reduction actions by industrial customers.

a. Expanded embedded product portfolio, with first mass production shipments of Spartan UltraScale+ FPGA, providing leading performance and advanced security for cost-sensitive low-power applications; Versal adaptive SoC adoption continues to expand in high-end applications, such as Bosch's next-gen autonomous taxi platform developed in Europe.

b. Future Outlook: Expected improvement in test and measurement, communication, and aerospace market demand in the second half of 2025 to drive quarter-over-quarter growth;

5. Outlook:

Strong demand across the company's full product line, with significant growth expected in the second half of the year driven by increased MI350 series production and EPYC market share gains. Accelerated development in server and PC CPU businesses, benefiting from high-performance computing demand growth, market share gains, premium product mix, and market investment. Embedded and gaming businesses returning to growth with strong design orders, laying a solid foundation for the long term. In the AI field, increased adoption of MI350 series and Rackham 7 has clarified a multi-billion-dollar annual revenue growth path; next-gen MI400 series planned for large-scale deployment in 2026, with strong customer interest. The early stage of industry-wide AI transformation will drive step-change growth in computing demand across markets, laying the foundation for future revenue and earnings growth. In the long term, the momentum of design wins continues to strengthen. As of now, it has improved compared to the same period last year, with the potential to exceed the $14 billion record set in 2024.

2.2 Q&A

Q: Regarding the client business, you previously expected the second half to be roughly flat with the first half, due to the need to guard against some demand being pulled forward. Did Q2 results include some demand being pulled forward? Will the second half remain flat? In the long term, Intel mentioned 18A, which may trigger an immediate reaction. What does this mean for AMD's long-term market share in the ASBs field?

A: Our client and gaming segments (especially the client business) performed strongly in the first half, with 68% year-over-year growth, with all segments contributing: strong gaming and desktop channel sales, leading X3D series GPUs, rising AI application rates, and growing enterprise sales.

Regarding demand being pulled forward, we believe it is minimal, with end-user consumption remaining strong. The main driver of Q3 growth is the data center (MI350 mass production), with expected growth in the client business, though slightly below seasonal levels, but overall performance is strong, and we continue to gain share in key areas.

Revenue growth in the client business in Q1 and Q2 was largely driven by increased average selling prices, a result of our focus on the high-end market with our product mix. Currently, our share in the enterprise business is still relatively low, which is a key area of investment for us, and we have seen progress in enterprise PCs from HP and Lenovo, with Dell also joining, and further growth expected in the second half. These will be growth drivers for the client business in the second half of 2025 and multiple quarters beyond.

Q: Regarding the China market, the guidance does not include MI308 revenue. If export licenses are obtained, is the supply chain ready? What is the current inventory situation? If MI308 shipments resume, how do you see its contribution compared to the $700 million in Q2 and the $800 million mentioned in April for the second half?

A: We are pleased with the progress made with government departments in recent months, working closely to support the global application of U.S. technology, with China being an important market. Several licenses are currently under review, and we are cooperating with the Department of Commerce to advance this. If licenses are approved, we will initiate MI308 shipments. On the supply chain side, most inventory is work-in-progress, requiring several quarters to process. Revenue contribution depends on the timing of approval, overall situation is better than 90 days ago. We hope to serve the China market with MI308 and are working closely with government departments on this.

Q: The sovereign AI project opportunities for AMD in 2026 could become quite significant. How should we assess its scale? What does the joint venture structure related to some contracts mean? Is this incremental to the current MI business growth rate, or does it replace existing business? How should we measure the incremental opportunities from sovereign customers in 2026? Does it depend on MI400? How should we view AMD's sovereign AI projects next year?

A: We are excited about the AI opportunities brought by MI355 and MI400 series in the second half of this year and 2026, with significant opportunities at the hyperscale enterprise, leading AI company, and sovereign nation levels, with active collaboration in the sovereign field. Sovereign projects need to address some regulatory issues, and we are working closely with government departments. Clearly, countries want to have autonomous computing capabilities, which is a global trend. For example, our collaboration with Saudi Arabia's HUMAIN will start with MI355 and continue to advance, with our open ecosystem being well-received by sovereign nations.

This is an incremental opportunity, important for both MI355 and MI400 series in the future.

Q: MI products initially have a dilutive effect on gross margins. What sales scale is needed to start contributing positively to gross margins? Assuming Q4 sales growth rate is similar to Q3, will gross margins remain at Q3 levels, or are there other factors that could impact gross margins?

A: MI product gross margins are slightly below the company average. Currently, we prioritize capturing larger, faster-growing revenue opportunities by providing customers with better TCO to expand market share. MI gross margins are influenced by customer, product generation, and other factors, with the operations team continuously improving efficiency to enhance gross margins, which is a long-term trend, not necessarily tied to quarterly revenue, and will continue to improve in the future. Overall, AI is one of the fastest-growing market opportunities, and we are more focused on capturing as much absolute gross profit as possible.

Q: Data center GPU business in the June quarter seems to slightly exceed $1 billion, with strong growth expected in the second half. What does this specifically mean? Can it reach $7 billion for the year? Can you provide some reference metrics for Q3 expectations?

A: We mentioned strong growth from Q2 to Q3, MI355 was in production and shipped in small quantities in the June quarter, with significant growth starting in Q3, and expected continued growth in Q4, with year-over-year improvement. Customer demand for MI355 is positive, with greater competitiveness compared to MI300's initial small-scale deployments, with customers willing to use it on a large scale. MI355 has strong inference performance, and we are also working with customers on training business, paving the way for the MI400 series in 2026. We are confident in MI355 and are advancing according to the product roadmap.

Q: The new developer cloud involves renting back some computing power sold to cloud providers and emerging cloud service providers. Will this be an important part of MI355 revenue in the second half of this year? How much demand will this stimulate? What is the ultimate goal of this cloud service?

A: The purpose of the developer cloud is to allow developers to easily use AMD Instinct GPUs. Looking back at the MI300 series, we focused on large hyperscale enterprises and major customers, but many other customers were also interested in our GPUs, just hoping for easier access. So the developer cloud provides ready-to-deploy containers for training and inference without long-term commitments.

It will not have a significant impact on revenue in the second half of this year, but it will allow more customers to experience AMD products. The larger revenue opportunity comes from large customer adoption and expanded deployment, which we are pushing to land as soon as possible. Additionally, the MI355 infrastructure is similar to MI300, allowing for quick and smooth customer adoption, which is also its advantage.

Q: MI355 will grow significantly in the second half, with year-over-year growth in Q3, similar to statements before Q1, excluding MI308. Given the slightly earlier launch of the MI350 series, how has its adoption changed compared to before Q1? Is the current growth exceeding, meeting, or falling short of previous expectations?

A: The adoption speed of MI355 is faster than expected. When we launch products, we complete comprehensive validation with customers, and the market has shown broad interest, with many new customers added in the past 90 days, which is very positive. Additionally, there is anticipation for the results of MI400 and the Helios rack. Given our strong product roadmap, many customers want to collaborate with us earlier in the product lifecycle, which is also a positive signal.

Q: You mentioned that the client business will perform slightly below seasonal levels in the second half. How do you view the overall situation of the client business in the second half? The gaming business grew significantly quarter-over-quarter in Q2, and you previously described some expectations. How do you view the overall seasonal performance of the gaming business in the second half?

A: Client Business: Expected single-digit quarter-over-quarter growth in Q3, with positive market response to products and mix.

Gaming Business: Q3 is expected to be roughly flat with Q2 (Q2 performed strongly, flat is expected). Q4 console business will decline significantly (double-digit decline), as customers typically complete stocking before the holiday season, with stocking ending in Q4. Therefore, the client and gaming business segments may decline in Q4.

Q: What is the delivery cycle for the Instinct series (MI350 and MI400)? As you move to larger cluster scales (e.g., MI350 collaboration with Oracle, and planned MI400), how much visibility and lead time do customers need to provide? What is the specific visibility for MI400?

A: Our delivery cycle is relatively long, about 8-9 months, due to multiple processing stages. Our supply chain is strong, and we have been preparing for capacity increases for the MI350 and MI400 series, with preparations ongoing, and we are confident in this.

In terms of customer demand visibility, we are working closely with customers to advance the MI350 series for recent deployments, aiming for quick implementation. The advantage of MI350 is that it can be directly deployed to existing data centers, and we are actively cooperating with these customers. The MI400 series involves full rack-scale design and implementation details, and we are currently working with major customers to ensure compatibility with their data center construction when the Helios rack is put into use in 2026. This demand visibility is important, as rack-scale architecture requires joint development and design, and the ZT team we introduced plays a significant role in internal platform building and close collaboration with customers on data center needs.

Q: If the Q2 expenses are added back, the implied gross margin in the guidance is roughly flat quarter-over-quarter, despite significant growth in data center GPU quarter-over-quarter, and no decline in Q3 console business. How do you explain the drivers behind this flat gross margin? How is gross margin maintained flat with a significant increase in data center GPU share (which seems to dilute gross margin)?

A: We expect Q3 gross margin to be around 54%, with Q2 excluding the $800 million expense also close to 54%. Gaming business gross margins are high, and the current business mix is actually unfavorable.

But we have drivers: first, expanding the server business, which has good gross margins; second, expanding the client commercial PC business, which helps improve gross margins; additionally, a strong operations team is driving gross margin improvement from an overall operational efficiency perspective. Therefore, despite strong growth in MI350 in Q3, we are still expected to continue improving profitability.

Q: What time frame might the multi-billion-dollar opportunity related to MI400 be realized? What factors help accelerate reaching this scale? Is it realistic to achieve a scale of over $20 billion by 2027?

A: We see many positive signs in AI customer adoption, including strong performance of the MI350 series, positive feedback on MI400, and our preparations to support large-scale deployments (including inference and training). The multi-billion-dollar scale corresponds to gigawatt-scale large deployments, which is critical. We are working with the right customers to drive growth, but we are not speculating on specific timing, which is undoubtedly our vision.

Q: As workloads evolve, are inference and training opportunities beginning to converge? In the inference field, as inference model complexity increases, is rack-scale more important to the inference market than expected? How does the complexity of inference affect AMD?

A: As models increase, GPUs remain the preferred computing tool for various models. In distributed inference and new technology applications, the importance of vertical and horizontal scaling architectures is highlighted, and we are increasing investment. More importantly, our future product roadmap is highly competitive and has been recognized by customers. Customer feedback also guides us on where to invest more resources and focus, which is helpful. The key is to become a comprehensive solution provider for large customer deployments, which is our direction.

Q: Data center business is expected to grow double digits quarter-over-quarter, and the MI355 series is improving. How should we understand the expectations on the server side? What is the current market share in traditional enterprise servers (excluding the cloud field)?

A: Data center business achieved strong double-digit quarter-over-quarter growth, with both server and MI growing quarter-over-quarter, with MI having the most significant increase. Server market share continued to increase from Q1, and although third-party reports have not yet been released, we are confident in the share growth in Q2.

Cloud capital expenditure data is positive, involving not only GPUs but also a large amount of CPU expenditure. We see stronger server CPU forecasts for multiple quarters in the future, as AI content also requires traditional CPU support, so we are optimistic about server opportunities. The team has excellent execution, with high adoption rates for Turin and Genoa in the current product mix, expanding applicable workloads, and increasing enterprise adoption, all of which are favorable for the server business in the second half of 2025 and beyond.

Q: In the MI308 opportunity in the China market, if licenses are obtained, does the previous $800 million write-down involve finished goods inventory? If licenses are approved, is there a possibility of reversal or recovery of this write-down?

A: Most of the $800 million is work-in-progress, with no finished goods ready for immediate shipment, so even if licenses are obtained, time is needed for preparation.

Q: How is the growth of the Instinct series in 2026? What is the timeline for transitioning from MI350 to MI400? How will Helios' contribution be reflected? From a customer contribution perspective, how will the share of traditional hyperscale vendors and sovereign cloud, new cloud vendors change?

A: The focus from the second half of this year to the first half of next year is on increasing MI355 production capacity. MI400 series and Helios platform development is progressing smoothly, with significant revenue contribution expected from Helios in 2026.

Regarding the contribution of various customers, it is too early to talk about specific proportions, but hyperscale vendors and new cloud vendors serving large AI-native enterprises may play an important role in the initial growth phase, with sovereign cloud contributions possibly coming later, related to the construction progress in different fields.

Q: After selling ZT for $3 billion in cash and stock, the company has existing debt of $3 billion. How do you plan to use the proceeds? Will it be kept as a reserve, used for strengthening acquisitions, or increased stock repurchases? What are your current thoughts?

A: Our business model generates a lot of free cash flow, reaching $1.2 billion in Q2. If the ZT sale is completed and more cash is obtained, the capital allocation principle remains: prioritize investment (especially in the huge AI opportunity), and then continue returning cash to shareholders. We have already repurchased $1.2 billion in stock in the first half of this year, and we are committed to continuing to return cash through repurchases.

Q: Is the $1 billion quarter-over-quarter revenue increase mainly from GPUs, excluding contributions from the China market? If GPU sales exceed $1 billion, does this mean an annualized sales rate of $2 billion, is this understanding accurate?

A: In the quarter-over-quarter revenue growth, the data center business (including GPU and CPU) showed strong double-digit growth, with GPU being the largest increment. Client business is expected to grow quarter-over-quarter, and embedded business will also resume quarter-over-quarter growth. Multiple businesses contributed, but the main growth came from the strong increase in MI355.

Q: If this is the case, sales are approaching billions of dollars. How has the company's sales growth performed after the launch of the MI350 series? What is the impact of licenses?

A: Capacity increases take time. It is now early August, and there will not be much improvement in Q3. But if licenses are obtained, we will plan, but it still takes some time.

<End Here>

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