What Management Said
Read the full Q3 2025 transcript ↗By now, you should have had the opportunity to review a copy of our earnings press release and the accompanying slides. We delivered an outstanding quarter with record revenue and profitability reflecting broad-based demand across our data center AI, server, and PC businesses. Net income rose 31% and free cash flow more than tripled, led by record EPYC, Ryzen, and Instinct processor sales. Our record third-quarter performance marks a clear step up in our growth trajectory as a combination of our expanding compute franchise and rapidly scaling data center AI business drives significant revenue and earnings growth.
Turning to our segments, data center segment revenue increased 22% year-over-year to a record $4.3 billion, led by the ramp of our Instinct MI350 series GPUs and server share gains. Server CPU revenue reached an all-time high as adoption of 5th Gen EPYC Turin processors accelerated rapidly, accounting for nearly half of overall EPYC revenue in the quarter. In cloud, we had record sales as hyperscalers expanded EPYC CPU deployments to power both their own first-party services and public cloud offerings. There are now more than 1,350 public EPYC cloud instances available globally, a nearly 50% increase from a year ago.
Adoption of EPYC in the cloud by large businesses more than tripled year-over-year, as our on-prem share gains are driving increased demand from enterprise customers for AMD cloud instances to support hybrid compute. We expect cloud demand to remain very strong as hyperscalers are significantly increasing their general-purpose compute capacity as they scale their AI workloads. Customer pull and engagement for Venice are the strongest we have seen, reflecting our competitive positioning and the growing demand for more data center compute. Revenue grew year-over-year, driven by the sharp ramp of MI350 series GPU sales and broader MI300 series deployments.
What went well
- Record total revenue of $9.2 billion, up 36% year-over-year and 20% sequentially, exceeding the high end of guidance.
- Data Center segment revenue hit a record $4.3 billion, up 22% year-over-year and 34% sequentially, led by the sharp ramp of Instinct MI350 series GPUs and server share gains.
- Client and gaming segment revenue reached a record $4 billion, up 73% year-over-year, with client revenue a record $2.8 billion (+46% YoY) and gaming revenue $1.3 billion (+181% YoY).
- Free cash flow was a record $1.5 billion (more than tripled) and diluted EPS rose 30% year-over-year to $1.20 from $0.92; operating income was $2.2 billion at a 24% operating margin.
- 5th Gen EPYC Turin ramped rapidly to nearly half of overall EPYC revenue, with server CPU revenue at an all-time high and more than 1,350 public EPYC cloud instances now available (a nearly 50% increase from a year ago).
- Announced a multi-year agreement with OpenAI to deploy 6 GW of Instinct GPUs (first gigawatt of MI450 in H2 2026), which management expects to generate well over $100 billion in revenue over the next few years.
What went wrong
- Q3 results included no revenue from MI308 GPU shipments to China, and the Q4 outlook also excludes any MI308 China revenue given the still-dynamic export-control situation.
- Embedded segment revenue was $857 million, down 8% year-over-year, with operating income falling to $283 million (33% of revenue) from $372 million (40%) a year ago on lower revenue and end-market mix.
- Data Center operating margin compressed to 25% ($1.1 billion) from 29% ($1.0 billion) a year ago, as higher R&D investment partially offset revenue growth.
- Operating expenses rose 42% year-over-year to approximately $2.8 billion due to aggressive R&D and go-to-market investment.
- Q4 guidance implies a decline in the client and gaming segment, with gaming revenue expected to be down strong double digits sequentially.
Guidance Changes
| Metric | Period | Current guidance |
|---|---|---|
| Total revenue | Q4 2025 | Approximately $9.6 billion, ±$300 million (~25% YoY growth at midpoint, ~4% sequentially); excludes any MI308 China revenue |
| Non-GAAP gross margin | Q4 2025 | Approximately 54.5% |
| Non-GAAP operating expenses | Q4 2025 | Approximately $2.8 billion |
| Net interest and other | Q4 2025 | Gain of approximately $37 million |
| Non-GAAP effective tax rate | Q4 2025 | 13% |
| Diluted share count | Q4 2025 | Approximately 1.65 billion shares |
| Segment growth mix | Q4 2025 | Double-digit growth in Data Center (strong server + continued MI350 ramp); client up but gaming down strong double digits; double-digit growth in Embedded |
Performance Breakdown
| Metric | YoY | Note |
|---|---|---|
| Total revenue | +36% ($9.2B) | Broad-based demand across data center AI, server, and PC businesses; record EPYC, Ryzen, and Instinct sales |
| Data Center revenue | +22% ($4.3B) | Ramp of Instinct MI350 series GPUs and 5th Gen EPYC server share gains (also +34% sequentially) |
| Client and Gaming revenue | +73% ($4.0B) | Strong demand for latest client and graphics processors plus stronger console gaming products |
| Client revenue | +46% ($2.8B) | Record Ryzen processor sales and a richer product mix |
| Gaming revenue | +181% ($1.3B) | Higher semi-custom revenue and strong demand for Radeon GPUs |
| Embedded revenue | -8% ($857M) | Soft end-market demand and mix, though up 4% sequentially as certain end markets strengthened |
| Diluted EPS | +30% ($1.20 vs $0.92) | Higher revenue and profitability across the business |
Earnings Call Themes & Trends
| Topic | Previous mention | Current period | Trend |
|---|---|---|---|
| MI308 China export controls | — | No MI308 China revenue recognized in Q3 or assumed in Q4 guide; some licenses received but demand environment still being worked with customers | — |
| MI350 / MI355X Instinct ramp | — | Sharp MI350 series ramp underway; Oracle first hyperscaler to offer MI355X instances; neocloud providers (Crusoe, DigitalOcean, TensorWave, Vultr) ramping; expected to keep ramping into H1 2026 | — |
| MI400 / MI450 series and Helios rack | — | Launching in 2026; Helios integrates MI400 GPUs, Venice EPYC CPUs, and Pensando NICs; strong OCP reception; MI450 online H2 2026 with a sharper ramp | — |
| OpenAI partnership | — | New multi-year 6 GW Instinct deal (first GW MI450 in H2 2026); potential for well over $100 billion in revenue over the next few years; includes warrant structure and joint Triton/software work | — |
| EPYC server / general-purpose compute | — | Turin ramping fast to ~half of EPYC revenue; hyperscalers forecasting significant CPU builds into 2026 as AI spawns general-purpose compute; demand seen as durable and multi-quarter; Venice (2nm) on track for 2026 | — |
| ROCm open software stack | — | ROCm 7 launched with up to 4.6x higher inference and 3x higher training vs ROCm 6; day-zero model support; contributions from Hugging Face, vLLM, SGLang; strengthened by large-customer collaboration | — |
| ZT Systems | — | ZT team is central to Helios development; ZT manufacturing business sold to Sanmina last week (reported as discontinued operations), with Sanmina now lead manufacturing partner for Helios | — |
| AI silicon TAM | ~$500 billion | Management sees the AI compute TAM going up and will provide updated, larger numbers at Financial Analyst Day | — |
Q&A Summary
Vivek Arya (BofA): CPU/GPU mix in Q3/Q4, and can data center AI keep growing in H1 2026 through the MI355-to-MI400 transition or should we expect a pause?
Lisa Su: Very strong Q3 in both server and data center AI (without any MI308). MI355 ramped sharply; server CPU strengthening with multi-quarter visibility. Q4 data center up double digits sequentially in both server and AI. Expect MI355 to keep ramping in H1 2026 and a sharper MI450 ramp in H2 2026.
Vivek Arya (BofA): How much visibility on OpenAI given industry debate about engaging multiple vendors amid power/CapEx constraints, and how does allocation broaden into 2027?
Lisa Su: Very excited about the significant OpenAI relationship; planning multiple quarters out for power and supply chain. First gigawatt deploys H2 2026 with work well underway; planning closely with OpenAI and CSP partners on Helios; good visibility into the MI450 ramp.
Thomas O'Malley (Barclays): On Helios, view on discrete sales versus system sales into next year and when the crossover happens, plus customer response after seeing it at OCP?
Lisa Su: OCP reception was phenomenal with customers bringing engineering teams; interest in MI450/Helios expanded, boosted by OpenAI, OCI, and Meta announcements. Early MI450 customers will center on rack-scale solutions, though other form factors will also be available.
Thomas O'Malley (Barclays): Where will the constraint be for large 2026 deployments—components (interconnect/memory), data center footprint, or power?
Lisa Su: The whole ecosystem must plan together; AMD is working with customers on power plans over the next ~two years and with supply-chain partners on silicon, memory, packaging and components. Feels good about a strong supply chain; expects things to be tight but is well positioned to grow significantly into H2 2026 and 2027.
Joshua Buchalter (TD Cowen): Sustainability of near-term data center CPU strength from AI/agentic workloads, any supply constraints, and should the data center CPU business be aseasonal or normal seasonality in H1 2026?
Lisa Su: CPU demand has broadened, with large hyperscalers forecasting significant 2026 CPU builds as AI requires general-purpose compute; Turin ramping fast alongside strong prior-gen demand; expects a positive, durable, multi-quarter demand environment into 2026. Jean Hu: Have supply to support growth, prepared for the 2026 ramp.
Joshua Buchalter (TD Cowen): Where does ROCm stand competitively, how broad is developer support, and what work remains?
Lisa Su: Great progress with ROCm 7 in performance and framework support, including day-zero model support; most new AMD customers have a smooth onboarding experience. More work remains on newer training/inference and reinforcement-learning workloads; will keep investing heavily.
CJ Muse (Cantor Fitzgerald): Framework for gross margins through calendar 2026 across the 355-to-400 transition and move to full rack scale?
Jean Hu: Not guiding 2026, but data center GPU gross margin typically dips at the start of a new product ramp then normalizes; priority is expanding top-line revenue and gross-margin dollars while also driving gross-margin percentage up.
CJ Muse (Cantor Fitzgerald): High-level view on OpenAI and other large customers and breadth of customer penetration through 2026-2027 (tens of billions in 2027)?
Lisa Su: More detail at Analyst Day, but great traction among the largest customers; OpenAI is extremely important and at multi-gigawatt scale, plus OCI, Department of Energy systems, and many other deep engagements. Expects multiple customers at very significant scale in the MI450 generation.
Stacy Rasgon (Bernstein): In data center, what grew more year-over-year in dollar/percentage terms—servers or GPUs?
Lisa Su: Data center grew nicely YoY in both servers and data center AI. Jean Hu: Directionally similar, but servers a little bit better.
Stacy Rasgon (Bernstein): On Q4 guidance, what does 'strong double digits' for servers mean and is the full-year GPU number still around ~$6.5 billion?
Jean Hu: Guided data center up double digits sequentially, with server up strongly and MI350 continuing to ramp—not the specific figure suggested. Lisa Su: Both server and data center AI will be up sequentially; the 'strong double-digit' comment applied to year-over-year.
Timothy Arcuri (UBS): Has the OpenAI deal influenced AMD's position with other customers, and how much single-customer risk exists if OpenAI becomes ~half of data center GPU revenue in 2027-2028?
Lisa Su: The OpenAI deal plus showing Helios at Open Compute have increased and accelerated customer interest and engagement at higher scale. A broad customer base is a key foundation; dimensioning the supply chain to support multiple customers at similar scale into 2027-2028.
Aaron Rakers (Wells Fargo): On server strength, how to think about unit growth versus ASP expansion through the Turin cycle?
Lisa Su: Turin carries more content so ASPs grow as it ramps, but Genoa demand also continues well since hyperscalers can't move everything to the latest generation immediately; broad-based CPU demand with strong early Venice pull.
Aaron Rakers (Wells Fargo): Updated view on the ~$500 billion AI silicon TAM given large megawatt deployments?
Lisa Su: Will give a full market picture at Analyst Day, but the AI compute TAM is going up—the opportunity now looks larger than the $500 billion figure first discussed.
Antoine Chkaiban (New Street): Is the OpenAI relationship a tailwind to the software stack, and has it made ROCm more robust?
Lisa Su: Yes—all large customers broaden and deepen the software stack; plans to work deeply with OpenAI on hardware, software, systems, and roadmap, with valuable joint work on Triton. Also using AI to accelerate ROCm kernel development.
Antoine Chkaiban (New Street): On GPU useful lives—any early indication CSPs may sweat GPUs longer than the typical 5-6 year depreciation?
Lisa Su: Some early indications of that; there's a desire for the latest GPUs (MI355 into new liquid-cooled facilities, MI450 next), but also strong ongoing use of older generations like MI300X, especially for inference—so a bit of both.
Joe Moore (Morgan Stanley): On MI308, if export relief comes are you ready to ship and how big a swing factor could it be?
Lisa Su: Still a dynamic situation, so no MI308 revenue in the Q4 guide; some licenses received and appreciated, still working with customers on demand and will update in coming months. Has some work in process to support the market but must see how demand shapes up.
Ross Seymore (Deutsche Bank): Amid many multi-gigawatt announcements, how does AMD truly differentiate to win the 6 GW and more?
Lisa Su: The world needs more AI compute and OpenAI is not alone in that demand. MI450 is an extremely strong rack-scale solution for inference and training; keys are time to market, TCO, and deep partnership—including MI500 and beyond—positioning AMD to capture a significant, meaningful piece of the market.
Ross Seymore (Deutsche Bank): Was the OpenAI warrant structure a unique agreement, or is AMD open to similar creative equity vehicles with other customers?
Lisa Su: It was a unique agreement for a unique time in AI, prioritizing deep, multi-year, multi-generation, significant-scale partnership with aligned incentives where AMD, OpenAI, and shareholders all win; expects other creative opportunities (including sovereign AI) but sees OpenAI as fairly unique.
More on Advanced Micro Devices Inc
Reported 2025-11-04 · figures from the Advanced Micro Devices Inc Q3 2025 earnings call.
See how Top Bucket AI works for your firm
Request DemoTop Bucket AI Research
Stay ahead of private markets
Research and market intelligence for private-markets professionals.
✓
You're subscribed.
Thanks for signing up.