What Management Said
Read the full Q4 2025 transcript ↗By now, you should have had the opportunity to review a copy of our earnings press release and accompanying slides. 2025 was a defining year for AMD, with record revenue, net income, and free cash flow driven by broad-based demand for high-performance computing and AI products. Looking at our fourth quarter, fourth quarter revenue grew 34% year-over-year to $10.3 billion, led by record EPYC, Ryzen, and Instinct processor sales. Net income increased 42% to a record $2.5 billion, and free cash flow nearly doubled year-over-year to a record $2.1 billion.
For the full year, revenue grew 34% to $34.6 billion, and we added more than $7.6 billion of data center segment and client revenue. Turning to our fourth quarter segment results, data center segment revenue increased 39% year-over-year to a record $5.4 billion, led by accelerating Instinct MI350 series GPU deployments and server share gains. In server, adoption of 5th-gen EPYC Turin CPUs accelerated in the quarter, accounting for more than half of the total server revenue. As a result, we had record server CPU sales to both cloud and enterprise customers in the quarter and exited the year with record share.
In cloud, hyperscaler demand was very strong as North American customers expanded deployments. EPYC-powered public cloud offerings grew significantly in the quarter, with AWS, Google, and others launching more than 230 new AMD instances. In the enterprise, we are seeing a meaningful shift in EPYC adoption driven by our leadership performance, expanded platform availability, broad software enablement, and increased go-to-market programs. As a result, the number of large businesses deploying EPYC on-prem more than doubled in 2025, and we exited the year with record server sell-through.
What went well
- Record Q4 revenue of $10.3 billion, up 34% year-over-year and 11% sequentially, led by record EPYC, Ryzen, and Instinct processor sales.
- Data center segment hit a record $5.4 billion, up 39% year-over-year and 24% sequentially, driven by the MI350 series GPU ramp and server share gains, with 5th-gen EPYC Turin accounting for more than half of total server revenue.
- Net income rose 42% to a record $2.5 billion and free cash flow nearly doubled year-over-year to a record $2.1 billion; Q4 operating income was a record $2.9 billion at a 28% operating margin and diluted EPS was a record $1.53, up 40%.
- Client business set a record at $3.1 billion, up 34% year-over-year and 13% sequentially, with commercial Ryzen notebook and desktop sell-through up more than 40% year-over-year.
- Q4 non-GAAP gross margin was 57%, up 290 basis points year-over-year (roughly 55% and up 80 bps excluding the MI308 items), helped by favorable product mix.
- Full-year 2025 delivered record revenue of $34.6 billion (up 34%) and record EPS of $4.17 (up 26%), adding more than $7.6 billion of data center and client revenue.
What went wrong
- Gaming revenue fell 35% sequentially on lower semi-custom sales, and management guided 2026 semi-custom revenue down a significant double-digit percentage as the console cycle enters its seventh year.
- Q1 2026 revenue is guided to about $9.8 billion, down roughly 5% sequentially, on seasonal declines in the client, gaming, and embedded segments.
- China MI308 remains a dynamic, uncertain situation: only about $100 million is forecast for Q1 with no additional China revenue assumed beyond that, and MI325 licenses are still pending approval.
- Operating expenses rose 42% year-over-year to $3.0 billion, and an analyst pressed management on OpEx repeatedly guiding up and coming in higher.
- The PC TAM is expected to be down a bit in 2026 with the second half a bit subseasonal to the first, driven by inflationary commodity and memory pricing, and embedded grew only 3% year-over-year with its operating margin slipping to 38% from 39%.
Guidance Changes
| Metric | Period | Current guidance |
|---|---|---|
| Total revenue | Q1 2026 | ~$9.8B ± $300M (incl. ~$100M MI308 China), up ~32% YoY at midpoint, down ~5% sequentially |
| Non-GAAP gross margin | Q1 2026 | ~55% (up 130 bps YoY) |
| Non-GAAP operating expense | Q1 2026 | ~$3.05B |
| Non-GAAP other net income | Q1 2026 | ~$35M |
| Non-GAAP effective tax rate | Q1 2026 | 13% |
| Diluted share count | Q1 2026 | ~1.65B shares |
| MI308 China revenue | Q1 2026 | ~$100M, with no additional China revenue forecast beyond that |
| Semi-custom revenue | FY2026 | Expected to decline by a significant double-digit percentage (7th year of console cycle) |
| Data center segment revenue | FY2026 / 3-5 yr | Long-term target of >60% annual growth is possible in 2026; AI business scaling to tens of billions by 2027 |
Performance Breakdown
| Metric | YoY | Note |
|---|---|---|
| Total revenue | +34% ($10.3B) | Broad-based demand led by record EPYC, Ryzen, and Instinct sales; up 11% sequentially. |
| Data center segment | +39% ($5.4B) | Accelerating MI350 series Instinct deployments and server share gains, with Turin >half of server revenue; up 24% sequentially. |
| Client and gaming segment | +37% ($3.9B) | Strong demand for leadership Ryzen processors, partially offset by lower semi-custom; down 3% sequentially. |
| Client business | +34% ($3.1B) | Strong channel and PC OEM demand plus market share gains; desktop CPU record for a fourth straight quarter; up 13% sequentially. |
| Gaming business | +50% ($843M) | Higher semi-custom revenue and strong Radeon RX 9000 GPU demand over the holidays; down 35% sequentially on lower semi-custom. |
| Embedded segment | +3% ($950M) | Improving end-customer demand led by test, measurement, and emulation plus aerospace; up 11% sequentially. |
Earnings Call Themes & Trends
| Topic | Previous mention | Current period | Trend |
|---|---|---|---|
| Instinct MI350/MI355 ramp | — | Record Instinct GPU revenue in Q4 led by MI350 series ramp; 8 of the top 10 AI companies now use Instinct in production; MI355 done well and continues ramping in 1H26. | — |
| MI400/MI450 series and Helios | — | Described as a major inflection point; second-half 2026 launch on track, revenue starting in Q3 and ramping significant volume in Q4 into 2027; multiple OEMs (HPE, Lenovo) to offer Helios racks. | — |
| OpenAI partnership | — | Multi-generation deal to deploy 6 GW of Instinct GPUs is on schedule to start in 2H26; active discussions with additional customers for at-scale multi-year MI450/Helios deployments. | — |
| China / MI308 export controls | — | ~$390M of MI308 China revenue in Q4 (from a license approved on early-2025 orders, not in guidance); only ~$100M forecast for Q1 and nothing beyond given the dynamic situation; MI325 licenses submitted. | — |
| EPYC server franchise / Venice | — | Record server CPU sales and record exit share; hyperscalers launched >500 AMD instances in 2025 (nearly 1,600 EPYC cloud instances, +50% YoY); next-gen Venice launching later in 2026 with high customer pull. | — |
| ZT Systems | — | Sale of ZT Systems manufacturing business to Sanmina closed in late October; results reported as discontinued operations and excluded from non-GAAP financials. | — |
| OpEx and operating leverage | OpEx grew as revenue rose in 2025 | OpEx up 42% YoY to $3.0B in Q4; management expects leverage in 2026 with OpEx growing slower than revenue, especially as 2H revenue inflects. | — |
| Gross margin trajectory | — | Favorable mix (Turin, MI355, richer client stack, embedded recovery) supports margins; each Instinct generation should carry higher margin long-term though early ramps dilute; MI450 ramp in Q4 to be mix-driven. | — |
Q&A Summary
Wells Fargo asked about MI450/Helios customer engagements and the shape of the 2027 AI ramp given the high-$20-billion 2027 expectation.
Su said MI450 development is going extremely well and on track for a second-half launch and start of production; the OpenAI ramp starts 2H26 into 2027, other customers are eager to ramp MI450 quickly across inference and training, and AMD feels very good about reaching tens of billions of data center AI revenue in 2027.
UBS asked for detail under the March guide and how data center GPU ramps through the year (investors modeling ~$14B).
Hu said Q1 is down ~5% sequentially but data center is up, with CPU revenue guided up sequentially versus a normally seasonally down quarter and GPU revenue (including China) also up; Su added server CPU strength continues all year and MI450 revenue starts in Q3 and ramps significant volume in Q4 into 2027.
Bank of America asked about China MI308 assumptions beyond Q1 and whether data center can grow at the 60%+ target in 2026.
Su said only ~$100M of China MI308 is forecast in Q1 with nothing beyond given the dynamic situation (MI325 licenses submitted); on data center she said the >60% long-term target is certainly possible in 2026, driven by EPYC (Turin/Genoa), Venice in 2H, and the significant MI450 ramp.
Cantor asked about sourcing incremental server CPU capacity from TSMC amid tightness and pricing implications.
Su said the server CPU TAM should grow strong double digits in 2026; AMD has increased server CPU supply capacity over recent quarters (enabling the higher Q1 guide) and is working with supply chain partners to keep growing supply throughout the year.
Cantor also asked Hu for a gross margin framework through the year as server CPU strengthens and GPU accelerates in 2H.
Hu said Q1 gross margin of ~55% is up 130 bps YoY despite the MI355 ramp, helped by favorable mix across data center (Turin, MI355), an improving client stack, and embedded recovery; when MI450 ramps in Q4 margin will be mix-driven, with more color to come.
Morgan Stanley asked whether MI455 will be 100% racks or include an eight-way server, and when revenue is recognized.
Su said there are multiple MI450 variants including an eight-way GPU form factor, but for 2026 the vast majority will be rack-scale solutions, and AMD recognizes revenue when it ships to the rack builder.
Morgan Stanley followed up on execution risk turning silicon into racks, referencing a competitor's issues.
Su said MI450 and Helios rack development is right on track with extensive rack-scale and silicon-level testing already done, much of it in parallel using customer input, and AMD expects to be on track for the second-half launch.
Bernstein asked how to think about OpEx ramp and whether leverage comes as GPU revenue inflects.
Su said AMD leaned into OpEx in 2025 for the right reasons but expects clear operating leverage in 2026 with OpEx growing slower than revenue, especially in the second half as revenue inflects.
Bernstein also asked if the $100M Q1 China revenue carries a zero-cost basis / margin headwind, and for the full-year 2025 Instinct number.
Hu said the $360M Q4 inventory reserve reversal already covers the $100M of MI308 to ship in Q1, so the Q1 gross margin guide is clean; Su declined to give the annual Instinct figure but noted data center AI grew from Q3 to Q4 even backing out the non-recurring China revenue.
TD Cowen asked whether client order patterns changed given memory inflation/pull-ins and the 2026 client outlook.
Su said client performed extremely well in 2025 on ASP mix-up and unit growth; for 2026 the PC TAM is likely down a bit with 2H a bit subseasonal on commodity/memory inflation, but AMD still expects to grow its PC business by focusing on enterprise and the premium high end.
TD Cowen also asked about competitive implications of SRAM/ASIC-based low-latency inference architectures.
Su framed it as expected AI-market maturation where tokens-per-dollar efficiency matters more; AMD's chiplet architecture lets it optimize across inference, training, and inference stages, and it has the full compute stack to serve workload-optimized products alongside continued inference and training focus.
Melius asked whether the OpenAI 6 GW deal and 3.5-year timeline remain on track for 2H.
Su said AMD is working in deep co-development with OpenAI and CSP partners, the ramp is on schedule to start in 2H, MI450 and Helios are doing well, and she reminded listeners a broad set of other customers is also excited to ramp MI450 in that timeframe.
Melius also asked about x86 versus ARM in agentic workloads and NVIDIA selling a standalone ARM CPU.
Su said high-performance CPUs are in great need for agentic workloads that spin off many traditional CPU tasks, the vast majority on x86 today; EPYC's workload optimization (best cloud, enterprise, and lower-cost storage variants) positions it well for the multi-year CPU cycle.
Barclays asked about the timing of memory/HBM procurement.
Su said AMD works with suppliers over a multi-year timeframe given HBM and wafer lead times, has been planning this CPU and GPU ramp for a couple of years independent of current conditions, and is now signing multi-year agreements extending further given supply tightness.
Barclays also asked about the evolution of AMD's system-based architecture versus competitors' KV-cache offload / discrete ASIC approaches.
Su emphasized there is no one-size-fits-all; the flexible chiplet and platform architecture lets AMD build different system solutions, with rack-scale best for the highest-end distributed inference and training while other form factors serve enterprise AI, and AMD invests across that spectrum.
Deutsche Bank asked whether gross margin changes across the MI300-to-400-to-500 transitions.
Hu said each generation adds more capability and memory so margin should progress higher over the long term, though early ramps tend to be lower before yield, test, and performance improvements lift margin within a generation.
Deutsche Bank also asked for the magnitude of the 2026 gaming decline given 2025 grew 50% versus an expected flat.
Hu said 2026 is the seventh year of the current console cycle, so semi-custom revenue is expected to come down significantly double-digit; Su added that a reversal is expected as the next generation (e.g., next-gen Xbox in 2027) ramps.
Goldman Sachs asked whether rack-scale supply constraints could limit 2H revenue growth (Q3 to Q4).
Su said AMD is planning at every component level and does not believe the data center AI ramp will be supply-limited; it is an aggressive but doable ramp and ensuring data center GPU and CPU ramps go well is the priority.
Goldman Sachs also asked about the largest 2025 investment areas and incremental 2026 OpEx priorities.
Su cited data center AI hardware roadmap acceleration, expanded software, the ZT Systems acquisition, and go-to-market/commercial CPU expansion as 2025 priorities; 2026 continues aggressive investment but with revenue expanding faster than OpEx to drive EPS expansion.
More on Advanced Micro Devices Inc
Reported 2026-02-03 · figures from the Advanced Micro Devices Inc Q4 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.