Post Highlights
-
AMD Business Model Explained
-
NVIDIA Q4 FY2026 Earnings Explained
-
TSMC Q4 2025 Earnings Explained
-
How Macro Conditions Affect NVIDIA
-
How to Read an Earnings Report as a Beginner
AMD reported Q4 2025 results on February 4, 2026, for the quarter ending December 27, 2025. This article assumes you already understand how AMD's four-segment business works, why the company is fabless, and what the CUDA vs ROCm software gap means for AMD's competitive position in AI. If you are starting fresh, read the AMD business model breakdown first.
The quarter produced record revenue, record net income, and record free cash flow across all segments. The data center number is the one that matters for anyone trying to understand where AMD is in the AI infrastructure race, and it deserves more than a headline.
The Quarter in One Sentence
Data center revenue grew 39% year over year to a record $5.4 billion. Everything else in this report should be read through that lens.
Q4 2025 Headline Results
| Metric | Q4 2025 | Q3 2025 | Q4 2024 | YoY Change |
|---|---|---|---|---|
| Revenue | $10.270B | $9.246B | $7.658B | +34% |
| GAAP gross margin | 54% | 52% | 51% | +3pp |
| Non-GAAP gross margin | 57% | 54% | 54% | +3pp |
| GAAP net income | $1.511B | $1.243B | $0.482B | +213% |
| Non-GAAP net income | $2.519B | $1.965B | $1.777B | +42% |
| GAAP diluted EPS | $0.92 | $0.75 | $0.29 | +217% |
| Non-GAAP diluted EPS | $1.53 | $1.20 | $1.09 | +40% |
| Adjusted EBITDA | $3.048B | $2.431B | $2.198B | +39% |
| Free cash flow | $2.082B | $1.530B | $1.091B | +91% |
| FCF margin | 20% | 17% | 14% | +6pp |
The GAAP vs non-GAAP gap is large and worth understanding. The difference between GAAP net income ($1.511B) and non-GAAP net income ($2.519B) comes from four main items: stock-based compensation ($486M), amortization of acquisition-related intangibles ($557M), acquisition-related and other costs ($50M), and a $109M loss from discontinued operations related to the ZT Systems manufacturing business AMD divested in Q4. None of these items affect cash directly. FCF of $2.082B is the cleaner read on what the business generated in the quarter.
The GAAP net income jump of 213% year over year looks large relative to the non-GAAP 42% growth. The gap is explained by a Q4 2024 base that included $186M in restructuring charges and a $419M tax provision, both of which normalized in Q4 2025.
Full Year 2025
| Metric | FY2025 | FY2024 | YoY Change |
|---|---|---|---|
| Revenue | $34.639B | $25.785B | +34% |
| Non-GAAP net income | $6.831B | $5.420B | +26% |
| Non-GAAP EPS | $4.17 | $3.31 | +26% |
| Adjusted EBITDA | $8.521B | $6.810B | +25% |
| Free cash flow | $5.519B | $2.405B | +129% |
| FCF margin | 16% | 9% | +7pp |
Free cash flow more than doubled year over year, from $2.405B to $5.519B. That improvement came partly from operating leverage and partly from working capital timing, but the direction is clear: AMD generates meaningfully more cash per dollar of revenue than it did in 2024.
Segment Breakdown
| Segment | Q4 2025 | Q3 2025 | Q4 2024 | YoY | Share of Revenue |
|---|---|---|---|---|---|
| Data Center | $5.380B | $4.341B | $3.859B | +39% | 52% |
| Client | $3.097B | $2.750B | $2.313B | +34% | 30% |
| Gaming | $0.843B | $1.298B | $0.563B | +50% | 8% |
| Embedded | $0.950B | $0.857B | $0.923B | +3% | 9% |
| Total | $10.270B | $9.246B | $7.658B | +34% | 100% |
| Segment | FY2025 | FY2024 | YoY |
|---|---|---|---|
| Data Center | $16.635B | $12.579B | +32% |
| Client | $10.640B | $7.054B | +51% |
| Gaming | $3.910B | $2.595B | +51% |
| Embedded | $3.454B | $3.557B | -3% |
| Total | $34.639B | $25.785B | +34% |
Data Center: The Segment That Defines This Stock
Data center generated $5.380 billion in Q4 2025, up 39% year over year and 24% sequentially. For the full year, the segment produced $16.635 billion, up 32% from $12.579 billion in 2024.
Two drivers split the segment: EPYC server CPUs and Instinct AI GPUs.
EPYC server CPUs. Fifth-generation Turin processors accounted for more than half of server revenue in Q4. Hyperscalers launched over 500 AMD-based cloud instances across 2025, bringing the total EPYC cloud instance count to nearly 1,600, a 50% increase year over year. AWS, Google, and others expanded EPYC deployments throughout Q4. On the enterprise side, the number of large businesses deploying EPYC on-premises more than doubled in 2025. Lisa Su flagged that the server CPU order book strengthened over the last 60 days of the year, and Q1 2026 server CPU guidance is for sequential growth, which is unusual given that Q1 is normally seasonally weak for enterprise hardware.
Instinct AI GPUs. MI350 series GPU revenue set a record in Q4 and is ramping into the first half of 2026. Eight of the top ten AI companies now use Instinct in production workloads. AMD also gained upstream integration in vLLM, one of the most widely used open-source inference engines, expanding the ROCm ecosystem beyond its historically narrow software support.
The figure that changed my read on this segment: Lisa Su disclosed an OpenAI partnership to deploy 6 gigawatts of Instinct GPUs across a multi-year timeline. That is not a hedge against NVIDIA supply shortages. Deploying infrastructure at 6GW scale requires a multi-year architectural commitment. You do not build that and swap it out when a competitor's supply normalizes.
Segment operating margin.
| Segment | Q4 2025 Operating Income | Operating Margin |
|---|---|---|
| Data Center | $1.752B | 33% |
| Client + Gaming | $0.725B | 18% |
| Embedded | $0.357B | 38% |
| All Other | -$1.082B | — |
| Total | $1.752B | 17% (GAAP) |
The "All Other" line absorbs stock-based compensation, intangible amortization, and acquisition costs that are not allocated to segments. Non-GAAP operating income was $2.854B at 28% margin.
Client and Gaming: Record Client, Multi-Year Gaming Headwind
Client revenue hit $3.097 billion in Q4, up 34% year over year and a record. Desktop CPU sales grew for the fourth consecutive quarter. Ryzen AI Max processor launched at CES 2026 with 128GB unified memory capable of running models up to 200 billion parameters locally, extending AMD's push into on-device AI at the edge.
Commercial notebook adoption grew over 40% year over year as enterprises shifted infrastructure buying toward AMD. Lisa Su cited large wins with telecom, financial services, aerospace, automotive, and technology customers in the quarter.
Gaming revenue was $843 million in Q4, up 50% year over year but down 35% sequentially. The sequential decline was expected. Semi-custom SoC sales to console manufacturers are entering the seventh year of the current cycle. AMD guided for a significant double-digit revenue decline in gaming for full-year 2026.
| Gaming Component | FY2025 | FY2026 Outlook |
|---|---|---|
| Semi-custom SoCs (console) | Strong | Significant double-digit % decline |
| Gaming GPU (Radeon) | Growing | Continued growth |
| Next Xbox (AMD SoC) | Development | 2027 launch |
The next Xbox uses an AMD semi-custom SoC and targets a 2027 launch. Steam Machine from Valve is on track to ship early 2026 with AMD hardware. Both are positive for gaming revenue in 2027, but 2026 is a down year for this segment.
Embedded: Stabilizing After Two Years of Inventory Correction
Embedded revenue was $950 million in Q4, up 3% year over year. Design win momentum was strong: AMD closed $17 billion in embedded design wins in 2025, up nearly 20% year over year, bringing the cumulative total since the Xilinx acquisition to over $50 billion.
Channel sell-through accelerated in Q4, led by test, measurement, and emulation end markets. New products launched in the quarter include Versal AI Edge Gen 2 SoCs for low-latency inference and Ryzen X100 series for physical AI and autonomous platforms.
The embedded segment inventory correction that ran through 2023 and 2024 appears to be working through. The design win backlog is the leading indicator for embedded growth in 2026 and 2027.
What Management Said About the Next 12 Months
Lisa Su laid out explicit targets on the Q4 call.
Data center growth target: 60% annual revenue growth over the next three to five years, with AI revenue scaling to tens of billions by 2027.
MI450 and Helios: The next-generation Instinct GPU series and rack-scale platform are on track for a second-half 2026 launch. Multiple OEMs including HPE and Lenovo have already announced plans to offer Helios rack systems. MI450 ramp starts in Q3 2026, with significant volume in Q4 2026 extending into 2027.
MI500 beyond that: Development underway on CDNA 6 architecture, built on 2nm process technology with HBM4E memory. On track for 2027.
Venice server CPU: Next-generation EPYC CPU launching later in 2026. Customer pull described as "very high" with large-scale cloud deployment engagements underway.
China: AMD recognized approximately $390 million in MI308 GPU revenue in Q4 after receiving export licenses for older inventory. Another $100 million is expected in Q1 2026. No additional China revenue is being forecasted due to the uncertain regulatory environment.
Q1 2026 Guidance
| Metric | Q1 2026 Guidance | Q1 2025 Actual | YoY |
|---|---|---|---|
| Revenue | ~$9.8B (+/- $300M) | ~$7.4B | +32% |
| Non-GAAP gross margin~55%~53%+2pp | |||
| Non-GAAP tax rate | 13% | — | — |
The $9.8 billion Q1 midpoint represents a sequential decline of roughly 5% from Q4's $10.3 billion, which is normal seasonality from gaming and client. Jean Hu confirmed that data center revenue is actually expected to grow sequentially in Q1, with both server CPU and data center GPU expected to be up quarter over quarter.
Balance Sheet
| Item | Dec 27, 2025 | Dec 28, 2024 |
|---|---|---|
| Cash and equivalents | $5.539B | $3.787B |
| Short-term investments | $5.013B | $1.345B |
| Total cash + investments | $10.552B | $5.132B |
| Inventories | $7.920B | $5.734B |
| Total assets | $76.926B | $69.226B |
| Long-term debt | $2.348B | $1.721B |
| Short-term debt | $0.874B | — |
| Total debt | $3.222B | $1.721B |
| Net cash (cash minus debt) | ~$7.330B | ~$3.411B |
| Goodwill | $25.126B | $24.839B |
| Acquisition intangibles, net | $16.705B | $18.930B |
AMD is in a net cash position, which is different from Broadcom's net debt position post-VMware. Total debt of $3.222 billion against $10.552 billion in cash and investments gives AMD roughly $7.3 billion in net cash. The balance sheet does not constrain the business.
Inventory grew from $5.734 billion to $7.920 billion, a $2.186 billion increase year over year. Management attributed this to building inventory to support strong data center demand, particularly for MI350 and early MI450 supply chain positioning. High inventory in a growth phase is expected. The risk is if demand weakens faster than inventory was built, which would require write-downs. AMD took inventory write-downs in 2023 during the embedded correction. Worth monitoring.
The $16.705 billion in acquisition-related intangibles is declining as Xilinx and prior acquisition intangibles amortize. That non-cash amortization runs through the income statement at roughly $1 billion per quarter, which explains most of the GAAP vs non-GAAP gap.
Capital Return
| Item | Q4 2025 | FY2025 |
|---|---|---|
| Share repurchases | — | $1.316B |
| Total returned to shareholders | — | ~$1.9B (est.) |
| Shares repurchased | 12.4M (FY) | — |
| Remaining buyback authorization | $9.4B | — |
AMD returned $1.3 billion to shareholders through buybacks in FY2025, repurchasing 12.4 million shares. The remaining $9.4 billion authorization provides meaningful capacity for continued repurchases. Unlike Broadcom, AMD is not returning more cash than it generates. At $5.5 billion in FY2025 FCF against roughly $1.9 billion in shareholder returns, AMD is retaining most of its cash generation and building the balance sheet.
Most coverage of AMD frames this as a NVIDIA competition story. It is not. NVIDIA has CUDA, better training performance, and a software ecosystem that took a decade to build. AMD is not catching that. Trying to beat NVIDIA at its own game is not the strategy here, and Q4 2025 makes that clearer than any previous quarter.
What AMD is actually doing is carving out a parallel lane. The OpenAI deal is the clearest signal. OpenAI is not buying MI450 because H100 supply is tight. They are making a deliberate architectural commitment to a second platform for reasons that come down to workload economics at scale. That is a different thing entirely from being a cheaper fallback.
The market has not fully processed that distinction. AMD's stock still gets discussed as though its AI upside depends on NVIDIA stumbling. It does not. The hyperscaler "train on NVIDIA, infer on AMD" pattern is real and growing, and the OpenAI multi-year deployment suggests it extends beyond inference into training workloads too.
Here is where I think the market is wrong: the OpenAI deal is being priced as confirmed revenue. It is not. It is a deployment execution bet. Six gigawatts of Instinct GPUs across multiple years requires MI450 to ship on schedule, ROCm to work reliably at that scale, and Helios rack infrastructure to perform without the kind of launch problems NVIDIA had with its own rack systems in 2024. Any one of those slips and the revenue timeline moves right by two to three quarters minimum.
The concentration risk is the second thing the market is underweighting. AMD's entire AI revenue story sits inside a handful of hyperscalers right now. Three to five customers are the AI thesis. That is not a platform. That is a customer concentration that introduces earnings volatility the current valuation does not fully reflect. If one major hyperscaler slows MI450 deployment for budget or architecture reasons, AMD's data center GPU line moves materially in the wrong direction with no diversified revenue base to absorb it. That is a real risk, not a tail risk.
The inventory build makes this more pointed. AMD grew inventory from $5.7 billion to $7.9 billion year over year, largely to support MI450 demand. If the ramp slips or demand is overestimated, that inventory becomes a margin compression and write-down problem. AMD has been here before with embedded in 2023. The data center version would be more damaging given the segment's weight in the growth story.
AMD is becoming a genuine second platform, but only inside hyperscaler environments, and only if MI450 executes cleanly. That is a specific claim. It is not the same as saying AMD is a NVIDIA alternative. It means AMD wins a defined lane, not the whole race.
Check By: Q4 2026 earnings. Data center AI revenue above $25 billion annualized AND at least two additional hyperscaler MI450 commitments confirmed publicly. If MI450 slips past Q4 2026, ROCm adoption stays confined to existing hyperscaler relationships, or NVIDIA Blackwell supply normalizes faster than expected and pulls customers back, the second-platform thesis is premature by at least two years.
Frequently Asked Questions
What was AMD's revenue in Q4 2025?
AMD reported $10.270 billion in revenue for Q4 2025, up 34% from $7.658 billion in Q4 2024. Results were reported on February 4, 2026.
How much did AMD's data center segment grow?
Data center revenue was $5.380 billion in Q4 2025, up 39% year over year. For full-year 2025, data center generated $16.635 billion, up 32% from $12.579 billion in 2024.
What is AMD's MI450 and when does it launch?
The MI450 series is AMD's next-generation Instinct AI GPU, part of the Helios rack-scale platform. AMD has guided for a second-half 2026 launch. The ramp starts in Q3 2026 with significant volume in Q4 2026 extending into 2027. AMD has a confirmed partnership with OpenAI to deploy MI450 at scale.
Why did AMD's gaming revenue fall sequentially in Q4?
Gaming revenue fell 35% sequentially in Q4 2025 because semi-custom SoC sales to console manufacturers declined as expected in the late stage of the current console cycle. AMD guided for a significant double-digit percentage decline in gaming revenue for full-year 2026. Recovery is expected when the next Xbox, using an AMD chip, launches in 2027.
What is ROCm and why does it matter?
ROCm is AMD's open-source software stack for running AI and machine learning workloads on Instinct GPUs. It competes with NVIDIA's proprietary CUDA platform. Historically, ROCm has had narrower software support and higher developer friction than CUDA. AMD added ROCm support to vLLM in Q4 2025, which expands compatibility for inference workloads. Whether ROCm can close enough of the CUDA gap to attract enterprise customers without large in-house AI engineering teams is the key software question for AMD's AI business in 2026.
What is AMD's Q1 2026 revenue guidance?
AMD guided Q1 2026 revenue at approximately $9.8 billion, plus or minus $300 million, representing 32% growth year over year. The sequential decline from Q4's $10.3 billion reflects seasonal weakness in client and gaming, partially offset by sequential growth in data center.
What does AMD's balance sheet look like?
As of December 27, 2025, AMD had $10.552 billion in cash and short-term investments against $3.222 billion in total debt, resulting in a net cash position of approximately $7.3 billion. Inventory was $7.920 billion, up from $5.734 billion a year ago, built in anticipation of strong data center demand. AMD's balance sheet does not constrain the business.
What is AMD's long-term revenue target?
Lisa Su guided for data center segment revenue growth of more than 60% annually over the next three to five years, with AI revenue scaling to tens of billions of dollars by 2027. For the full company, AMD's financial analyst day targets include revenue CAGR above 35% over the next three to five years and non-GAAP EPS above $20 in that timeframe.