Advanced Micro Devices ($AMD)
AI Compute | EPYC Server CPUs · Instinct AI GPUs · Helios | FY2025 Revenue: $34.6B
Sources: SEC EDGAR 10-K (FY2025, filed Feb 2026; FY2024), Q4 2025 Earnings Release & Call (Feb 3, 2026), Q1 2026 Earnings Release & Call (May 5, 2026). CIK: 0000002488
Company Overview & Thesis
Advanced Micro Devices, Inc. (NASDAQ: $AMD) is a fabless semiconductor company headquartered in Santa Clara, California, founded in 1969. AMD designs high-performance CPUs, GPUs, FPGAs, and adaptive SoCs. All wafer fabrication is contracted primarily to TSMC. AMD operates in three reportable segments: Data Center (EPYC server CPUs, Instinct AI GPUs, Pensando networking), Client and Gaming (Ryzen desktop/laptop CPUs, Radeon GPUs, semi-custom SoCs for consoles), and Embedded (FPGAs, adaptive SoCs, and embedded x86 processors, primarily from the 2022 Xilinx acquisition for $48.8B).
AMD reported record revenue of $34.6B (+34% YoY) in FY2025 and $10.3B (+38% YoY) in Q1 2026. Data Center segment revenue of $16.6B (+32% YoY) in FY2025 now represents nearly half of total revenue. Management guided for Q2 2026 revenue of ~$11.2B (+46% YoY) — driven by accelerating EPYC CPU and Instinct GPU demand — and has outlined ambitions to grow data center AI revenue to tens of billions annually by 2027.
The core thesis: AMD is executing a two-pronged AI infrastructure strategy. First, EPYC server CPUs are displacing Intel Xeon across cloud hyperscalers and enterprise data centers — a multi-year market share gain with durable economics and expanding TAM driven by agentic AI. Second, Instinct AI GPUs (MI300X → MI350 → MI400 series → MI500 planned 2027) are establishing AMD as the second-source AI accelerator behind NVIDIA — with particular strength in inference workloads demanding high HBM memory capacity. The OpenAI (6 GW) and Meta (6 GW) partnership agreements provide a multi-year demand floor for Instinct GPUs, while the Helios rack-scale platform (EPYC + Instinct + Pensando) positions AMD to compete at the system level with NVIDIA's DGX platform.
Revenue & Financial Metrics
| Metric | FY2025 | FY2024 | Change |
|---|---|---|---|
| Total Revenue | $34,639M | $25,785M | +34% YoY |
| Cost of Revenue | $17,487M | $13,060M | +34% |
| Gross Profit | $17,152M | $12,725M | +35% |
| Gross Margin (GAAP) | 50% | 49% | +1 ppt |
| Gross Margin (Non-GAAP) | 52% | 53% | -1 ppt |
| Research & Development | ~$6,800M | ~$5,700M | +~19% |
| GAAP Operating Income | $3,694M | $1,900M | +94% |
| GAAP Operating Margin | 11% | 7% | +4 ppts |
| Non-GAAP Operating Income | $7,768M | $6,138M | +27% |
| Non-GAAP Operating Margin | 22% | 24% | -2 ppts |
| GAAP Net Income | $4,335M | $1,641M | +164% |
| GAAP Net Margin | 12.5% | 6.4% | +6.1 ppts |
| Non-GAAP Net Income | $6,831M | $5,420M | +26% |
| GAAP EPS (Diluted) | $2.65 | $1.00 | +165% |
| Non-GAAP EPS (Diluted) | $4.17 | $3.31 | +26% |
| Metric | Dec 2025 | Dec 2024 |
|---|---|---|
| Cash, CE & Short-Term Investments | $10,600M | $5,100M |
| Inventory | ~$7,900M | ~$5,000M |
| Total Debt (principal) | $3,300M | $1,800M |
| Net Cash (Cash minus Debt) | ~$7,300M | ~$3,300M |
| Capital Returned (FY2025) | $1,300M (12.4M shares) | — |
| Share Repurchase Authorization Remaining | $9,400M | — |
| Q4 2025 Free Cash Flow | $2,100M (record) | — |
| Q1 2026 Free Cash Flow | $2,600M (record) | — |
| Period | Total Rev. | Data Center | Client | Gaming | Embedded | Non-GAAP GM% | Non-GAAP EPS |
|---|---|---|---|---|---|---|---|
| Q4 2024 | $7,658M (+24% YoY) | $3,857M (+69%) | $2,294M (+58%) | $563M (-59%) | $923M (+4%) | 54% | $1.09 |
| Q1 2025 | $7,438M (+36% YoY) | ~$3,700M | ~$2,302M | ~$648M | ~$823M | 54% | $0.96 |
| Q2 2025 (1) | $7,685M | $3,200M (+14%) | ~$2,700M | ~$940M | ~$846M | 43% | $0.48 |
| Q3 2025 | $9,246M (+36% YoY) | $4,300M (+22%) | $2,800M (+46%) | $1,300M (+181%) | $857M (-8%) | 54% | $1.20 |
| Q4 2025 | $10,270M (+34% YoY) | $5,400M (+39%) | $3,100M (+34%) | $843M (+50%) | $950M (+3%) | 57% | $1.53 |
| Q1 2026 | $10,253M (+38% YoY) | $5,800M (+57%) | $2,900M (+26%) | $720M (+11%) | $873M (+6%) | 55% | $1.37 |
(1) Q2 2025 Non-GAAP GM% of 43% reflects $800M MI308 export control inventory charges; underlying Non-GAAP GM would have been ~54% excluding charges. Q1 2025 and Q2 2025 segment figures partially estimated. Q4 2024, Q3 2025, Q4 2025, Q1 2026 segment data from official earnings releases.
Key Business Segments
| Segment | FY2025 | FY2024 | YoY | % of Total |
|---|---|---|---|---|
| Data Center (total) | $16,600M | ~$12,600M | +32% YoY | 47.9% |
| — EPYC Server CPUs | Majority of segment | — | +>50% mix | — |
| — Instinct AI GPUs | Rapidly scaling | — | Growing | — |
| Client (Ryzen PCs) | $10,600M | ~$7,020M | +51% YoY | 30.6% |
| Gaming (Radeon + Semi-Custom) | $3,900M | ~$2,583M | +51% YoY | 11.3% |
| Embedded (FPGAs / Adaptive) | $3,500M | ~$3,608M | -3% YoY | 10.1% |
AMD's largest and fastest-growing segment. EPYC server CPUs (4th Gen 'Genoa,' 5th Gen 'Turin'/Zen 5) power cloud instances at AWS, Azure, Google Cloud, Oracle, and Tencent — approximately 1,600 EPYC cloud instance types by end of 2025 (+50% YoY). AMD Instinct AI GPUs (MI300X, MI350 Series, now ramping MI400 Series) compete directly with NVIDIA H200/Blackwell in AI training and inference. AMD Pensando 'Pollara' and 'Vulcano' NICs address data center networking. Q4 2025 DC: $5.4B (+39% YoY). Q1 2026: $5.8B (+57% YoY). Segment operating margin Q4 2025: 33%.
Ryzen desktop CPUs (incl. 7000/9000 X3D series), Ryzen mobile processors, and Ryzen PRO commercial CPUs. Client revenue was a record $10.6B (+51% YoY) in FY2025, driven by continued Intel market share gains, a richer product mix (higher-ASP Ryzen AI 300/400 series), and commercial enterprise momentum (Ryzen Pro sell-through +40%+ YoY in Q4 2025). The Ryzen AI 400 series supports Copilot+ AI PC workloads with an integrated NPU. 6th Gen EPYC 'Venice' (Q3/Q4 2026) extends to enterprise and AI server headnodes.
AMD Radeon RX 9000 Series (RDNA 4 architecture) discrete GPUs for gaming PCs. Semi-custom SoCs for Microsoft Xbox Series X/S and Sony PlayStation 5 consoles. FY2025 gaming revenue was $3.9B (+51% YoY), with the YoY rebound driven by semi-custom recovery and Radeon demand. Semi-custom revenue is expected to decline significantly in 2026 (7th year of the current console cycle). Microsoft Xbox next-gen SoC (AMD semi-custom, targeting 2027 launch) and Valve Steam Machine (AMD-powered, shipping 2026) are the next semi-custom catalysts.
FPGAs (Artix, Kintex, Virtex, Versal AI Series), adaptive SoCs, and embedded x86 CPUs (EPYC Embedded, Ryzen Embedded, Ryzen AI Embedded). Primarily from the 2022 Xilinx acquisition. Revenue was $3.5B (-3% YoY) in FY2025, reflecting customer inventory normalization. $17B in embedded design wins in 2025 (+~20% YoY; $50B+ cumulative since Xilinx). End markets: aerospace/defense, industrial automation, test/measurement, communications, automotive, medical. Embedded segment operating margin remains the highest in AMD's portfolio at ~38-40%.
| Metric | Value | Period / Note |
|---|---|---|
| Data Center Revenue — FY2025 | $16,600M | FY2025 (12-mos ended Dec 27, 2025) |
| Data Center Revenue — FY2024 | ~$12,600M | FY2024 (12-mos ended Dec 28, 2024) |
| Data Center YoY Growth | +32% | FY2025 vs FY2024 |
| Data Center % of Total Revenue | 47.9% | FY2025 |
| Data Center Revenue — Q4 2025 | $5,400M (+39% YoY) | Q4 2025 (ended Dec 27, 2025) |
| Data Center Revenue — Q1 2026 | $5,800M (+57% YoY) | Q1 2026 (ended Mar 28, 2026) |
| Data Center Op. Income — Q4 2025 | $1,800M (33% margin) | Q4 2025 |
| Data Center Op. Income — Q1 2026 | $1,600M (28% margin) | Q1 2026 |
| 5th Gen EPYC ('Turin') Revenue Share | >50% of server CPU revenue | Q4 2025 |
| EPYC-Powered Cloud Instances | ~1,600 (+50% YoY) | As of end of 2025 |
| OpenAI GPU Deployment Partnership | 6 GW AMD GPUs (MI450+) | Multi-year, from H2 2026 |
| Meta GPU Deployment Partnership | Up to 6 GW AMD Instinct GPUs | Multi-gen partnership |
| MI308 Net Export Control Impact | ~$440M net inventory charge | FY2025 (Q2 charge; Q4 partial reversal) |
| Server CPU TAM Estimate (2030) | >$120B (>35% CAGR) | Management forecast, May 2026 |
SEC Filing Key Findings
FY2025 total revenue reached $34,639M — a 34% YoY increase from $25,785M in FY2024. Data Center segment revenue was a record $16,600M (+32% YoY), now representing 47.9% of total company revenue. Q1 2026 Data Center revenue of $5,800M (+57% YoY) was another record and the first time AMD disclosed that 'Data Center is now the primary driver of our revenue and earnings growth.' AMD's Data Center segment has grown from ~$6.0B in FY2023 to $16.6B in FY2025, approaching 3x growth over two fiscal years.
FY2025 10-K (CIK: 0000002488, Acc-no: 0000002488-26-000018, filed Feb 2026); Q4 2025 Earnings Release, Feb 3, 2026; Q1 2026 Earnings Release, May 5, 2026
During Q2 2025, AMD recorded approximately $800 million in inventory and related charges on AMD Instinct MI308 Data Center GPU products due to new U.S. export restrictions on certain semiconductors to China. AMD subsequently applied for and received licenses to ship MI308 to certain China-based customers. In Q4 2025, AMD reversed approximately $360 million of those charges and recognized approximately $390 million of MI308 revenue to China. The net FY2025 impact was approximately $440 million in inventory and related charges. This charge caused Q2 2025 GAAP gross margin to fall to 40% and Non-GAAP gross margin to 43%, masking AMD's underlying execution.
FY2025 10-K (CIK: 0000002488); Q2 2025 8-K (Aug 5, 2025); Q4 2025 Earnings Release & Call, Feb 3, 2026
In March 2025, AMD completed the acquisition of ZT Systems for $3.2 billion in cash and 8.3 million shares of AMD common stock. AMD retained select intellectual property and employees from the ZT Design Business. In October 2025, AMD sold the ZT Manufacturing Business to Sanmina Corporation for $2.4 billion in cash, plus 1.2 million shares of Sanmina common stock, with up to $450 million in contingent cash consideration. Sanmina became AMD's preferred manufacturing partner. The transaction gives AMD deep systems design capability (how AI rack-scale platforms are architected) without the capital intensity of a contract manufacturer.
FY2025 10-K (CIK: 0000002488); Q3 2025 Press Release, Nov 4, 2025; Q4 2025 Earnings Call, Feb 3, 2026
In October 2025, AMD entered into a product purchase agreement with OpenAI to deploy 6 gigawatts of AMD GPUs, starting with 1 GW of AMD Instinct MI450-based capacity. Concurrent with the agreement, AMD issued to OpenAI a warrant to purchase up to 160 million shares of AMD common stock at $0.01 per share. The warrant vests in tranches based on AMD Instinct GPU purchase milestones and AMD stock price targets, and is exercisable through October 5, 2030. As of December 27, 2025, no warrant shares had met the vesting conditions and the warrant had no financial statement impact. This partnership, alongside the similar Meta deal (6 GW of AMD Instinct GPUs), establishes AMD as a core hyperscale AI GPU partner alongside NVIDIA.
FY2025 10-K (CIK: 0000002488); Q3 2025 Earnings Call, Nov 4, 2025; Q4 2025 Earnings Call, Feb 3, 2026
AMD launched the 5th Gen EPYC family of server processors ('Turin,' built on Zen 5 architecture) in 2025. In Q4 2025, 5th Gen EPYC accounted for more than 50% of AMD server CPU revenue — the fastest adoption ramp of any EPYC generation. By end of 2025, the number of EPYC-powered public cloud instances reached approximately 1,600 (+50% YoY), as AWS, Azure, Google Cloud, Oracle, and Tencent all launched new AMD-based instances. The number of large businesses deploying EPYC on-premises more than doubled in 2025. AMD exited 2025 at record server market share. The 6th Gen EPYC ('Venice,' Zen 6 on TSMC 2nm) is on track to launch later in 2026.
FY2025 10-K (CIK: 0000002488); Q4 2025 Earnings Call, Feb 3, 2026; Q1 2026 Earnings Call, May 5, 2026
Cash, cash equivalents and short-term investments totaled $10.6 billion as of December 27, 2025, up from $5.1 billion at the end of 2024. The $5.5 billion net increase in cash reflects the strong operating free cash flow generation, partially offset by the ZT Systems acquisition net cost (~$0.8B net after divestiture proceeds) and $1.3 billion in share repurchases (12.4 million shares). Total debt increased from $1.8B to $3.3B, primarily related to ZT acquisition financing. AMD ended 2025 with $9.4 billion remaining under its share repurchase authorization. Q4 2025 free cash flow was a record $2.1B; Q1 2026 FCF was a record $2.6B (25% of revenue).
FY2025 10-K (CIK: 0000002488); Q4 2025 Earnings Release, Feb 3, 2026; Q1 2026 Earnings Release, May 5, 2026
AMD unveiled its 'Helios' rack-scale AI platform at OCP Global Summit in October 2025. Helios integrates AMD Instinct MI450 GPUs, 5th/6th Gen EPYC CPUs, and AMD Pensando 'Vulcano' AI networking NICs in a unified rack-scale design targeting hyperscale AI superclusters — AMD's equivalent of NVIDIA's DGX rack platform. Multiple OEMs (HPE, Lenovo) announced Helios systems in 2025. Oracle Cloud announced the first publicly available Helios-based AI supercluster with 50,000 MI450 GPUs in Q3 2026. Production shipments are on track for H2 2026. The MI400 series expands the Instinct portfolio: MI455X/Helios for AI superclusters, MI430X for HPC/sovereign AI, MI440X for enterprise 8-GPU servers.
FY2025 10-K (CIK: 0000002488); Q3 2025 Earnings Call, Nov 4, 2025; Q4 2025 & Q1 2026 Earnings Calls
The Embedded segment (primarily FPGA/adaptive SoCs from the 2022 Xilinx acquisition) reported $3,500M in FY2025 revenue, down 3% YoY, reflecting customer inventory normalization. Despite the revenue headwind, AMD closed $17 billion in embedded design wins in 2025 (+~20% YoY), bringing total wins to more than $50 billion since the Xilinx acquisition. New product launches included Versal AI Edge Gen 2 SoCs, Spartan UltraScale+ devices, EPYC Embedded 2005 series, and Ryzen AI Embedded P100 series. Design wins are a leading indicator of future revenue; embedded recovery and growth is expected in 2026.
FY2025 10-K (CIK: 0000002488); Q4 2025 Earnings Call, Feb 3, 2026
Earnings Call Intelligence
Sources: Q4 FY2025 Earnings Call, February 3, 2026 (Dr. Lisa Su, CEO; Jean Hu, CFO). Q1 2026 Earnings Call, May 5, 2026 (Dr. Lisa Su, CEO; Jean Hu, CFO).
2025 was a defining year for AMD, with record revenue, net income, and free cash flow driven by broad-based demand for our high-performance computing and AI products. We ended the year with significant momentum with every part of our business performing very well. We are entering a multiyear demand supercycle for high performance in AI computing that is creating significant growth opportunities across each of our businesses.
Q4 FY2025 Earnings Call Prepared Remarks, February 3, 2026 — Discussing full-year 2025 record results.
These results mark a clear inflection in our growth trajectory and a structural shift in our business. Data center is now the primary driver of our revenue and earnings growth, and as AI adoption scales, demand is increasing not only for accelerators, but also for the high performance CPUs that power and orchestrate those workloads.
Q1 2026 Earnings Call Prepared Remarks, May 5, 2026 — Announcing record Q1 2026 results with 38% revenue growth.
We have strong and increasing confidence in our ability to deliver tens of billions of dollars in annual data center AI revenue in 2027, and to exceed our long-term growth target of greater than 80% in the coming years. With the momentum we are seeing across the business and the expanding market opportunity, we see a clear path to exceed our long-term financial targets, including delivering more than $20 in EPS over the strategic time frame.
Q1 2026 Earnings Call Prepared Remarks, May 5, 2026 — Discussing data center AI growth confidence and long-term EPS targets.
AMD executed very well in 2025, delivering record revenue of $34.6 billion, up 34% year over year, driven by 32% growth in our data center segment and 51% growth in our client and gaming segment. Gross margin was 52%, and we delivered record earnings per share of $4.17, up 26% year over year, while continuing to invest aggressively in AI and the data center to support our long-term growth.
Q4 FY2025 Earnings Call Prepared Remarks, February 3, 2026 — Presenting full-year 2025 financial summary.
During the quarter, we generated $3 billion in cash from continuing operations and a record $2.6 billion in free cash flow, or 25% of revenue, demonstrating the cash generating power of our business model. At the end of the quarter, cash, cash equivalents, and short-term investments were $12.3 billion.
Q1 2026 Earnings Call Prepared Remarks, May 5, 2026 — Reporting record Q1 2026 free cash flow of $2.6B.
At our Financial Analyst Day in November, we outlined a server CPU market growing at approximately 18% annually over the next three to five years. Based on the demand signals we are seeing today and the structural increase in CPU compute requirements driven by agentic AI, we now expect the server CPU TAM to grow at greater than 35% annually, reaching over $120 billion by 2030.
Q1 2026 Earnings Call Prepared Remarks, May 5, 2026 — Revising server CPU TAM estimate upward due to agentic AI demand.
| Metric | Value | Period | Source |
|---|---|---|---|
| Q2 2026 Revenue Guidance | $11.2B (±$300M) | Q2 2026 (ends ~Jun 2026) | Q1 2026 Earnings Release, May 5, 2026 |
| Q2 2026 YoY Growth (mid) | +46% YoY | Q2 2026 | Q1 2026 Earnings Release |
| Q2 2026 Non-GAAP Gross Margin | ~56% | Q2 2026 | Q1 2026 Earnings Release |
| Q2 2026 Non-GAAP OpEx | ~$3.3B | Q2 2026 | Q1 2026 Earnings Call |
| Q2 2026 Non-GAAP Other Income | Gain of ~$60M | Q2 2026 | Q1 2026 Earnings Call |
| Q2 2026 Non-GAAP Tax Rate | ~13% | Q2 2026 | Q1 2026 Earnings Call |
| Server CPU Revenue Growth — Q2 2026 | >70% YoY | Q2 2026 | Q1 2026 Earnings Call, May 5, 2026 |
| Helios + MI450 Production Ramp | H2 2026 | Expected start | Q4 2025 & Q1 2026 Earnings Calls |
| 6th Gen EPYC ('Venice') Launch | Later 2026 | Calendar 2026 | Q1 2026 Earnings Call |
| Data Center AI Revenue Target | Tens of billions in 2027 | FY2027 | Q1 2026 Earnings Call, May 5, 2026 |
| DC Revenue CAGR Target (3–5 yr) | >60% annually | FY2026–FY2030 | Q4 2025 Earnings Call, Feb 3, 2026 |
| Long-Term EPS Target | >$20 EPS | Strategic time frame | Q4 2025 Earnings Call; Financial Analyst Day, Nov 2025 |
Risk Factors
U.S. export regulations restrict AMD from selling certain AI accelerators (MI308 and successors) to China-based customers without licenses. The MI308 restrictions caused ~$440M in net inventory charges in FY2025 and material Q2 2025 gross margin compression. Any expansion of export controls to future Instinct generations (MI400 series, MI500 series) could materially impair AMD's data center AI revenue growth trajectory. China remains a significant portion of the global AI hardware TAM.
NVIDIA holds estimated 70–90% AI GPU market share by revenue. Its CUDA software ecosystem — 20+ years of developer tooling, library integrations with PyTorch/JAX/TensorFlow, and community momentum — creates an extremely high switching cost. AMD's ROCm software stack has improved substantially with ROCm 7, but significant gaps remain in multi-GPU scale-up (NVLink equivalent), software maturity, and enterprise support tooling. At the same time, AMD Instinct is gaining traction for inference workloads due to its high HBM memory capacity.
AMD is a fabless semiconductor company — all leading-edge wafer production is contracted to TSMC (N3, N4 process nodes; 2nm for 'Venice'). Any disruption to TSMC — geopolitical conflict, natural disaster, or capacity allocation prioritization — would severely limit AMD's ability to ship EPYC, Instinct, and Ryzen products. Intensifying competition for TSMC capacity with NVIDIA, Apple, and Qualcomm is an ongoing risk to supply availability and wafer cost.
AMD Instinct GPUs use High Bandwidth Memory (HBM), sourced primarily from SK Hynix, Samsung, and Micron. HBM supply is tight industry-wide given surging AI GPU demand. AMD has disclosed collaborative agreements with Samsung on HBM4 supply for future Instinct MI455X products. Supply chain constraints — including CoWoS advanced packaging capacity at TSMC and HBM availability — could limit AMD's ability to ship Instinct GPUs at the scale required to meet hyperscaler commitments.
A significant portion of AMD's Data Center revenue is concentrated among a small number of hyperscale customers: AWS, Azure, Google Cloud, Oracle, Meta, and OpenAI. AMD's multi-year partnership agreements (OpenAI: 6 GW, Meta: 6 GW) provide revenue visibility but also create revenue concentration risk. Reduced CapEx spending by a major hyperscaler — due to AI strategy pivots, recession, or cost constraints — could disproportionately impact AMD's near-term Data Center revenue.
Major hyperscalers (Google TPU, Amazon Trainium/Inferentia, Microsoft Maia, Meta MTIA) continue to develop proprietary AI accelerators that reduce reliance on merchant silicon. AMD's Instinct GPUs compete in the merchant GPU market but cannot address custom ASIC workloads. Intel's Gaudi 3 has gained limited traction. The risk is partially offset by AMD's broad applicability across cloud, enterprise, and sovereign AI customers, many of whom cannot develop custom silicon.
While AMD has accelerated ROCm development (ROCm 7 released in 2025; day-zero support for leading open models; MLPerf leadership results in Q1 2026), the CUDA ecosystem remains significantly more mature in tooling, documentation, community expertise, and enterprise support infrastructure. Developers default to CUDA for new AI projects. AMD's ability to reach tens of billions of data center AI revenue in 2027 depends in part on ROCm maturation keeping pace with the expanding customer deployment base.
AMD issued a warrant to OpenAI in October 2025 for up to 160 million shares of AMD common stock at $0.01/share (effectively free shares), exercisable through 2030 upon meeting GPU purchase and stock price milestones. Similar warrant-like arrangements may apply to other large customer partnerships. If these warrants vest and are exercised, they represent meaningful dilution to existing shareholders — up to approximately 10% of AMD's current diluted share count of ~1.66 billion shares.
Competitive Positioning
AMD competes on two fronts simultaneously. In server CPUs, AMD EPYC is displacing Intel Xeon across hyperscaler and enterprise data centers. Intel's Sapphire Rapids and Emerald Rapids have failed to close the performance-per-watt gap; AMD exited 2025 at record x86 server market share. AMD's 6th Gen EPYC 'Venice' (Zen 6, TSMC 2nm) is designed to extend this advantage further, with management citing "substantially higher performance per socket and per Watt versus competitive x86 offerings, and more than 2x throughput per socket versus leading ARM-based AI solutions."
In AI accelerators, AMD Instinct competes against NVIDIA's dominant H200/Blackwell platform. AMD's structural advantage in inference is high HBM memory capacity and bandwidth (MI300X ships with 192GB HBM3, addressing LLM inference memory requirements that exceed GPU memory on NVIDIA H100). For training, AMD lags NVIDIA in multi-GPU scale-up (no direct NVLink equivalent at scale), software maturity (ROCm vs CUDA), and system-level integration. The Helios platform addresses the system-level gap, while ROCm investments are closing the software gap. AMD's MLPerf Inference 6.0 results (MI355X) showed strong competitive performance across the full suite with leadership in multiple categories.
| Dimension | AMD | NVIDIA | Intel | Custom ASICs |
|---|---|---|---|---|
| Server CPU Leadership | ✓ EPYC (record share) | No product | Xeon (losing share) | None (x86) |
| AI GPU — Inference HBM | ✓ MI300X: 192GB HBM | H200: 141GB HBM | Gaudi 3 (limited) | Domain-specific |
| AI GPU — Training Scale-Up | ROCm + Infinity Fabric | ✓ NVLink 72-GPU | No equiv. | Internal only |
| AI Software Ecosystem | ROCm (accelerating) | ✓ CUDA (dominant) | oneAPI (limited) | Proprietary only |
| Rack-Scale AI Platform | Helios (H2 2026) | ✓ DGX/HGX (market) | Gaudi rack (minimal) | Custom (some) |
| DC Revenue (FY2025) | $16.6B (+32% YoY) | $193.7B (+68% YoY) | <$1B AI | Internal use only |
| Hyperscale GPU Partnerships | OpenAI 6GW, Meta 6GW | All major CSPs | Minimal | Self-supply |
Verikal Assessment
AMD delivered $34.6 billion in FY2025 revenue (+34% YoY), $4.3 billion in GAAP net income (+164% YoY), and non-GAAP EPS of $4.17 (+26% YoY). These results reflect the strongest absolute revenue and earnings in AMD's history, driven by three concurrent drivers: EPYC server CPU market share gains, Instinct AI GPU ramp, and Ryzen PC growth. Data Center revenue of $16.6 billion (+32% YoY) now represents nearly half of AMD's total revenue — a structural shift from AMD's historical dependence on PC and gaming. Q1 2026 continued the acceleration: revenue $10.3B (+38% YoY), Data Center $5.8B (+57% YoY), and record free cash flow of $2.6B (25% of revenue).
The EPYC server CPU business is a proven, durable market share gain story. AMD exited 2025 at record server market share, with 5th Gen EPYC ('Turin') accounting for >50% of AMD server revenue in Q4 2025 — the fastest generational adoption rate in EPYC history. AMD revised its server CPU TAM estimate upward in May 2026 from 18% CAGR to >35% CAGR, driven by agentic AI workflows requiring high-performance CPUs as head nodes and orchestration for GPU clusters. With 6th Gen EPYC 'Venice' (Zen 6, TSMC 2nm) launching later in 2026, AMD is on track to widen its performance-per-watt advantage over Intel Xeon. Venice is also the first EPYC purpose-built for AI infrastructure co-design with Instinct.
The Instinct AI GPU franchise — the highest-risk, highest-reward component of AMD's data center story — is inflecting. Eight of the top 10 AI companies used Instinct for production workloads by Q4 2025. The OpenAI (6 GW, Oct 2025) and Meta (6 GW, Feb 2026) partnership agreements — both multi-generation, multi-year commitments — provide strong visibility and fundamentally change AMD's position in the AI hyperscale GPU market. Helios production shipments (H2 2026) represent the next step: a co-designed, full-stack EPYC+Instinct+Pensando rack solution, directly addressing the same hyperscale AI factory market as NVIDIA's DGX platform. Management expressed strong confidence in delivering tens of billions in annual data center AI revenue in 2027.
The MI308 export control episode in FY2025 is a cautionary data point on China regulatory risk. The $800M gross charge in Q2 2025 compressed gross margins to 40% GAAP / 43% Non-GAAP — not representative of AMD's underlying profitability. The subsequent $360M Q4 reversal and $390M in China MI308 revenue demonstrates AMD's ability to navigate export restrictions through licensing, but ongoing U.S. policy tightening on AI chip exports to China remains an open risk variable. AMD currently assumes approximately $100M per quarter of MI308 China revenue in its near-term guidance, a small fraction of its overall data center addressable market.
Sources & Methodology
All financial data sourced from SEC EDGAR: Advanced Micro Devices, Inc. (CIK: 0000002488) Form 10-K Annual Report for fiscal year ended December 27, 2025 (FY2025), filed February 2026 (Acc-no: 0000002488-26-000018); Form 10-K Annual Report for fiscal year ended December 28, 2024 (FY2024), filed February 2025 (Acc-no: 0000002488-25-000012); Form 10-Q for quarter ended March 28, 2026 (Q1 2026), filed May 2026 (Acc-no: 0000002488-26-000076); Q2 2025 8-K (filed August 5, 2025); Q3 2025 8-K (filed November 4, 2025). Management quotes sourced from Q4 FY2025 Earnings Call Transcript (February 3, 2026) and Q1 2026 Earnings Call Transcript (May 5, 2026). Q1 2025 and Q2 2025 segment figures partially estimated from available full-year and quarterly disclosed data. FY2024 segment figures calculated from disclosed YoY growth rates. No forward projections or price targets are provided.