Equity Research: Advanced Micro Devices (AMD)

Recent News & Earnings

Business Summary

Advanced Micro Devices designs and sells high-performance computing products across four segments. The company competes with Intel in x86 server and PC processors, with NVIDIA in AI accelerator GPUs, and with a range of embedded chip makers in FPGAs and adaptive SoCs. AMD's $49B acquisition of Xilinx in 2022 added the embedded portfolio and roughly $25B of goodwill to the balance sheet.

Data Center (48% of TTM Revenue, $16.6B)

The growth engine. This segment includes EPYC server CPUs and Instinct AI GPU accelerators. EPYC has been taking share from Intel Xeon for five consecutive years, now powering nearly 1,600 cloud instances across AWS, Google, Azure, and others. Instinct GPUs (MI300/MI350 series) target AI training and inference workloads. Eight of the top ten AI companies deploy Instinct in production. Data center operating income hit $1.8B in Q4 2025, a 33% margin.

Client (~30% of TTM Revenue)

Ryzen desktop and mobile CPUs for consumer and commercial PCs. The client business hit a record $3.1B in Q4 2025 revenue, up 34% year over year. Desktop channel sell-out was a record for four consecutive quarters. The commercial PC opportunity is growing, with Ryzen sell-through for commercial notebooks up more than 40% year over year. The Ryzen AI 400 series positions AMD in the emerging AI PC category.

Gaming (~12% of TTM Revenue)

Semi-custom SoCs for PlayStation and Xbox consoles, plus Radeon discrete GPUs for PC gaming. This segment faces a multi-year headwind: the current console cycle is entering its seventh year, and management guided for a "significant double-digit percentage" decline in semi-custom revenue in 2026. Microsoft's next-gen Xbox, featuring an AMD SoC, is on track for 2027. Radeon GPU revenue grew year over year in Q4, led by the RX 9000 series launch.

Embedded (10% of TTM Revenue, $3.5B)

Xilinx-acquired FPGA, adaptive SoC, and embedded CPU business targeting industrial, automotive, aerospace, and communications markets. After a painful inventory correction in 2023-2024, this segment returned to year-over-year growth in Q4 2025. Design win momentum is the standout metric: $17B in new design wins in 2025 (+20% YoY), with cumulative wins exceeding $50B since the Xilinx acquisition. These design wins convert to revenue over 3-7 year product lifecycles.

Data Center: CPUs + AI GPUs

Data Center is where AMD's investment thesis lives. This segment contributes 48% of trailing twelve-month revenue but is driving more than 80% of incremental growth. It consists of two distinct growth vectors: EPYC server CPUs (the share-gain story) and Instinct AI GPUs (the market-expansion story). Both are compounding simultaneously.

EPYC: The Server Share Engine

Fifth-generation Turin EPYC CPUs accounted for more than half of server CPU revenue in Q4 2025. Fourth-generation Genoa continues to sell well alongside it, which is unusual and signals strong demand across all price points. Enterprise on-premises EPYC deployments more than doubled in 2025. The server OEMs now offer more than 3,000 EPYC-powered solutions. The next-generation Venice CPU launches later in 2026, and Lisa Su described customer pull as "very high."

I find the CPU story underappreciated. As AI workloads scale, the need for high-performance host CPUs grows in parallel, because every GPU cluster needs CPUs to manage data pipelines, scheduling, and pre/post-processing. Lisa Su made this point directly: "Agentic and emerging AI workloads require high-performance CPUs to power head nodes and run parallel tasks alongside GPUs." This means EPYC demand is a beneficiary of, not a competitor to, the GPU ramp.

Instinct GPU: The AI Growth Vector

MI350 series GPUs ramped through Q4 2025 and are the current production platform. MI450 and the Helios rack-scale system launch in H2 2026, which management calls a "major inflection point." The vast majority of MI450 revenue will come from RackScale solutions shipped to rack builders, with revenue recognized on shipment. The MI500 series (cDNA6 architecture on 2nm with HBM4e memory) is on track for 2027.

The product roadmap velocity is a competitive strength. AMD is shipping new GPU generations on an annual cadence: MI300 (2024), MI350/355 (2025), MI450 (2026), MI500 (2027). This gives customers confidence in a long-term platform commitment, which is why OpenAI signed a multi-year, multi-generation partnership for six gigawatts of Instinct GPUs.

ROCm Software Ecosystem

The biggest question in AMD's AI thesis is software. NVIDIA's CUDA ecosystem is the dominant platform for AI development, and switching costs are real. AMD's answer is ROCm, an open-source GPU computing platform. Millions of large language models now run out of the box on AMD GPUs with day-zero support. AMD has upstream integration in vLLM (one of the most widely used inference engines) and launched an enterprise AI suite for production deployments. The software gap is narrowing, but it remains AMD's primary competitive vulnerability.

Data Center Revenue by Sub-Segment (TTM, $B)

Q4 2025 operating margin of 33% includes the benefit of a $306M inventory reserve release. Clean operating margin was approximately 30%, consistent with prior quarters.

Industry Context

The AI infrastructure spending cycle is the defining force in the semiconductor industry. Hyperscalers (Microsoft, Google, Meta, Amazon) collectively committed to more than $200B in capital expenditures for 2026, the majority directed at AI training and inference compute. Demand for AI accelerators is doubling annually, and the total addressable market (TAM) for data center GPUs is projected to exceed $100B by 2028. Alongside this, the x86 server CPU market (~$30B) is growing at strong double-digit rates as AI workloads require high-performance host processors alongside GPUs.

AMD operates in this market as the credible second source. NVIDIA dominates AI GPUs with an estimated 80%+ market share and a massive CUDA software moat. Intel is losing server CPU share to AMD for the fifth consecutive year. AMD's strategy is to be the alternative supplier across both categories, capturing share through competitive performance and lower total cost of ownership (TCO). This "second-source" position is structurally advantageous: hyperscalers actively want to avoid single-supplier dependency, and AMD is the only company that can offer both a competitive x86 CPU and a competitive AI GPU. That combination is unique in the industry.

Competitive Advantages & Moat

1. Dual CPU + GPU Architecture

AMD is the only company with leadership products in both server CPUs and AI GPUs. Intel makes CPUs but lacks a competitive GPU. NVIDIA makes GPUs but has no x86 CPU. This allows AMD to offer bundled data center solutions and creates a unique cross-selling opportunity. No other company can deliver a complete x86 CPU + AI accelerator stack from a single vendor.

2. EPYC Share Momentum and Switching Costs

Winning server CPU share is extremely sticky. Once a hyperscaler qualifies and deploys EPYC at scale, the switching costs are significant: software validation, power and thermal design, procurement contracts, and multi-year deployment cycles. AMD's cloud instance count grew 50% year over year to nearly 1,600, creating a growing installed base. Enterprise on-premises deployments more than doubled. This momentum compounds because each new deployment validates EPYC for the next customer.

3. Product Roadmap Velocity

AMD ships new CPU and GPU generations on aggressive annual cadences. On the CPU side: Turin (2025), Venice (2026). On the GPU side: MI350 (2025), MI450 (2026), MI500 (2027). This clear multi-year roadmap gives customers confidence in long-term platform commitment. The OpenAI multi-year, multi-generation partnership is direct evidence that large customers trust AMD's execution.

4. Open-Source Software Strategy

While CUDA dominates, AMD's ROCm is open-source and gaining upstream community support. Millions of large language models run out of the box on AMD GPUs. The open-source approach reduces lock-in concerns for enterprise customers who want to diversify away from NVIDIA. This is not yet a moat, but it is becoming a credible alternative, and for customers who want supplier diversification, AMD is the only real option.

5. Xilinx Embedded Design Win Pipeline

$50B+ in cumulative design wins since the Xilinx acquisition gives AMD visibility into multi-year embedded revenue. These design wins convert to production revenue over 3-7 year lifecycles across industrial, automotive, aerospace, and communications markets. The embedded segment is just beginning its AI adoption cycle, with the Versal AI Edge Gen2 platform targeting low-latency inference at the edge.

Revenue Growth & Profitability

Historical Income Statement ($B)

FY2022-2023 profitability decline reflects the Xilinx acquisition: $3.5B+ in annual intangible amortization, the embedded inventory correction, and elevated integration costs. FY2025 represents a return to strong profitability with the data center AI ramp.

Earnings Per Share

AMD reports significant non-GAAP adjustments because of the Xilinx acquisition. The largest is intangible asset amortization, which runs approximately $2.3B per year and is declining as acquired intangibles depreciate. All EPS figures in this report and analyst consensus forecasts use non-GAAP adjusted earnings to strip out this non-cash charge. GAAP earnings are substantially lower: $2.65 for FY2025 vs. $4.17 non-GAAP.

EPS History & Forecast

FY2022 GAAP EPS collapse reflects Xilinx acquisition dilution (shares jumped from 1.23B to 1.56B) and $3.5B+ in intangible amortization. Non-GAAP EPS provides a cleaner view of underlying profitability. FY2026E and FY2027E are consensus estimates as of Feb 2026. Lisa Su has stated a long-term target of EPS exceeding $20.

Balance Sheet

AMD's balance sheet is conservatively managed with near-zero leverage. The $6.7B net cash position and 2.85x current ratio give the company significant flexibility to invest in R&D, ramp AI GPU production, and accelerate buybacks. The main item to monitor is $25.1B in goodwill from the Xilinx acquisition, which represents 33% of total assets. If the embedded segment's recovery stalls, goodwill impairment would be a meaningful negative event, even though it is non-cash. Inventory of $7.9B is elevated but intentional, as management is building supply to support the MI450 ramp in H2 2026.

3-Year Forecast & Scenario Analysis

The forecast spans FY2025 to FY2028. Three scenarios are differentiated by how quickly AMD's AI GPU business scales and how much operating leverage the company generates as revenue mix shifts toward higher-margin data center products. All pricing based on Mar 2026 entry price of $192. Price targets apply a forward P/E to FY2028E non-GAAP diluted EPS.

Base Case — Annual Projections ($B)

FY2026E and FY2027E revenue and EPS align with analyst consensus. FY2028E is my estimate assuming revenue growth decelerates to ~27% and non-GAAP operating margins expand to ~34% as intangible amortization declines and data center mix improves. Share count assumes modest dilution from stock-based compensation, partially offset by buybacks.

Valuation

I ran a discounted cash flow (DCF) model with a 5-year explicit forecast period (FY2026-FY2030), followed by a Gordon Growth terminal value. The weighted average cost of capital (WACC) uses an adjusted beta of 1.0. AMD's raw beta is 2.02, but this reflects semiconductor cyclicality and retail trading volatility, not fundamental business risk. For a 5-year DCF, I believe a normalized beta of 1.0 better reflects AMD's underlying risk profile as a diversified large-cap semiconductor company with strong cash generation. This results in a WACC of 9.75%.

The DCF confirms that AMD is approximately fairly valued at $192 on a fundamental cash flow basis. This means the current price fully reflects AMD's existing business trajectory. The 3-year price target of $300 (+56.3%) is derived from a multiples-based approach: 22x FY2028E non-GAAP EPS of $13.60. I use a multiples-based target because AMD is a high-growth company where near-term EPS acceleration (61-63% growth in FY2026-2027) drives more value than a terminal-value-heavy DCF can capture. The 22x exit multiple is conservative: it is below AMD's current forward P/E of 28.6x and below the semiconductor sector average for high-growth names (25-30x). The PEG ratio of 0.68 is the most compelling single metric. It means investors are getting $1 of AMD's earnings growth for only $0.68, compared to more than $1.67 for Alphabet or $1.50 for NVIDIA.

DCF Sensitivity Analysis

The table shows implied DCF intrinsic value per share across a range of WACC and terminal growth rate (TGR) assumptions. The highlighted cell represents the base case (WACC 9.75%, TGR 2.5%). All values in USD per share.

Highlighted cell (TGR 2.5% / WACC 9.75%) represents the base case DCF intrinsic value of ~$192, matching the current stock price. The 3-year price target of $300 is a multiples-based estimate that captures the near-term EPS growth trajectory the DCF inherently understates.

The sensitivity range of $165-$229 tells a clear story. Even in the most optimistic DCF scenario (WACC 9.0%, TGR 3.0%), the implied value of $229 is well below the $300 base price target. This gap exists because AMD is in the steep part of its growth curve: the DCF, anchored by a terminal value decades in the future, cannot fully price 60%+ annual EPS growth happening in the next two years. The multiples-based approach, which directly prices that near-term growth, is the more appropriate valuation method for a company at this stage. The DCF serves as a floor: it confirms there is no fundamental overvaluation at the current $192 entry price.

Key Risks

1. NVIDIA Competitive Dominance

NVIDIA controls an estimated 80%+ of the AI GPU market and has a massive CUDA ecosystem moat. If NVIDIA's Blackwell and Rubin platforms maintain their performance lead and customers remain unwilling to diversify GPU suppliers, AMD's Instinct market share gains could stall below 10%. The ROCm software gap remains AMD's single biggest competitive vulnerability. A world where CUDA lock-in deepens rather than loosens is the primary threat to the thesis.

2. AI Capex Cycle Reversal

AMD's growth thesis depends on hyperscalers continuing to spend aggressively on AI infrastructure. If AI revenue growth for cloud providers disappoints expectations and capex budgets are cut, GPU and server CPU demand could slow abruptly. AMD's high beta (2.0) means the stock would get hit disproportionately in a capex downturn. This is a macro risk that is difficult to hedge at the individual stock level.

3. Console Cycle Decline

Semi-custom SoC revenue (PlayStation, Xbox) is entering the seventh year of the current cycle. Management guided for a "significant double-digit percentage" decline in 2026. This creates a multi-billion-dollar revenue headwind that data center growth must more than offset. The next-gen Xbox AMD SoC launches in 2027, but the gap year creates a real drag on total revenue growth. If the console decline is steeper than expected while the MI450 ramp is slower, FY2026 revenue could disappoint.

4. Gross Margin Dilution from GPU Mix

AI GPUs carry lower gross margins than server CPUs. As GPU revenue becomes a larger share of total revenue, blended gross margins could compress even as revenue grows. The Q4 2025 clean gross margin of ~55% (excluding the $306M inventory reserve release) sets the current baseline. If MI450 margins are lower than MI350, or if AMD needs to price aggressively to win GPU share, margin expansion may disappoint relative to expectations.

5. China Export Controls

AMD generated $390M in MI308 China sales in Q4 2025, then guided only $100M for Q1 2026 due to "a very dynamic situation." Further export restrictions could eliminate this revenue entirely and reduce the global addressable market for AMD's AI GPUs. Management explicitly stated they are not forecasting additional China GPU revenue beyond Q1. This is a binary risk: either regulations stabilize and China becomes a meaningful revenue stream, or the market closes entirely.

6. Xilinx Goodwill Impairment

The $25.1B in goodwill from the Xilinx acquisition is the largest single balance sheet item and represents 33% of total assets. If the embedded segment's recovery stalls or design wins do not convert to revenue at expected rates, a goodwill write-down would be a significant negative event. While non-cash, a goodwill impairment would signal that the Xilinx acquisition has not delivered its expected strategic value, which would weigh heavily on investor sentiment.

Final Verdict