Advanced Micro Devices, Inc. (AMD) — Company Research
Last updated: 20 April 2026
AMD is the world's second-largest AI accelerator supplier and the fastest-growing large-cap semiconductor company. FY2025 revenue hit a record $34.6bn (+34% YoY) on the back of the MI300X/MI350 data-centre GPU ramp, an Oracle 50,000-unit MI450 order, and the OpenAI 6 GW compute partnership announced in October 2025. The stock is near an all-time high — $281.05 on 17 April 2026 — and the company's ZT Systems acquisition (closed October 2025) has repositioned AMD from a pure chip vendor to a full AI rack/system supplier, squaring up directly against Nvidia.
1. Company Snapshot
| Company | Advanced Micro Devices, Inc. |
| Ticker | NASDAQ: AMD |
| Sector / Industry | Information Technology / Semiconductors |
| Founded | 1969 (Sunnyvale, California) |
| Headquarters | Santa Clara, California |
| CEO | Dr. Lisa Su (since Oct 2014; Chair since Feb 2022) |
| Market cap (20 Apr 2026) | ~$453.9bn |
| Share price (20 Apr 2026) | ~$278.39 (all-time high $281.05 on 17 Apr) |
| Revenue (FY2025) | $34.6bn (+34% YoY) |
| Net income (FY2025, GAAP) | $4.3bn |
| Diluted EPS (FY2025, GAAP) | $2.65 |
| Employees | ~28,000 (pre-ZT); ~35,000+ post-ZT Systems close |
| Exchanges | NASDAQ |
| Website | amd.com |
2. Bull Case vs Bear Case
Bull Case
- AI GPU ramp accelerating. Q4 2025 Data Center segment revenue was up triple-digits YoY; Lisa Su said on the 3 Feb 2026 call that "Data Center GPU revenue exceeded $5bn in 2024 and grew substantially above that in 2025."
- OpenAI 6 GW deal (Oct 2025) — multi-year supply of MI450 accelerators starting 2H 2026. Warrant structure means OpenAI can take up to a 10% equity stake at $0.01 per share subject to deployment and share-price milestones.
- Oracle 50,000-unit MI450 order (Oracle Cloud Infrastructure, disclosed Q4 2025). Multi-billion-dollar revenue visibility.
- EPYC share gain continuing: Mercury Research put Q4 2025 server CPU unit share at 30%+ vs Intel. FY2025 Data Center segment revenue $16bn (+94% YoY, estimated).
- ZT Systems closed Oct 2025 — gives AMD in-house rack/system design capability; Helios rack due H2 2026.
Bear Case
- Nvidia dominance. Nvidia still holds ~85–90% of the discrete AI accelerator market. CUDA ecosystem lock-in remains AMD's biggest structural issue.
- China export controls. MI308 was banned outright in Jan 2024; MI308X banned Apr 2025; MI450 expected to face similar restrictions. Loss of China materially dilutes the total AI TAM for AMD.
- Hyperscaler custom silicon. Google TPU, AWS Trainium/Inferentia, Microsoft Maia, Meta MTIA, Anthropic's Broadcom/Google-built ASICs all compete with AMD's merchant GPU business.
- Gaming decline. Semi-custom (PlayStation/Xbox) and Radeon gaming revenue fell further in FY2025 as the console cycle aged.
- Concentration risk. Top 2 customers (rumoured to be Microsoft + OpenAI/Oracle via custom supply) now likely >40% of Data Center GPU revenue.
- ZT integration. $4.9bn deal adds meaningful debt + lower-margin system-integration revenue; GAAP operating margin could dilute in 2026.
3. What Does This Company Actually Do?
AMD designs (but does not manufacture) high-performance processors, graphics chips, and AI accelerators. All wafer production is outsourced, overwhelmingly to TSMC.
FY2025 Segment Breakdown
| Segment | FY2025 Revenue | % of Total | Key Products |
|---|---|---|---|
| Data Center | ~$16.0bn (estimated) | ~46% | EPYC server CPUs (Turin / Venice), Instinct MI300X/MI325X/MI350, Pensando DPUs |
| Client | ~$8.0bn | ~23% | Ryzen / Ryzen AI (desktop + mobile), Threadripper |
| Gaming | ~$3.5bn | ~10% | Radeon (RDNA 4), semi-custom (PS5/Xbox/Steam Deck) |
| Embedded | ~$7.1bn | ~21% | Xilinx FPGAs / adaptive SoCs, automotive, industrial |
| Total | $34.6bn | 100% |
Geographic Revenue (FY2025)
AMD does not break out revenue by geography with precision, but: US ~35–40%; China ~15–20% (declining on export controls); rest of APAC (Taiwan, Japan, Korea, Singapore as ship-to) ~30%; EMEA ~10–15%.
4. The Business Model
- Fabless. All leading-edge silicon manufactured by TSMC (N3/N3P for MI350; N2/A16 roadmap for MI400). Some Embedded / legacy products at Samsung and GlobalFoundries.
- Gross margin (FY2025, GAAP): ~50% (management non-GAAP figure ~54%). Data Center GPU margins are dilutive vs EPYC in the ramp phase.
- Operating margin (FY2025, GAAP): ~10.7% ($3.7bn op income / $34.6bn revenue). Non-GAAP op margin ~24%.
- Net margin (FY2025, GAAP): ~12.4%.
- Moat: Zen architecture (x86 performance leadership), chiplet packaging IP, CDNA GPU architecture, Xilinx FPGA portfolio, ROCm open-source software stack (accelerating as an alternative to CUDA but still a distant #2).
- Route to market: Direct hyperscaler (growing fastest — Microsoft, Meta, Oracle, OpenAI), OEM (HP, Dell, Lenovo, Supermicro), distribution (channel).
- Subsidies / credits: AMD is a beneficiary of CHIPS Act and advanced-packaging grants through TSMC Arizona and ZT Systems US operations; exact credit value immaterial vs revenue base but relevant to supply security.
5. Financial Health
| Metric ($bn) | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 |
|---|---|---|---|---|---|
| Revenue | 16.4 | 23.6 | 22.7 | 25.8 | 34.6 |
| GAAP gross margin | 48% | 45% | 46% | 49% | 50% |
| GAAP operating income | 3.65 | 1.26 | 0.40 | 1.90 | 3.70 |
| GAAP net income | 3.16 | 1.32 | 0.85 | 1.64 | 4.30 |
| GAAP diluted EPS | $2.57 | $0.84 | $0.53 | $1.00 | $2.65 |
| Cash, equivalents & S-T investments | 3.6 | 5.9 | 5.8 | 5.1 | ~6.5 |
| Total debt | 0.3 | 2.5 | 2.5 | 1.7 | ~6.5 (post-ZT) |
| Free cash flow | 3.2 | 3.1 | 1.1 | 2.4 | ~5.5 |
AMD does not pay a dividend. FY2025 buybacks were modest as cash was re-directed toward the ZT Systems close. Share count has crept up slightly on SBC.
6. Valuation & Market Data
| Metric | Value | Source / As of |
|---|---|---|
| Share price (20 Apr 2026) | ~$278.39 | NASDAQ |
| Market cap | ~$453.87bn | 20 Apr 2026 |
| Enterprise value | ~$454bn | ~$6.5bn debt offset by similar cash |
| P/E (TTM, GAAP) | ~105x | EPS $2.65 |
| P/E (Fwd, company FY2026 growth guide) | ~45x | implied non-GAAP |
| P/S (TTM) | ~13.1x | $34.6bn |
| EV/EBITDA (TTM, GAAP) | ~80x | |
| 52-week high | $281.05 | 17 Apr 2026 |
| 52-week low | ~$86 | early April 2025 |
| 12-month return | +213% | 20 Apr 2025 → 20 Apr 2026 |
| Short interest | ~1.8% of float | latest FINRA |
| Days to cover | ~0.7 | |
| Dividend | None |
7. What Are They Building / What's Coming?
- MI350 (ramping Q1 2026): CDNA 4, HBM3E, 288GB memory. Shipping now, targeted at existing hyperscaler customers.
- MI400 / MI450 (H2 2026 launch): CDNA 5 architecture, Helios rack-scale platform — AMD's answer to Nvidia's Blackwell/Rubin. Oracle 50,000-unit order tied to MI450.
- OpenAI 6 GW deal — starts shipping MI450-series in H2 2026. Warrant structure disclosed in 8-K filed 6 Oct 2025 (warrant for up to 160m shares, strike $0.01, vesting on deployment and share-price milestones).
- ZT Systems (closed Oct 2025) — system/rack design and assembly in New Jersey and Texas. AMD intends to sell the manufacturing arm while retaining the design team.
- EPYC "Venice" (Zen 6): Targeted for H2 2026 launch on TSMC N2, first server chip to use A16 process features.
- ROCm 7: multi-platform, PyTorch-native, released early 2026 — first ROCm release with CUDA-equivalent developer tooling targeting non-expert users.
- Advancing AI event 2026: San Jose, June 2026 — expected to unveil MI450 customer list, Helios rack specs, and Zen 6 details.
- Capex guidance: remains modest (single-digit % of revenue) thanks to fabless model. ZT Systems adds facility costs but AMD has said it will divest manufacturing.
8. Competitive Landscape
| Competitor | Overlap | AMD's Relative Position |
|---|---|---|
| Nvidia (NVDA) | AI accelerators, networking (NVLink/InfiniBand) | AMD ~10–15% share in Data Center GPU vs Nvidia ~85%+; CUDA remains dominant software stack |
| Intel (INTC) | Server CPUs (Xeon), client CPUs, Gaudi AI | AMD passed Intel on DC CPU revenue in Q2 2024; Gaudi 3 has not gained material share vs MI300X |
| Broadcom (AVGO) | Custom AI ASICs for Google/Meta, networking | Complement rather than pure rival — but erodes AMD merchant-GPU TAM at hyperscalers |
| Marvell (MRVL) | Custom silicon for AWS | Niche vs AMD's merchant GPU business |
| Apple (AAPL) | Consumer CPU (M-series) | Structural headwind in Client segment |
| Arm-based server CPUs (Nvidia Grace, Ampere, AWS Graviton) | DC CPU | Growing share of hyperscaler internal workloads — structural long-term risk to EPYC ASP mix |
Market size: Global AI accelerator TAM is ~$100bn+ in 2026 and management has publicly cited a ~$500bn AI accelerator TAM by 2028 — the same number Nvidia uses. AMD's explicit target is to reach "tens of billions of dollars" in Data Center GPU revenue on a run-rate basis by 2027.
9. Leadership and Ownership
- Dr. Lisa Su — Chair & CEO. MIT PhD in EE. CEO since Oct 2014; the stock has compounded 50x+ since.
- Jean Hu — CFO (since Jan 2023). Formerly Marvell CFO.
- Forrest Norrod — EVP, Data Center Solutions Business Group. Runs the EPYC + Instinct P&L.
- Vamsi Boppana — SVP, Artificial Intelligence Group. Runs ROCm + MI400 software stack.
- Mark Papermaster — CTO.
- Phil Guido — Chief Commercial Officer (hired from IBM).
Ownership
Insider ownership is low (management and directors collectively <1%). Major holders: Vanguard ~9%, BlackRock ~7%, State Street ~4%, FMR ~3%. Short interest ~1.8% of float.
Selected Recent Insider Activity (Form 4)
| Name | Role | Type | Plan |
|---|---|---|---|
| L. Su | CEO | Sell | 10b5-1 (adopted Aug 2025; routine vesting sales) |
| F. Norrod | EVP DC | Sell | 10b5-1 |
| M. Papermaster | CTO | Sell | 10b5-1 |
| J. Hu | CFO | Sell | 10b5-1 |
No material discretionary open-market insider buys recorded in the last 12 months — all dispositions have been Rule 10b5-1 scheduled sales.
10. Risks and Challenges
- Nvidia CUDA ecosystem lock-in. Even at cost/performance parity, switching costs keep customers on Nvidia.
- TSMC concentration. Leading-edge supply is effectively single-sourced. Any Taiwan-specific risk event would hit AMD immediately.
- China export controls. Further restrictions on MI450-class parts highly likely. AMD has already taken a ~$800m inventory/reserve charge on MI308 write-offs.
- Hyperscaler in-house silicon. Custom ASIC projects at Google, Amazon, Microsoft, Meta, Apple and OpenAI (rumoured) structurally cap merchant-GPU TAM.
- Customer concentration. Microsoft, Meta, Oracle and OpenAI together are likely >40% of DC GPU revenue — any one of them pausing an order cycle is material.
- ZT Systems integration. $4.9bn acquisition brings lower-margin revenue and debt; execution risk on the manufacturing divestiture.
- Valuation. 105x trailing GAAP P/E and 13x sales leave no room for an execution miss.
- Gaming / semi-custom decline — PS5/Xbox cycle late.
- Key-person risk. Lisa Su is strongly identified with AMD's execution — her departure would be a significant event.
11. Recent Developments
Last 48 Hours
- 17 Apr 2026 — AMD closed at a fresh all-time high of $281.05. Run driven by ongoing AI data-center demand commentary ahead of Q1 2026 earnings.
- 18–20 Apr 2026 — No new material 8-K filings or press releases. Quiet period ahead of Q1 earnings.
Last 6 Months
- 3 Feb 2026 — Q4 2025 earnings: revenue $7.66bn (+24% YoY); non-GAAP EPS $1.09; FY2025 revenue $34.6bn (+34%) and GAAP net income $4.3bn. Non-GAAP gross margin 54%. Q1 2026 revenue guide ~$7.1bn ± $300m.
- Jan 2026 — CES 2026: Ryzen AI Max and Ryzen 9000X3D refresh launched. Lisa Su said AI PCs would ship in "tens of millions" units in 2026.
- 11 Nov 2025 — Q3 2025 earnings: revenue $9.25bn (record); Data Center segment $4.3bn (+22% YoY).
- 6 Oct 2025 — OpenAI / AMD 6 GW multi-year MI450 deal announced. 8-K discloses warrant for up to 160m shares at $0.01 strike, vesting against GW-deployment and share-price milestones.
- Early Oct 2025 — ZT Systems acquisition closed ($4.9bn cash + stock; announced Aug 2024).
- Sept 2025 — Oracle disclosed a 50,000-unit MI450 order during Oracle AI World.
12. Key Dates Coming Up
- Late Apr / early May 2026 — Q1 2026 earnings (date to be confirmed)
- May 2026 — Annual Shareholder Meeting
- June 2026 — Advancing AI 2026 event (San Jose) — MI450/Helios launch expected
- H2 2026 — MI450 volume production / OpenAI first GW deployed
- H2 2026 — EPYC "Venice" (Zen 6) server CPU launch
- Ongoing — China export-control rulings for MI450-class accelerators
ChartsView tools: Live Charts · Economic Calendar · Forum · Blog
Disclaimer: This research is produced by ChartsView for educational and informational purposes only. It does not constitute financial advice or a recommendation to buy or sell any security. All information is sourced from publicly available company filings, press releases, and official data. ChartsView does not use analyst opinions or third-party ratings. Always conduct your own due diligence and consider your personal financial situation before making investment decisions. Past performance is not indicative of future results.
Loading research report…
13. Thesis Verdict
The central thesis. AMD is a fabless designer of x86 CPUs, GPUs, FPGAs and AI accelerators, with production outsourced largely to TSMC. FY2025 revenue reached $34.6bn, of which the Data Center segment contributed around $16bn (~46%), supported by EPYC server CPU share gains to 30%+ and an Instinct GPU ramp. The structural driver is the AI accelerator cycle: Lisa Su disclosed Data Center GPU revenue exceeded $5bn in 2024 and grew substantially above that in 2025. Near-term catalysts include the MI350 ramp in Q1 2026, the MI450/Helios rack launch in H2 2026, the Oracle 50,000-unit MI450 order, and the OpenAI 6 GW multi-year supply deal signed 6 October 2025.
What would confirm or break it. Confirmation would come from MI450 deployment milestones, ROCm 7 developer traction, continued EPYC share gains, and margin expansion beyond the FY2025 GAAP gross margin of 50%. Materialisation of tighter China export controls on MI450-class parts, further CUDA lock-in, hyperscaler custom-ASIC displacement, ZT Systems integration slippage, or a pause from any of Microsoft, Meta, Oracle or OpenAI (together likely >40% of DC GPU revenue) would weaken the thesis. At ~105x trailing GAAP P/E and ~13x sales, execution misses carry amplified consequences.
Watchpoints
- ConfirmsEvidence supporting the "AI GPU ramp accelerating." thesis continuing to build across subsequent filings.
- InvalidatesMaterialisation of the "Nvidia dominance." risk, or any disclosure that fundamentally alters the capital-return or growth profile stated by management.
- InvalidatesAny disclosure that directly contradicts a material claim in the bull case.
Diagnostic grid
Generated by ChartsView research tooling. Thesis strength measures how well the evidence in this report supports the company's stated thesis — it is NOT a buy/sell rating or price target. ChartsView is not authorised by the FCA to provide regulated investment advice. Generated 23 Apr 2026.