Advanced Micro Devices (AMD): What This Company Actually Does

How AMD makes money

AMD designs semiconductor chips — the "brains" of computers, servers, and AI data centers — but doesn't manufacture them itself. Actual chip production is outsourced to foundries like Taiwan's TSMC (a "fabless" model), letting AMD focus its resources on design and R&D (23% of FY2025 revenue) instead of expensive factories. Finished chips are sold across four end markets: cloud/AI companies, PC makers, console makers, and industrial/automotive customers.

Source: 10-K FY2025, Item 1 "Manufacturing Arrangements," p.10; Note 4, Segment Reporting, p.71.

Where the revenue comes from

Starting in 2025, Client and Gaming are combined into one reportable segment (detailed revenue is still disclosed separately for reference). Embedded is the smallest segment by revenue but the most profitable, at a 36.0% operating margin.

Source: 10-K FY2025, Note 4, Segment Reporting, p.71.

By billing location, the U.S. accounts for 32.8% of FY2025 revenue, China (including Hong Kong) 22.4%, Taiwan 15.0%, Singapore 12.4%, and other regions 17.5%. These figures reflect where invoices are sent, not where end products are ultimately used — the high Taiwan/Singapore/China shares largely reflect manufacturing and assembly hubs, and also mean direct exposure to Taiwan Strait geopolitical risk and U.S.-China export controls.

Source: 10-K FY2025, Note 4, Segment Reporting (revenue by geography), p.72.

Customers and competitors

Nearly all B2B: core customers are cloud/big-tech companies building AI data centers, PC makers (Dell, HP, Lenovo), and console makers (Sony PlayStation, Microsoft Xbox). No single customer accounted for 10% or more of revenue in FY2024 or FY2025 — though in FY2023, one Client-and-Gaming customer (likely a console maker) accounted for 18% of total revenue. Growing AI revenue has diversified the base recently, but dependence on a handful of hyperscale customers could grow again.

• Nvidia — the dominant #1 in AI data-center GPUs; AMD is effectively the #2 challenger.
• Intel — the traditional rival in PC and server x86 CPUs, and the only major competitor that also owns its own manufacturing (foundry).
• Customers' own in-house chips (Google TPU, Amazon Trainium, etc.) — as big-tech customers build more of their own AI chips, they become competitors as much as customers.

Source: 10-K FY2025, Note 4, Segment Reporting, p.72; Item 1, "Competition," p.9.

The metric that matters most in this sector

Data Center segment revenue now accounts for nearly half of AMD's total revenue, and AI-accelerator demand is close to the sole driver of the stock's story. This one segment's growth rate matters more to AMD's investment case than total revenue or profit combined.

Source: 10-K FY2025, Note 4, p.71; 10-K FY2023, Note 4, p.67; 10-Q Q2 FY2026, Note 4, p.34.

Leadership and ownership

Dr. Lisa Su has served as CEO since October 2014 (about 12 years) and as Chairman since February 2022. Before AMD, she was a semiconductor engineer and executive at IBM, Freescale, and Texas Instruments — she is not a founder (AMD was founded by Jerry Sanders in 1969; Su joined in 2012). Combined ownership by all 16 executives and directors is under 0.5% of shares outstanding. The largest shareholders are Vanguard (8.8%) and BlackRock (7.7%) — a professional-management structure with very low insider ownership and large institutional holders as the top shareholders.

Source: DEF 14A (filed Mar 27, 2026), ownership table p.36; board letter, p.17-19.

Capital returns

AMD pays no dividend and has no plans to, per its 10-K, prioritizing reinvestment in growth. Share buybacks totaled $985M (2023), $862M (2024), and $1,316M (2025).

Source: 10-K FY2025, Consolidated Statements of Cash Flows, p.62; Consolidated Statements of Operations (diluted shares), p.58; Item 5, "Issuer Purchases of Equity Securities," p.45.

How this company could fail

• Dependence on TSMC for manufacturing — as a fabless company, AMD relies almost entirely on TSMC for advanced-node chip production; geopolitical tension in the Taiwan Strait translates directly into AMD's production risk.
• U.S.-China export controls — in April 2025, new U.S. government restrictions on AI chip exports to China led AMD to recognize roughly $800 million in inventory-related losses on its MI308 product ($360M of that reversed in a later quarter). With China (including Hong Kong) over 22% of revenue, this exposure could keep shifting with regulatory changes.
• Nvidia's lead plus customers building their own chips — Nvidia dominates the AI GPU market, while AMD's would-be largest customers (big-tech companies) are simultaneously ramping up their own AI chips (Google TPU, Amazon Trainium), which could erode long-term demand.

Source: 10-K FY2025, Item 1A "Risk Factors," p.14; MD&A, p.47 (export-control loss figures).

Five-year financials

Source: 10-K FY2025, Consolidated Statements of Operations (p.58), Cash Flows (p.62), Balance Sheets (p.60); 10-K FY2023, same statements (p.58, 62, 60).

What we still don't know

• When and how much revenue the October 2025 "6-gigawatt GPU supply" agreement with OpenAI will actually generate can't be determined from this data alone — future quarterly results and earnings calls need to confirm it.
• Whether a proposed U.S. government arrangement to take 15% of licensed MI308 revenue to China will actually become formal policy hadn't been finalized as of August 2026.
• How the Data Center segment's 21.7% operating margin compares to Nvidia's can't be assessed from this card alone — it would require separate competitor analysis.