Advanced Micro Devices (AMD)

What it is

Advanced Micro Devices (AMD, NASDAQ: AMD) is a US semiconductor company headquartered in Santa Clara, California, that designs CPUs and GPUs for PCs, servers, and AI data centers. Its two main compute franchises are the EPYC x86 server CPU line, which competes with Intel Xeon, and the Instinct GPU accelerator line, which competes with Nvidia. As of mid-2026, AMD ranks second in AI compute by accelerator revenue. The Radeon line covers consumer and professional graphics. AMD also produces field-programmable gate arrays (FPGAs) and adaptive computing products, a capability acquired with Xilinx in 2022.

History

Jerry Sanders co-founded AMD on 1 May 1969 in Sunnyvale, California, initially as a second-source manufacturer for Intel processor designs. A 1982 cross-licensing agreement with Intel secured AMD's perpetual rights to the x86 instruction-set architecture, the legal and commercial foundation of its CPU business. AMD beat Intel to 64-bit x86 computing in 2003 with the Opteron and Athlon 64 processors, but fell behind in server performance through the late 2000s. The 2006 acquisition of Canadian GPU maker ATI Technologies for US$5.4 billion added a discrete graphics capability. Lisa Su became CEO in October 2014. Her team's Zen CPU microarchitecture, launched in January 2017, restored competitive server performance and drove gross-margin expansion from roughly 32% to above 49% by 2024. AMD closed its US$49 billion acquisition of Xilinx in February 2022, adding FPGAs and the adaptive computing market to its portfolio.

Current state

For full year 2025, AMD reported US$34.6 billion in revenue, up 34% year-on-year, with data-center revenue of US$16.6 billion, up 32%. Non-GAAP net income reached a record US$6.8 billion. EPYC server CPUs held roughly one-third of the x86 server market by unit shipments as of Q4 2025. In AI accelerators, the Instinct MI350X, built on 3nm with 288 GB of HBM3e memory and AMD's CDNA 4 architecture, began shipping to hyperscale partners in Q3 2025, targeting inference workloads where its memory capacity exceeds Nvidia's Blackwell B200. AMD's Helios rack-scale platform pairs 72 MI455X accelerators to target approximately 3 AI exaflops per rack. The MI400-series (CDNA Next architecture) is targeted for second-half 2026. AMD's MI308 GPU for China markets faced new US export restrictions in late 2025, adding a revenue risk to the data-center segment. At AMD's November 2025 analyst day, Lisa Su projected that the global AI data-center market would reach US$1 trillion by 2030 and that AMD's data-center business would grow at roughly 50% compound annually.

Relationships

AMD competes with Nvidia across the AI accelerator spectrum and with Intel in x86 CPUs. On interconnect standards, AMD co-designed UALink for scale-up GPU networking within a rack, detailed in UALink发布四项规范，正式确立对抗NVLink的开放方案, and participates in the Ultra Ethernet Consortium for rack-to-rack AI networking, covered in Ultra Ethernet 1.0 turns commodity Ethernet into an AI back-end fabric. In October 2025, AMD and OpenAI announced a strategic partnership to deploy 6GW of Instinct GPUs across multiple chip generations, AMD's largest single commercial commitment, with the first 1GW of MI450 targeted for 2H 2026; the full context is in AMD的MI450切入OpenAI的6GW协议，对Nvidia开辟第二战线. AMD fabricates its advanced chips at TSMC in Taiwan and sources HBM memory from SK Hynix, Samsung, and Micron. AMD announced a £2 billion UK investment and over US$10 billion in Taiwan commitments in early 2025 to expand R&D and ecosystem relationships.

What to watch

• Whether ROCm, AMD's open-source GPU software stack, narrows CUDA's developer lead enough to hold hyperscaler commitments after the OpenAI contract ramp begins
• MI400 supply execution and competitive pricing against Nvidia's Vera Rubin generation, which will shape AMD's AI revenue trajectory through 2027
• Scope of US export controls on AMD China-market accelerators, where any broadening or narrowing of the 2025 restrictions directly affects data-center revenue
• EPYC server CPU share gains, which increasingly pair with Instinct accelerators in the same AI rack, making CPU and GPU share interdependent in AMD's largest growth vertical