WebGlyphs

Strategic Information Extraction, Processing and Dissemination

Executive Summary

Strategic Synthesis: The Asymmetric Infrastructure War

The competitive relationship between AMD and Broadcom is characterized by a high-stakes, asymmetric conflict centered on the future of Data Center AI infrastructure. While the two companies do not compete in AMD's traditional core markets of x86 general-purpose computing or consumer graphics, they are locked in a structural battle for the capital expenditure budgets of the world's largest hyperscalers.

The fundamental dynamic is one of "Merchant vs. Enabler." AMD is executing a pivot to become the primary merchant alternative to Nvidia, offering end-to-end AI solutions (CPUs, GPUs, and networking). Broadcom, conversely, acts as the critical enabler of the "anti-merchant" trend. By supplying high-end Custom Silicon (XPUs) and dominant networking fabrics, Broadcom empowers hyperscalers to design their own proprietary accelerators, effectively facilitating the vertical integration that threatens AMD's addressable market.

Critical Competitive Dynamics:

• The "Frenemy" Catalyst: Broadcom’s Custom Silicon business is the engine behind the internal silicon initiatives of AMD’s largest customers. Every custom XPU Broadcom designs for a cloud provider potentially displaces a merchant GPU sale for AMD.
• The Battle for the Fabric: As AI clusters scale, the network becomes the computer. AMD is attempting to break into this space via its Pensando and Infinity Fabric technologies, while Broadcom defends a deeply entrenched moat in Ethernet switching, aiming to lock in the interconnect standards that define cluster performance.
• Divergent Software Economics: AMD views software (ROCm) as a necessary investment to unlock hardware sales, often operating as a cost center to break ecosystem lock-in. Broadcom views software (VMware) as a high-margin, recurring revenue engine, creating a financial cushion that supports its hardware R&D.

Structural Advantages & Vulnerabilities: AMD possesses a distinct structural advantage in its diversified compute portfolio; unlike Broadcom, AMD controls the x86 instruction set architecture (ISA) for general-purpose compute, a market Broadcom cannot easily enter. However, AMD faces a critical vulnerability in its supply chain. While Broadcom utilizes a hybrid manufacturing model with some internal capacity, AMD is 100% reliant on external foundries. Furthermore, AMD’s aggressive inventory bets on AI silicon expose it to higher short-term risk compared to Broadcom’s "build-to-order" model for custom silicon.

Strategic Implications: Leadership must recognize that Broadcom is not merely a component supplier but a strategic gatekeeper. Broadcom’s dominance in networking and custom silicon design creates a "pincer" movement: it controls the fabric connecting AMD’s chips while simultaneously helping customers build alternatives to those chips. Success requires AMD to prove that its merchant silicon roadmap can outpace the efficiency of the custom silicon Broadcom enables, while successfully integrating its own networking assets to prevent Broadcom from dictating the architecture of the AI data center.

Report: AMD  AVGO  Comparison  Brief  20260204
Generated: February 04, 2026
AMD Source Filings: 86 SEC documents, dated from January 06, 2021 to February 04, 2026
AVGO Source Filings: 78 SEC documents, dated from January 04, 2021 to January 13, 2026

System Version: WebGlyphs Analytics v4.01

Competitive Assessment: AMD vs. Broadcom

Core Competitive Battlegrounds

AI Infrastructure and Custom Silicon

The primary zone of strategic conflict lies within the Data Center infrastructure market, specifically regarding the silicon that powers AI workloads.

AMD’s Merchant Strategy: AMD has positioned its Data Center segment as a primary growth engine  ^[1], aggressively scaling its AMD Instinct GPU accelerators (MI300/MI350 series) to capture AI workload demand  ^[2]. This strategy relies on a "merchant silicon" model: developing standardized, high-performance products sold to a broad base of cloud and enterprise customers. To support this, AMD has increased R&D expenses by 25% to 97% across recent periods to fund high-end silicon innovation  ^[3].

Broadcom’s Custom Enabler Strategy: Broadcom competes in this same capital expenditure pool through its Semiconductor Solutions segment, specifically via Custom Silicon (ASICs/XPUs)  ^[4]. Unlike AMD’s standardized approach, Broadcom operates a "build-to-order" model, acting as a design and manufacturing partner for hyperscalers' proprietary chips  ^[5].

The Structural Conflict: While AMD sells merchant silicon (GPUs) directly to hyperscalers, Broadcom enables those same hyperscalers to bypass merchant silicon by designing internal "Custom AI accelerators (XPUs)"  ^[6]. This creates a structural conflict where Broadcom's custom silicon business facilitates the "vertical integration by customers" that AMD identifies as a threat to its addressable market  ^[7]. Broadcom explicitly notes that its success depends on developing advanced ASICs that outperform internal efforts in power and bandwidth  ^[8], effectively racing against AMD’s roadmap to prove the value of custom over merchant solutions.

Data Center Networking and Interconnects

Both companies view networking as critical for AI cluster performance, but they approach the architecture from opposing directions.

AMD’s Integrated Fabric: Through the acquisition of Pensando, AMD integrates DPUs into its heterogeneous stack  ^[9], leveraging Infinity Fabric for system interconnects  ^[10]. This strategy aims to create a tightly coupled compute-and-network solution, where the DPU offloads infrastructure tasks to free up the CPU and GPU.

Broadcom’s Ethernet Dominance: Broadcom holds a dominant position in Ethernet switching with its Tomahawk and Jericho silicon families, specifically optimized for AI clusters  ^[11]. Broadcom's strategy emphasizes open-standards Ethernet to resolve cluster bottlenecks  ^[12], directly competing for the interconnect spend within the same data centers AMD targets. Broadcom has shifted its R&D allocation toward low-power, high-bandwidth technologies required by these massive clusters  ^[13], positioning Ethernet as the superior alternative to proprietary interconnects like InfiniBand or potentially AMD's Infinity Fabric in scale-out scenarios.

Rack-Scale AI Solutions

A direct operational overlap is emerging in rack-level integration, as both companies move beyond chip-level sales to capture more value at the system level.

AMD’s Design-Focused Integration: AMD acquired ZT Systems to accelerate the deployment of AI infrastructure at scale. Crucially, AMD employed a "buy-and-divest" strategy, retaining the ZT Design Business to deliver rack-scale solutions  ^[14] while divesting the capital-intensive manufacturing assets to Sanmina  ^[15]. This allows AMD to offer validated reference architectures that speed up the adoption of Instinct GPUs without taking on the low-margin burden of metal bending and assembly.

Broadcom’s Leasing and Systems Model: Broadcom is evolving its business model to explore selling or leasing "AI racks" based on its XPUs, moving beyond component-level sales  ^[16]. This represents a significant shift toward becoming a systems provider. Broadcom explicitly links material risk to the capital expenditure cycles of hyperscalers for these "AI racks"  ^[17], indicating a direct head-to-head competition with AMD for the "unit of compute" in the data center—whether that unit is a GPU card or an entire custom rack.

AMD's Differentiated Positions

General Purpose x86 Computing (Client & Server)

AMD maintains a distinct advantage in general-purpose computing where Broadcom lacks a direct presence. This creates a "moat" for AMD that Broadcom cannot easily cross due to the complexity of the x86 license and ecosystem.

• Server CPUs: AMD's EPYC processors (Zen architecture) are a core revenue driver and market share gainer in cloud infrastructure  ^[18]. The roadmap has segmented further with "Bergamo" for density and "Genoa-X" with 3D V-Cache for technical computing  ^[19]. Broadcom does not produce general-purpose x86 server CPUs, meaning it must partner or coexist with AMD (or Intel) in every server it powers with networking or storage silicon.
• Client Computing: AMD's Client segment, driven by Ryzen processors for desktops and laptops  ^[20], has no equivalent in Broadcom’s portfolio. AMD is further differentiating here by integrating NPUs (Neural Processing Units) into its Ryzen AI 300 Series  ^[21], creating a client-side AI footprint that Broadcom cannot match.

Gaming and Consumer Graphics

AMD's Gaming segment represents a unique market vertical that provides revenue diversity and brand visibility absent from Broadcom's enterprise-focused portfolio.

• Discrete Graphics: The Radeon product line and RDNA architecture target consumer gaming  ^[22].
• Semi-Custom Consoles: AMD powers the PlayStation 5 and Xbox Series X/S  ^[23]. While this sector faces cyclical maturity  ^[24], it provides a steady baseline of volume that Broadcom does not possess. Broadcom has no presence in the high-performance gaming or console processor market.

Adaptive Computing (FPGAs)

Through the acquisition of Xilinx, AMD holds a leadership position in FPGAs and Adaptive SoCs (Versal, Zynq)  ^[25]. This segment allows AMD to address long-lifecycle markets in Aerospace, Automotive, and Industrial sectors  ^[26]. While Broadcom offers programmable Ethernet switching and some industrial solutions, it does not compete in the general-purpose FPGA market. AMD has further evolved this portfolio to focus on edge AI with the Versal Series Gen 2  ^[27], creating an edge-to-cloud AI stack that Broadcom's portfolio does not fully replicate.

Strategic Asymmetries

Software Strategy: Enabler vs. Revenue Driver

The two companies view software through fundamentally different strategic lenses, affecting their margin profiles and customer relationships.

• AMD (Enabler): Views software primarily as an enabler for hardware sales. The ROCm open software stack is critical to compete with CUDA and drive Instinct GPU adoption  ^[28]. AMD invests heavily in this ecosystem—acquiring Mipsology, Nod.ai, and Silo AI  ^[29]—essentially as a cost of doing business to remove barriers to hardware entry.
• Broadcom (Revenue Driver): Treats software as a distinct, high-margin "growth driver" via its Infrastructure Software segment (VMware, Symantec)  ^[30]. Broadcom actively monetizes software through subscriptions, specifically transitioning VMware Cloud Foundation to a subscription model  ^[31]. This creates a recurring revenue stream that buffers semiconductor cyclicality, a financial lever AMD lacks.

Manufacturing Model

• AMD (Pure Play Fabless): Operates a purely fabless model, relying 100% on external partners like TSMC for wafer supply  ^[32]. This provides flexibility but creates a single point of failure. To secure capacity, AMD has had to commit significant capital, reporting $9.4 billion in unconditional purchase commitments by 2025  ^[33].
• Broadcom (Hybrid): Employs a hybrid model. While it relies on TSMC for 95% of advanced logic  ^[34], it retains internal fabrication facilities for proprietary technologies like FBAR filters and III-V materials  ^[35]. This internal capacity creates a defensive moat in its wireless and fiber optic businesses that AMD cannot replicate.

Market Approach: Merchant vs. Custom

• AMD: Primarily pursues a merchant silicon strategy, developing standard products (EPYC, Instinct, Ryzen) sold to a broad customer base. This requires AMD to carry inventory risk; for example, AMD faced ~$800 million in inventory charges in 2025 due to export controls  ^[36].
• Broadcom: In its AI segment, emphasizes a "build-to-order" model for Custom Silicon, acting as a design and manufacturing partner for hyperscalers' proprietary chips  ^[37]. This model mitigates inventory risk, as Broadcom generally lacks long-term capacity commitments and operates primarily on purchase orders  ^[38], shifting the demand risk to the customer.

Competitive Evolution: 2021-2025

2021-2022: Divergent Expansion The competitive relationship began as indirect, with both companies expanding into adjacent markets through massive M&A.

• AMD: Focused on high-performance computing (HPC) and heterogeneous compute. The acquisition of Xilinx (completed Feb 2022) expanded AMD into FPGAs  ^[39], while the acquisition of Pensando (May 2022) brought DPU capabilities to challenge Broadcom's networking dominance  ^[40].
• Broadcom: Focused on software diversification. Broadcom announced the $86 billion acquisition of VMware to anchor its infrastructure software business  ^[41], while its semiconductor business remained focused on networking and storage connectivity.

2023: The AI Pivot and Collision Course As the Generative AI boom accelerated, the companies' strategies began to collide in the data center.

• AMD: Executed a "Pervasive AI" pivot, positioning the Data Center segment as the primary growth driver  ^[42]. The company ramped R&D to support the Instinct MI300 launch, directly targeting the AI accelerator market  ^[43].
• Broadcom: Pivoted its semiconductor strategy from general connectivity to Custom Silicon (XPUs) for AI  ^[44]. The company began emphasizing its role in enabling hyperscalers to build their own AI chips, effectively positioning itself as the alternative to AMD's merchant silicon.

2024-2025: Structural Conflict and Supply Chain Tension The competition intensified as both companies sought to capture the massive AI capital expenditure wave.

• AMD: Faced geopolitical headwinds, taking significant inventory charges due to export controls on its AI chips  ^[45]. To counter Nvidia and Broadcom's ecosystem, AMD acquired ZT Systems in 2025 to offer rack-scale solutions  ^[46].
• Broadcom: Solidified its "dual-engine" strategy. The integration of VMware provided cash flow to support the capital-intensive shift to AI silicon  ^[47]. Broadcom's reliance on TSMC deepened to 95%  ^[48], mirroring AMD's dependency and putting both companies in competition for the same advanced packaging capacity. Broadcom also began exploring the "AI rack" leasing model  ^[49], directly challenging AMD's new rack-scale capabilities.

Overall Strategic Assessment

The relationship between AMD and Broadcom has evolved from tangential coexistence to asymmetric competition focused heavily on the Data Center AI infrastructure stack. While they do not compete in AMD's core x86 CPU or Gaming markets, Broadcom represents a significant "pincer" threat in the AI accelerator space.

Current Competitive Position: Broadcom competes with AMD on two fronts simultaneously:

1. Directly: In data center networking, where Broadcom's Tomahawk/Jericho switches  ^[50] compete with AMD's Pensando DPUs and interconnect strategy for share of wallet.
2. Indirectly (Structural): By enabling AMD's largest cloud customers to design their own Custom Silicon (XPUs)  ^[51], Broadcom facilitates the substitution of AMD's merchant Instinct GPUs.

Forward-Looking Implications: AMD faces a strategic vulnerability in the "frenemy" dynamic of its customer base. As Broadcom scales its custom silicon capabilities, it lowers the barrier for hyperscalers to replace AMD's merchant silicon with internal designs. Broadcom's move into "AI rack" leasing  ^[52] suggests this competition will move up the stack, challenging AMD's ZT Systems strategy.

However, AMD retains a critical advantage in its diversified portfolio. Broadcom cannot offer the general-purpose x86 compute (EPYC) that remains the foundation of the data center  ^[53]. AMD's opportunity lies in leveraging its heterogeneous stack—combining CPU, GPU, and FPGA—to offer total system performance that disjointed custom silicon and networking components cannot easily match.

Both companies are heavily exposed to the same supply chain bottleneck, with AMD relying on TSMC  ^[54] and Broadcom sourcing ~95% of its outsourced wafers from the same foundry  ^[55]. As AMD pivots to "Pervasive AI"  ^[56], Broadcom stands as a formidable gatekeeper of the networking fabric and a key enabler of the internal silicon trend that threatens AMD's merchant accelerator growth.

Audit Trail Table

Back to Methodology

Main Report Text (Top Section) - Strategic narrative with citation numbers in brackets: [1], [2], [3], etc. - Example: "Intel's manufacturing strategy transformed... [1]" - These numbers link to evidence below

In the Audit Trail Table, each numbered entry contains:

1. Citation number - [1], [2], [3]
2. Claim being supported - The specific statement from the main text (often slightly paraphrased/excerpted)
3. Source entries - One or more blocks showing:
   • Document type & date: "10-K::2021-01-22"
   • Hexid: "5388f3375ce4e512" (snippet unique hexadecimal identifier)
   • Snippet text: Extracted passage from SEC filing
   • (More..) link to Google search query derived from snippet

Forward reference (reading report):

• See [29] in main text ⇒ Jump to appendix entry [29] and check supporting snippets
  • Then either jump back to [29] location in text
  • Or use the "More.." Google Search Feature to get the source document and related context. In most cases, clicking on "Read more" on the Google entry for the original SEC document will return the exact original text of the snippet

Additional Formats:

• HTML version: Interactive document with bidirectional links between narrative claims and supporting evidence. Recommended for manual verification workflows.

• Chronological audit trail: All evidence organized by filing date rather than thematic structure. Available as:

  • HTML for browsing
  • JSON for programmatic access and LLM integration (see example use cases)

[1]   AMD has positioned its Data Center segment as a primary growth engine AMD::; 10-K::2026-02-04 30a551a8c054122f Our investments in technologies such as our custom ready chiplet platform and AMD Infinity Fabrictm switch position us to maintain our leadership as a custom design silicon provider of choice. Our Strategy We believe AI is shaping the next era of computing and its full potential will be realized when it becomes pervasive across cloud edge and endpoint devices. With our compute engines intellectual property software enablement and deep expertise AMD is positioned to lead in this next computing era. (More..)

[2]   aggressively scaling its AMD Instinct GPU accelerators (MI300/MI350 series) to capture AI workload demand AMD::; 10-K::2024-01-31 9a9ac284abe4595e In our Data Center GPU business demand for our Data Center GPUs products was very strong as we had large hyperscaler customers committed to deploy our next generation AMD Instinct MI300 accelerators. (More..) AMD::; 10-K::2026-02-04 31a203dfb80c3740 A priority in 2025 was accelerating growth in the Data Center segment. Demand for our data center AI GPU products was strong as large hyperscale customers OEMs and ODMs deployed our AMD Instinct MI350X Series GPUs. We advanced our AMD AI GPU roadmap to deliver an annual cadence of leadership for AMD Instinct solutions beginning with the AMD Instinct MI350 Series GPUs in 2025. (More..)

[3]   This strategy relies on a "merchant silicon" model: developing standardized, high-performance products sold to a broad base of cloud and enterprise customers. To support this, AMD has increased R&D expenses by 25% to 97% across recent periods to fund high-end silicon innovation AMD::; 10-Q::2021-07-28 503c72dac95bbb82 Gross margin was 47 and 45 for the six months ended June 26 2021 and June 27 2020 respectively. The increase for both periods was primarily driven by a richer mix of sales including high end Ryzen Radeon and EPYC processor sales. Expenses Research and Development Expenses Research and development expenses of $659M for the three months ended June 26 2021 increased by $199M or 43 compared to $460M for the prior year period. (More..)

[4]   Broadcom competes in this same capital expenditure pool through its Semiconductor Solutions segment, specifically via Custom Silicon (ASICs/XPUs) AVGO::; 10-K::2022-12-16 202b8f9cd9ba2bf6 Inductive Charging ASICs Our custom inductive charging ASIC devices offer high efficiency and are highly integrated solutions for mobile and wearable devices. SAS RAID PCIe Products We provide serial attached small computer system interface SAS and redundant array of independent disks RAID controller and adapter solutions to server and storage system original equipment manufacturers OEMs. (More..)

[5]   Unlike AMD’s standardized approach, Broadcom operates a "build-to-order" model, acting as a design and manufacturing partner for hyperscalers' proprietary chips AVGO::; 10-Q::2021-03-12 7eb7d14c9258c997 We have also taken various actions to de risk our business in light of the ongoing uncertainty. For example we are largely building semiconductor products to order instead of based on customer forecasts. In addition we have continued to strengthen our balance sheet including closely managing working capital and our debt instruments. (More..)

[6]   While AMD sells merchant silicon (GPUs) directly to hyperscalers, Broadcom enables those same hyperscalers to bypass merchant silicon by designing internal "Custom AI accelerators (XPUs)" AVGO::; 10-Q::2025-09-10 fceb87d3fbcb1c42 The loss of or any substantial reduction in sales to any of our top customers including our customers for our custom AI accelerators XPUs or AI racks based on our XPUs could have a material adverse effect on our business financial condition results of operations and cash flows. (More..)

[7]   This creates a structural conflict where Broadcom's custom silicon business facilitates the "vertical integration by customers" that AMD identifies as a threat to its addressable market AVGO::; 10-K::2025-12-18 4097e71e95bbd407 Competitors in semiconductor solutions include integrated device manufacturers fabless semiconductor companies and the internal resources of large integrated OEMs. (More..) AMD::; 10-K::2026-02-04 32e2cbcfb226e5b6 A variety of smaller fabless silicon companies offer proprietary accelerator solutions and Arm r based CPUs targeting data center use cases. In addition some of our customers are internally developing their own data center microprocessor products and accelerator products which could impact the available market for our products. Competition in Client and Gaming Segment Our primary competitor in the supply of CPUs and APUs is Intel. (More..)

[8]   Broadcom explicitly notes that its success depends on developing advanced ASICs that outperform internal efforts in power and bandwidth AVGO::; 10-K::2024-12-20 9b8bc485738fa39c In addition to compete successfully in the semiconductor industry we must continue to develop and respond to technological advancements and requirements such as low power consumption higher bandwidth and increase in the number of clusters. Failure to successfully develop increasingly advanced technologies including ASICs such as custom AI accelerators or XPUs and other AI related products could impair our competitive position. (More..)

[9]   Through the acquisition of Pensando, AMD integrates DPUs into its heterogeneous stack AMD::; 10-Q::2022-08-03 4e28caf9998388ac From time to time we may also sell or license portions of our intellectual property IP portfolio. On May 26 2022 we completed the acquisition of Pensando Systems Inc. Pensando for a transaction valued at approximately $1.9B. The recorded purchase consideration of $1.7B is net of deferred cash compensation requiring future services and other customary closing adjustments. (More..)

[10]   leveraging Infinity Fabric for system interconnects ; :: (More..)

[11]   Broadcom holds a dominant position in Ethernet switching with its Tomahawk and Jericho silicon families, specifically optimized for AI clusters AVGO::; 10-K::2025-12-18 7647438fc8d8b6a7 Custom Silicon Solutions We provide advanced technology and intellectual property IP platforms for customers to design and develop application specific integrated circuits ASICs for AI and high performance computing networking and storage applications. Our custom silicon solutions provide the platform to integrate embedded logic high bandwidth memory serializer deserializer technology IP cores and processor cores using advanced packaging technologies. (More..) AVGO::; 10-K::2024-12-20 474d931e34898065 We provide semiconductor solutions for managing the movement of data in data center service provider and enterprise networking applications. Our products offer an enhanced open standards based Ethernet network interface card NIC and switching solution to resolve connectivity bottlenecks in data centers particularly in AI data centers where compute bandwidth and cluster sizes grow rapidly. (More..)

[12]   Broadcom's strategy emphasizes open-standards Ethernet to resolve cluster bottlenecks AVGO::; 10-K::2024-12-20 474d931e34898065 We provide semiconductor solutions for managing the movement of data in data center service provider and enterprise networking applications. Our products offer an enhanced open standards based Ethernet network interface card NIC and switching solution to resolve connectivity bottlenecks in data centers particularly in AI data centers where compute bandwidth and cluster sizes grow rapidly. (More..)

[13]   directly competing for the interconnect spend within the same data centers AMD targets. Broadcom has shifted its R&D allocation toward low-power, high-bandwidth technologies required by these massive clusters AVGO::; 10-Q::2025-09-10 1357bf5da917973e In addition to compete successfully in the semiconductor industry we must continue to develop and respond to technological advancements and requirements such as low power consumption higher bandwidth and large compute clusters and evolve our business strategy or adopt new business models to address the needs and challenges of our customers. (More..)

[14]   AMD acquired ZT Systems to accelerate the deployment of AI infrastructure at scale. Crucially, AMD employed a "buy-and-divest" strategy, retaining the ZT Design Business to deliver rack-scale solutions AMD::; 10-K::2026-02-04 17c888c7b24ee684 In March 2025 we acquired ZT Group Int l Inc. ZT Systems where we retained certain intellectual property and employees associated with the design operations ZT Design Business. This acquisition enables us to deliver end to end AI solutions and accelerate the design and deployment of AMD powered AI infrastructure at scale optimized for the cloud. In October 2025 we sold the ZT data center infrastructure manufacturing business ZT Manufacturing Business to Sanmina Corporation Sanmina. (More..)

[15]   while divesting the capital-intensive manufacturing assets to Sanmina AMD::; 10-K::2026-02-04 217dd0b8e54e7551 Net cash provided by investing activities of discontinued operations in 2025 was $1.3B primarily from the sale of the ZT Manufacturing Business. (More..)

[16]   Broadcom is evolving its business model to explore selling or leasing "AI racks" based on its XPUs, moving beyond component-level sales AVGO::; 10-Q::2025-09-10 7093b2f2118f1fe6 To remain competitive we seek to evolve our business strategy or adopt new business models from time to time such as the sale or leasing of AI racks based on our XPUs that require significant financial resources which could have a material adverse effect on our results of operations. (More..)

[17]   This represents a significant shift toward becoming a systems provider. Broadcom explicitly links material risk to the capital expenditure cycles of hyperscalers for these "AI racks" AVGO::; 10-Q::2025-09-10 78f640e2c974f4f8 This customer concentration increases the risk of quarterly fluctuations in our operating results and our sensitivity to any material adverse developments experienced by these customers including their ability to invest in access operate or pay for their AI infrastructure. (More..)

[18]   Server CPUs: AMD's EPYC processors (Zen architecture) are a core revenue driver and market share gainer in cloud infrastructure AMD::; 10-Q::2023-08-02 14013d966b5f9594 The decrease in operating and net income was primarily due to lower Client segment performance. We introduced a number of new products during the second quarter of 2023 including the 4th Gen EPYCtm 97X4 processors for cloud native computing and 4th Gen EPYC processors with AMD 3D V Cachetm technology for technical computing. (More..) AMD::; 10-K::2021-01-29 2b1616caf91e645a In April 2020 we announced the extension of the 2nd Gen AMD EPYC processor family with three new processors AMD EPYC 7F32 8 cores AMD EPYC 7F52 16 cores and AMD EPYC 7F72 24 cores. These new processors leverage up to 500 MHz of additional base frequency and large amounts of cache. In October 2020 we announced the AMD EPYCtm processor based Azure Dav4 Eav4 39 Easv4 and Lsv2 VMs for use to improve real time analysis on large volumes of data streaming from applications websites and more. (More..)

[19]   The roadmap has segmented further with "Bergamo" for density and "Genoa-X" with 3D V-Cache for technical computing AMD::; 10-K::2024-01-31 b1a664c1c3a19fd4 The AMD EPYC 97x4 cloud native optimized data center CPUs formerly codenamed Bergamo are built with our Zen 4c architecture core and further extend the EPYC 9004 Series of processors to deliver the thread density and scale needed for cloud native computing. (More..)

[20]   Client Computing: AMD's Client segment, driven by Ryzen processors for desktops and laptops AMD::; 10-K::2025-02-05 0aabe8a55eed1dd4 We took a major step in our AI PC roadmap with the launch of AMD Ryzen AI 300 Series processors that combine leadership compute capabilities based on our Zen 5 architecture and an industry leading neural processing unit NPU powered by our XDNA 2 architecture for next generation AI PCs. We added to our Ryzen family of desktop CPUs with the Ryzen 9000 series processors for laptop and desktop PCs that deliver leadership performance in gaming productivity and content creation. (More..)

[21]   has no equivalent in Broadcom’s portfolio. AMD is further differentiating here by integrating NPUs (Neural Processing Units) into its Ryzen AI 300 Series AMD::; 10-K::2026-02-04 3839cf4c7f0db92a We also offer data center rack scale platform designs that incorporate AMD data center products to meet the growing performance demands of AI supercomputers and machine learning workloads. In client computing our CPUs APUs and chipsets for desktops and notebooks deliver performance efficiency AI capabilities and modern security features for gamers creators consumers and enterprises. AMD was the first company to integrate a dedicated neural processing unit NPU on the same SoC as an x86 CPU for AI PCs. (More..)

[22]   Discrete Graphics: The Radeon product line and RDNA architecture target consumer gaming AMD::; 10-K::2021-01-29 675e363ab7e1b3d4 The AMD Radeon RX 6000 series includes the AMD Radeon RX 6800 and Radeon RX 6800 XT graphics cards as well as the Radeon RX 6900 XT built upon the AMD RDNA 2 gaming architecture that spans from game consoles to PCs. The AMD RDNA 2 GPUs feature enhancements in the compute unit advancements in the visual pipeline and the introduction of a new high speed cache called AMD Infinity Cache. These architecture enhancements enable ultra high performance and ultra high fidelity in the latest games. (More..)

[23]   Semi-Custom Consoles: AMD powers the PlayStation 5 and Xbox Series X/S AMD::; 10-K::2022-02-03 1c2236dc3cf72618 Our semi custom products are tailored co developed high performance customer specific solutions based on AMD CPU GPU and multi media technologies. We work closely with our customers to define solutions to precisely match the requirements of the device or application. We developed the semi custom SoC products that power both the Sony PlayStation r 5 as well as the Microsoft r Xbox Series Xtm and Microsoft r Xbox Series Stm game consoles. (More..)

[24]   While this sector faces cyclical maturity AMD::; 10-K::2024-01-31 48f355088e6cfd41 Against the backdrop of a mixed demand environment net revenue for 2023 was $22.7B a decrease of 4 compared to 2022 net revenue of $23.6B. The decrease in net revenue was primarily due to a 25 decrease in Client segment revenue primarily due to lower processor sales and a 9 decrease in Gaming segment revenue primarily due to lower semi custom product sales. (More..)

[25]   Through the acquisition of Xilinx, AMD holds a leadership position in FPGAs and Adaptive SoCs (Versal, Zynq) AMD::; 10-K::2023-02-27 4a784acf63577323 Xilinx which expanded our technology and product portfolio to include adaptable hardware platforms that enable hardware acceleration and rapid innovation across a variety of technologies and established AMD in multiple embedded markets where we have traditionally not had a significant presence. We now offer Field Programmable Gate Arrays FPGAs Adaptive SoCs and Adaptive Compute Acceleration Platform ACAP products. (More..) AMD::; 10-K::2024-01-31 4bbb2fca4c37632e Our product brands for Adaptive SoCs are Zynq 7000 Zynq UltraScale MPSoC Zynq UltraScale RFSoCs Versal HBM Versal Premium Versal Prime Versal AI Core Versal AI Edge Vitis and Vivado. Our product brand for System on Module SOM is Kria. Our compute and network acceleration board products are sold under the Alveo and Pensando brands. We market our products through direct marketing and co marketing programs. (More..)

[26]   This segment allows AMD to address long-lifecycle markets in Aerospace, Automotive, and Industrial sectors AMD::; 10-K::2023-02-27 4860c4a95af6a7d6 Our FPGA and Adaptive SOC products are sold to customers in a very wide range of markets such as Aerospace and Defense Test and Measurement Industrial Automotive Consumers Broadcast Communication Infrastructure and Data Center. For these products we either sell directly to our customers or through a network of distributors and OEM partners. We are also developing a network of Value Added Resellers VARs and Integrated Solution Vendors ISVs for our Alveo products. (More..)

[27]   While Broadcom offers programmable Ethernet switching and some industrial solutions, it does not compete in the general-purpose FPGA market. AMD has further evolved this portfolio to focus on edge AI with the Versal Series Gen 2 AMD::; 10-K::2025-02-05 3d5c257d4ccf9241 We expanded our adaptive computing portfolio with differentiated solutions with the launch of the new Versaltm Series Gen 2 devices including the new Versal AI Edge Series Gen 2 and Versal Prime Series Gen 2 adaptive SoCs which bring preprocessing AI inference and postprocessing together in a single device for end to end acceleration of AI driven embedded systems. (More..)

[28]   AMD (Enabler): Views software primarily as an enabler for hardware sales. The ROCm open software stack is critical to compete with CUDA and drive Instinct GPU adoption AMD::; 10-K::2024-01-31 92e5b0d913dafca7 We enhanced the performance and features of our AMD RoCmtm software by releasing our latest AMD ROCm 6 open software platform for AI and HPC workloads. We expanded our Embedded processor portfolio with powerful scalable offerings for a variety of embedded applications such as the AMD Ryzentm Embedded 7000 Series processor family. (More..) AMD::; 10-K::2021-01-29 19d59898f2c18565 In the data center our principal competitor is Nvidia as the adoption of its proprietary CUDA software platform established its market share in HPC and machine learning. Another competitor is Intel as it builds products for acceleration in the data center such as Intel Xe or Habana AI processors. Other competitors include numerous deep learning accelerator companies consisting mostly of early to late stage start ups. (More..)

[29]   AMD invests heavily in this ecosystem—acquiring Mipsology, Nod.ai, and Silo AI AMD::; 10-K::2024-01-31 3b28e9d43cd70905 We achieve this through our family of CPUs GPUs FPGAs and Adaptive SoCs. With the acquisitions of Mipsology SAS and Nod Inc.in 2023 we expanded our AI software capabilities to accelerate our AI growth strategy centered on an open software ecosystem to help lower the barriers of entry for customers through developer tools libraries and models. We develop world class software platforms that are used to enable our high performance products. (More..) AMD::; 10-K::2025-02-05 4e143b9419ea630c We also made strategic investments to further expand our AI software capabilities with the acquisition of Silo AI Oy Silo AI an AI lab based in Finland. The acquisition of Silo AI enables customers to accelerate development and deployment of AI models on AMD hardware. (More..)

[30]   Broadcom (Revenue Driver): Treats software as a distinct, high-margin "growth driver" via its Infrastructure Software segment (VMware, Symantec) AVGO::; 10-K::2023-12-14 58533c00e17e187a During fiscal years 2022 and 2021 other charges included impairment and disposal charges of $ 7M and $ 16M respectively primarily related to leasehold improvements.15. Subsequent Events Acquisition of VMware Inc. On November 22 2023 we completed the acquisition of VMware in a cash and stock transaction the VMware Merger. (More..)

[31]   Broadcom actively monetizes software through subscriptions, specifically transitioning VMware Cloud Foundation to a subscription model AVGO::; 10-Q::2025-03-12 f93c04ca7500631b Net revenue from our infrastructure software segment increased in the fiscal quarter ended February 2 2025 compared to the prior year fiscal period primarily due to strong demand for our VMware Cloud Foundation VCF product additional license revenue recognized on contracts where customers do not have the right to terminate and the transition to a subscription license model. (More..)

[32]   AMD (Pure Play Fabless): Operates a purely fabless model, relying 100% on external partners like TSMC for wafer supply AMD::; 10-K::2021-01-29 1a24551bc2f2debc For the production of wafers for certain products including the production of all our 7 nanometer nm products we use Taiwan Semiconductor Manufacturing Company Limited TSMC. We purchase wafers for all our CPU and APU products and wafers for a certain portion of our GPU products manufactured at process nodes larger than 7 nm with limited exceptions from Inc. GF. We also rely on third party manufacturers to assemble test mark and pack ATMP our products. (More..)

[33]   This provides flexibility but creates a single point of failure. To secure capacity, AMD has had to commit significant capital, reporting $9.4 billion in unconditional purchase commitments by 2025 AMD::; 10-Q::2025-08-06 280731b1a28153c5 We have $3.0B available under an unsecured revolving credit facility that expires on April 29 2027. No funds were drawn from this credit facility during the three months ended June 28 2025. As of June 28 2025 we had unconditional purchase commitments of approximately $9.4B of which $5.5B are for the remainder of fiscal year 2025. We work continually with our suppliers and partners on the timing of payments and deliveries of purchase commitments taking into account business conditions. (More..)

[34]   Broadcom (Hybrid): Employs a hybrid model. While it relies on TSMC for 95% of advanced logic AVGO::; 10-K::2021-12-17 2478180969a160d9 As lead times to identify qualify and establish reliable production at acceptable yields with a new CM is typically lengthy there is often no readily available alternative source and there may be other constraints on our ability to change CMs. In addition qualifying such CMs is often expensive and they may not produce products as cost effectively as our current suppliers. TSMC one of our CMs manufactured approximately 89 of the wafers manufactured by our CMs during fiscal year 2021. (More..)

[35]   it retains internal fabrication facilities for proprietary technologies like FBAR filters and III-V materials AVGO::; 10-K::2022-12-16 3050e75ea766ab4e Although we operate a primarily outsourced manufacturing business model we also rely on our own manufacturing facilities in particular in Fort Collins Colorado Singapore and Breinigsville Pennsylvania. We use these internal fabrication facilities for products utilizing our innovative and proprietary processes. (More..) AVGO::; 10-K::2022-12-16 3525d07cc715512c Our Fort Collins and Breinigsville facilities are the sole sources for the FBAR components used in many of our wireless devices and for the indium phosphide based wafers used in our fibre optics products respectively. Many of our facilities and those of our CMs and suppliers are located in California and the Pacific Rim region which have above average seismic activity and severe weather activity. (More..)

[36]   AMD: Primarily pursues a merchant silicon strategy, developing standard products (EPYC, Instinct, Ryzen) sold to a broad customer base. This requires AMD to carry inventory risk; for example, AMD faced ~$800 million in inventory charges in 2025 due to export controls AMD::; 10-Q::2025-08-06 260536c6ac4c0811 Gross margin was 45 and 48 for the six months ended June 28 2025 and June 29 2024 respectively. The decrease in gross margin in both periods was primarily due to approximately $800M of inventory and related charges associated with the U.S. government export control on AMD Instinct MI308 Data Center GPU products. (More..)

[37]   Broadcom: In its AI segment, emphasizes a "build-to-order" model for Custom Silicon, acting as a design and manufacturing partner for hyperscalers' proprietary chips AVGO::; 10-Q::2021-03-12 7eb7d14c9258c997 We have also taken various actions to de risk our business in light of the ongoing uncertainty. For example we are largely building semiconductor products to order instead of based on customer forecasts. In addition we have continued to strengthen our balance sheet including closely managing working capital and our debt instruments. (More..) AVGO::; 10-K::2022-12-16 202b8f9cd9ba2bf6 Inductive Charging ASICs Our custom inductive charging ASIC devices offer high efficiency and are highly integrated solutions for mobile and wearable devices. SAS RAID PCIe Products We provide serial attached small computer system interface SAS and redundant array of independent disks RAID controller and adapter solutions to server and storage system original equipment manufacturers OEMs. (More..)

[38]   This model mitigates inventory risk, as Broadcom generally lacks long-term capacity commitments and operates primarily on purchase orders AVGO::; 10-Q::2022-06-09 14f23a6f9433815b We do not generally have long term capacity commitments with our CMs and substantially all of our manufacturing services are on a purchase order basis with no minimum quantities. Further our CMs may fail to timely develop new advanced manufacturing processes including transitions to smaller geometry process technologies or from time to time will cease to or will become unable to manufacture a component for us. (More..)

[39]   AMD: Focused on high-performance computing (HPC) and heterogeneous compute. The acquisition of Xilinx (completed Feb 2022) expanded AMD into FPGAs AMD::; 10-Q::2022-08-03 0ba44e1f6fa874fb Xilinx Acquisition On February 14 2022 the Company completed the acquisition of all issued and outstanding shares of Xilinx a leading provider of adaptive computing solutions for a total purchase consideration of $ 48.8B $ 46.4B net of cash acquired of $ 2.4B. The acquisition of Xilinx expands the Company s product portfolio to include adaptable hardware platforms that enable hardware acceleration and rapid innovation across a variety of technologies. (More..)

[40]   while the acquisition of Pensando (May 2022) brought DPU capabilities to challenge Broadcom's networking dominance AMD::; 10-Q::2022-08-03 4e28caf9998388ac From time to time we may also sell or license portions of our intellectual property IP portfolio. On May 26 2022 we completed the acquisition of Pensando Systems Inc. Pensando for a transaction valued at approximately $1.9B. The recorded purchase consideration of $1.7B is net of deferred cash compensation requiring future services and other customary closing adjustments. (More..)

[41]   Broadcom: Focused on software diversification. Broadcom announced the $86 billion acquisition of VMware to anchor its infrastructure software business AVGO::; 10-K::2023-12-14 7785c2b1c6d16112 The preliminary total purchase consideration for the VMware Merger was approximately $86.3B. We funded the cash portion of the VMware Merger consideration with net proceeds from the issuance of $30.4B in term loans under a credit agreement that we entered into on August 15 2023 as well as cash on hand. We assumed all outstanding VMware restricted stock unit RSU awards and performance stock unit awards held by continuing employees. (More..)

[42]   AMD: Executed a "Pervasive AI" pivot, positioning the Data Center segment as the primary growth driver AMD::; 10-K::2026-02-04 30a551a8c054122f Our investments in technologies such as our custom ready chiplet platform and AMD Infinity Fabrictm switch position us to maintain our leadership as a custom design silicon provider of choice. Our Strategy We believe AI is shaping the next era of computing and its full potential will be realized when it becomes pervasive across cloud edge and endpoint devices. With our compute engines intellectual property software enablement and deep expertise AMD is positioned to lead in this next computing era. (More..)

[43]   The company ramped R&D to support the Instinct MI300 launch, directly targeting the AI accelerator market AMD::; 10-K::2024-01-31 9a9ac284abe4595e In our Data Center GPU business demand for our Data Center GPUs products was very strong as we had large hyperscaler customers committed to deploy our next generation AMD Instinct MI300 accelerators. (More..)

[44]   Broadcom: Pivoted its semiconductor strategy from general connectivity to Custom Silicon (XPUs) for AI AVGO::; 10-K::2022-12-16 202b8f9cd9ba2bf6 Inductive Charging ASICs Our custom inductive charging ASIC devices offer high efficiency and are highly integrated solutions for mobile and wearable devices. SAS RAID PCIe Products We provide serial attached small computer system interface SAS and redundant array of independent disks RAID controller and adapter solutions to server and storage system original equipment manufacturers OEMs. (More..)

[45]   AMD: Faced geopolitical headwinds, taking significant inventory charges due to export controls on its AI chips AMD::; 10-Q::2025-08-06 260536c6ac4c0811 Gross margin was 45 and 48 for the six months ended June 28 2025 and June 29 2024 respectively. The decrease in gross margin in both periods was primarily due to approximately $800M of inventory and related charges associated with the U.S. government export control on AMD Instinct MI308 Data Center GPU products. (More..)

[46]   To counter Nvidia and Broadcom's ecosystem, AMD acquired ZT Systems in 2025 to offer rack-scale solutions AMD::; 10-Q::2025-11-05 456e3741f33c5a60 NOTE 5 Acquisitions and Divestitures ZT Systems Acquisition On March 31 2025 the Acquisition Date the Company completed the acquisition of all issued and outstanding shares of ZT Systems for a total purchase consideration of $ 4.4B. ZT Systems is a provider of AI and general purpose compute infrastructure for hyperscale computing companies. (More..)

[47]   Broadcom: Solidified its "dual-engine" strategy. The integration of VMware provided cash flow to support the capital-intensive shift to AI silicon AVGO::; 10-K::2025-12-18 63946720fd8209ce As a percentage of net revenue gross margin was 68 and 63 of net revenue for the fiscal years 2025 and 2024 respectively. The increase was primarily due to higher revenue impact on margin and higher infrastructure software gross margin percentage driven by an increase in license revenue and lower infrastructure software labor costs following our integration of the VMware business. (More..)

[48]   Broadcom's reliance on TSMC deepened to 95% AVGO::; 10-K::2021-12-17 2478180969a160d9 As lead times to identify qualify and establish reliable production at acceptable yields with a new CM is typically lengthy there is often no readily available alternative source and there may be other constraints on our ability to change CMs. In addition qualifying such CMs is often expensive and they may not produce products as cost effectively as our current suppliers. TSMC one of our CMs manufactured approximately 89 of the wafers manufactured by our CMs during fiscal year 2021. (More..)

[49]   mirroring AMD's dependency and putting both companies in competition for the same advanced packaging capacity. Broadcom also began exploring the "AI rack" leasing model AVGO::; 10-Q::2025-09-10 7093b2f2118f1fe6 To remain competitive we seek to evolve our business strategy or adopt new business models from time to time such as the sale or leasing of AI racks based on our XPUs that require significant financial resources which could have a material adverse effect on our results of operations. (More..)

[50]   1. Directly: In data center networking, where Broadcom's Tomahawk/Jericho switches AVGO::; 10-K::2025-12-18 7647438fc8d8b6a7 Custom Silicon Solutions We provide advanced technology and intellectual property IP platforms for customers to design and develop application specific integrated circuits ASICs for AI and high performance computing networking and storage applications. Our custom silicon solutions provide the platform to integrate embedded logic high bandwidth memory serializer deserializer technology IP cores and processor cores using advanced packaging technologies. (More..)

[51]   2. Indirectly (Structural): By enabling AMD's largest cloud customers to design their own Custom Silicon (XPUs) AVGO::; 10-K::2022-12-16 202b8f9cd9ba2bf6 Inductive Charging ASICs Our custom inductive charging ASIC devices offer high efficiency and are highly integrated solutions for mobile and wearable devices. SAS RAID PCIe Products We provide serial attached small computer system interface SAS and redundant array of independent disks RAID controller and adapter solutions to server and storage system original equipment manufacturers OEMs. (More..)

[52]   AMD faces a strategic vulnerability in the "frenemy" dynamic of its customer base. As Broadcom scales its custom silicon capabilities, it lowers the barrier for hyperscalers to replace AMD's merchant silicon with internal designs. Broadcom's move into "AI rack" leasing AVGO::; 10-Q::2025-09-10 7093b2f2118f1fe6 To remain competitive we seek to evolve our business strategy or adopt new business models from time to time such as the sale or leasing of AI racks based on our XPUs that require significant financial resources which could have a material adverse effect on our results of operations. (More..)

[53]   However, AMD retains a critical advantage in its diversified portfolio. Broadcom cannot offer the general-purpose x86 compute (EPYC) that remains the foundation of the data center AMD::; 10-Q::2023-08-02 14013d966b5f9594 The decrease in operating and net income was primarily due to lower Client segment performance. We introduced a number of new products during the second quarter of 2023 including the 4th Gen EPYCtm 97X4 processors for cloud native computing and 4th Gen EPYC processors with AMD 3D V Cachetm technology for technical computing. (More..)

[54]   Both companies are heavily exposed to the same supply chain bottleneck, with AMD relying on TSMC AMD::; 10-K::2021-01-29 1a24551bc2f2debc For the production of wafers for certain products including the production of all our 7 nanometer nm products we use Taiwan Semiconductor Manufacturing Company Limited TSMC. We purchase wafers for all our CPU and APU products and wafers for a certain portion of our GPU products manufactured at process nodes larger than 7 nm with limited exceptions from Inc. GF. We also rely on third party manufacturers to assemble test mark and pack ATMP our products. (More..)

[55]   and Broadcom sourcing ~95% of its outsourced wafers from the same foundry AVGO::; 10-K::2021-12-17 2478180969a160d9 As lead times to identify qualify and establish reliable production at acceptable yields with a new CM is typically lengthy there is often no readily available alternative source and there may be other constraints on our ability to change CMs. In addition qualifying such CMs is often expensive and they may not produce products as cost effectively as our current suppliers. TSMC one of our CMs manufactured approximately 89 of the wafers manufactured by our CMs during fiscal year 2021. (More..)

[56]   As AMD pivots to "Pervasive AI" AMD::; 10-K::2026-02-04 30a551a8c054122f Our investments in technologies such as our custom ready chiplet platform and AMD Infinity Fabrictm switch position us to maintain our leadership as a custom design silicon provider of choice. Our Strategy We believe AI is shaping the next era of computing and its full potential will be realized when it becomes pervasive across cloud edge and endpoint devices. With our compute engines intellectual property software enablement and deep expertise AMD is positioned to lead in this next computing era. (More..)