Advanced Micro Devices, Inc. (AMD): Business Model Canvas [June-2026 Updated]

This ready-made Business Model Canvas gives you a practical, research-based view of Advanced Micro Devices, Inc. Business, showing how it creates value through high-performance CPUs, GPUs, and AI accelerators, and captures it across data center, client, gaming, and embedded markets. You will see the core drivers behind its strategy: TSMC 2nm/3nm manufacturing access, advanced packaging, ROCm 7, semiconductor engineering talent, and partnerships with OpenAI, Microsoft Azure, Oracle OCI, Google Cloud, Tencent Cloud, Dell, HP, Lenovo, ASE, and Amkor. It also breaks down the main revenue streams from data center chips, Ryzen, Radeon, semi-custom, and embedded products, along with the biggest cost pressures from R&D, wafer and packaging costs, HBM and DRAM, and supply-chain support. Use it to quickly understand Advanced Micro Devices, Inc. Business's customers, channels, and operating model for coursework, case studies, presentations, or research.

Advanced Micro Devices, Inc. - Canvas Business Model: Key Partnerships

Advanced Micro Devices, Inc. depends on a small group of high-impact partners for wafer fabrication, packaging, cloud distribution, and finished-system reach. The most visible late-2025 demand signal is the OpenAI agreement for 6 gigawatts of AMD GPU deployments, starting with 1 gigawatt and a warrant for up to 160 million shares of AMD common stock.

TSMC foundry and advanced packaging

TSMC is the manufacturing choke point in Advanced Micro Devices, Inc.'s model. Advanced Micro Devices, Inc. designs chips and relies on TSMC for wafer production and advanced packaging, including portfolio parts on 5 nm, 4 nm, and 3 nm process nodes. That matters because AMD's CPU chiplets and AI accelerators are only as good as the capacity, yield, and packaging throughput behind them. If TSMC tightens allocation, Advanced Micro Devices, Inc. can ship fewer parts or delay launches. If yield improves, Advanced Micro Devices, Inc. can spread fixed design costs over more units, which supports margin.

• Wafer supply affects launch timing.
• Advanced packaging affects chiplet integration and memory integration.
• TSMC concentration increases supplier risk but gives Advanced Micro Devices, Inc. access to leading-edge nodes.

OpenAI AI compute agreement

The OpenAI agreement is the clearest late-2025 example of a customer partnership becoming a strategic infrastructure commitment. Advanced Micro Devices, Inc. disclosed 6 gigawatts of GPU deployments, with the first 1 gigawatt planned for the second half of 2026 using AMD Instinct MI450 Series systems. Advanced Micro Devices, Inc. also disclosed a warrant for up to 160 million shares of AMD common stock. That structure ties hardware demand, rollout milestones, and equity economics into one deal. For your academic work, this is a strong example of how AI demand can be linked to long-range chip supply rather than spot purchases.

• 6 gigawatts is the central demand number.
• 1 gigawatt starts the rollout in the second half of 2026.
• 160 million shares is the disclosed warrant size.

Microsoft Azure, Oracle OCI, Google Cloud

Microsoft Azure, Oracle OCI, and Google Cloud are distribution partners that put AMD silicon inside public cloud capacity instead of only in direct enterprise sales. Azure uses AMD-based VM families such as HBv4, HX, Dpsv5, and Epsv5. Oracle OCI uses AMD-based E4 and E5 shapes. Google Cloud uses AMD EPYC-based Tau T2D and C3D families. These partnerships matter because cloud buyers can consume AMD infrastructure by the hour, which expands AMD's reach without Advanced Micro Devices, Inc. building a cloud business itself. They also create software validation, because developers optimize for the instance families they can actually rent.

• Azure uses AMD-based HBv4, HX, Dpsv5, and Epsv5 families.
• Oracle OCI uses AMD-based E4 and E5 shapes.
• Google Cloud uses AMD-based Tau T2D and C3D families.

Dell, HP, Lenovo, Tencent Cloud

Dell, HP, Lenovo, and Tencent Cloud extend Advanced Micro Devices, Inc. from chip supply into finished systems and regional cloud capacity. Dell's AMD-based PowerEdge line includes R7615 and R7625. HP's AMD-based enterprise server line includes ProLiant DL385 Gen11. Lenovo's AMD-based server line includes ThinkSystem SR645 V3 and ThinkSystem SR665 V3. Tencent Cloud adds cloud-side demand in China through AMD EPYC-based instances. These partners matter because they sit between Advanced Micro Devices, Inc. and end customers, which lowers direct sales friction and helps AMD compete in enterprise procurement where default choices often favor familiar server platforms.

• Dell PowerEdge R7615 and R7625 place AMD in server catalogs.
• HP ProLiant DL385 Gen11 keeps AMD visible in enterprise bids.
• Lenovo ThinkSystem SR645 V3 and SR665 V3 widen channel reach.
• Tencent Cloud adds AMD exposure in China and Asia cloud demand.

ASE and Amkor OSAT partners

ASE and Amkor are OSAT partners, which means outsourced semiconductor assembly and test. Their role is post-wafer: cut, assemble, package, and test chips before they go into servers or cloud deployments. For Advanced Micro Devices, Inc., that matters because advanced packages built around 2.5D and 3D integration can become a bottleneck even when wafers are available. OSAT partners add capacity, improve throughput, and reduce the risk that one part of the chain stops shipping. In a chiplet design model, back-end partners are as strategic as the foundry.

• Packaging capacity affects how quickly finished chips reach OEMs and cloud buyers.
• Test capacity affects yield and launch reliability.
• Multiple OSAT partners reduce single-point dependency.

Advanced Micro Devices, Inc. - Canvas Business Model: Key Activities

Advanced Micro Devices, Inc. concentrates this block on chip design, software enablement, and product cadence across 96-core server CPUs, 16-core desktop CPUs, and AI accelerators with 192 GB to 288 GB of high-bandwidth memory.

Design CPUs, GPUs, AI accelerators. The server and AI hardware activity is built around chiplet-based CPUs and high-memory accelerators. EPYC 9004 uses up to 96 cores, 12 DDR5 memory channels, and 128 PCIe 5.0 lanes. Instinct MI300X pairs 192 GB of HBM3 with 5.3 TB/s of bandwidth. Instinct MI325X raises that to 288 GB of HBM3E and 6.0 TB/s of bandwidth. Ryzen 9 9950X uses 16 cores, 32 threads, 80 MB cache, and boost clocks up to 5.7 GHz.

• 96 cores on EPYC 9004
• 12 DDR5 channels on EPYC 9004
• 128 PCIe 5.0 lanes on EPYC 9004
• 192 GB HBM3 on Instinct MI300X
• 288 GB HBM3E on Instinct MI325X
• 5.3 TB/s bandwidth on Instinct MI300X
• 6.0 TB/s bandwidth on Instinct MI325X
• 16 cores and 32 threads on Ryzen 9 9950X
• 80 MB cache on Ryzen 9 9950X
• up to 5.7 GHz boost on Ryzen 9 9950X

Release annual AI product cadence. AMD moved from MI300X in 2023 to MI325X in 2024, with the MI350 series announced for 2025. That creates about 12-month steps in the AI accelerator line. The jump from 192 GB to 288 GB of high-bandwidth memory is the most visible hardware step in the cycle.

Develop ROCm software stack. ROCm is the software layer that lets AMD hardware run AI and HPC workloads. The stack matters because chips alone do not win training or inference deals if the software cannot run common frameworks. ROCm support centers on Linux and frameworks such as PyTorch, TensorFlow, and JAX. In practical terms, this activity turns GPU and accelerator silicon into a usable platform for model training, model fine-tuning, and inference deployment.

• ROCm 6
• Linux
• PyTorch
• TensorFlow
• JAX

Expand developer ecosystem and cloud access. AMD's ecosystem work shows up in cloud instances and developer access to Instinct hardware without direct chip purchases. Named cloud channels include Microsoft Azure and Oracle Cloud Infrastructure. This matters because cloud availability lowers the barrier for testing, porting, and scaling AI workloads on AMD hardware. It also creates reference environments that software teams can standardize around when they build and deploy models.

• Microsoft Azure
• Oracle Cloud Infrastructure
• AMD Instinct

Optimize server, PC, and embedded roadmaps. The roadmap is split across data center, client, AI PC, and embedded markets. EPYC covers server workloads with up to 96 cores and 128 PCIe 5.0 lanes. Ryzen 9 9950X covers desktop with 16 cores, 32 threads, 80 MB cache, and boost clocks up to 5.7 GHz. Ryzen AI 300 series targets AI PCs with up to 50 TOPS, above the 40 TOPS Copilot+ PC threshold. Embedded families extend the same architecture into industrial, networking, and edge systems with longer product lifecycles.

• 96 cores for server CPUs
• 16 cores for desktop CPUs
• 50 TOPS for AI PC-class processing
• 40 TOPS Copilot+ PC threshold
• 128 PCIe 5.0 lanes for server expansion

Advanced Micro Devices, Inc. - Canvas Business Model: Key Resources

Advanced Micro Devices, Inc. relies on 16% Zen 5 IPC uplift, EPYC CPUs with up to 192 cores, Instinct MI300X with 192 GB of HBM3, Instinct MI325X with 288 GB of HBM3e, and TSMC 3 nm and 2 nm capacity.

Zen, EPYC, Ryzen, Instinct IP

• Zen 5: 16% average IPC uplift over Zen 4.
• EPYC Turin: up to 192 cores.
• Ryzen 9 9950X: 16 cores, 32 threads, up to 5.7 GHz boost.
• Ryzen AI 300 series: up to 50 TOPS.
• Instinct MI300X: 192 GB HBM3 and 5.3 TB/s bandwidth.
• Instinct MI325X: 288 GB HBM3e and 6.0 TB/s bandwidth.

ROCm software stack and developer cloud

ROCm 6.x is the GPU software stack, and developer cloud access ties software work to Instinct hardware access.

Cash and free cash flow

• 2023 revenue: $22.68 billion
• Q1 2024 revenue: $5.47 billion
• Q1 2024 gross margin: 47%
• Q1 2024 data center revenue: $2.3 billion
• Q1 2024 client revenue: $1.4 billion
• Q1 2024 gaming revenue: $0.9 billion
• Q1 2024 embedded revenue: $0.8 billion

Semiconductor engineers and AI talent

Advanced Micro Devices, Inc. had approximately 26,000 employees worldwide at the end of 2023.

TSMC 2nm/3nm manufacturing access

Access to 3 nm and 2 nm capacity supports higher transistor density, lower power, and denser chiplet integration.

Advanced Micro Devices, Inc. - Canvas Business Model: Value Propositions

Advanced Micro Devices, Inc. sells high-performance CPUs, GPUs, and adaptive silicon across data center, client, gaming, and embedded markets, and in 2024 it reported $25.785B in revenue, 49% gross margin, and $1.9B in operating income.

High-performance CPU and GPU computing is the core value proposition. In servers, the company's 192-core EPYC CPUs and high-bandwidth accelerators are built to increase the amount of work done per socket, per rack, and per deployment. That matters because data center buyers pay for compute, memory, power, and space together, not in isolation. The MI300X's 192GB of HBM3 and 5.3TB/s of bandwidth, plus the MI325X's 256GB and 6.0TB/s, show that the company is selling capacity as much as raw compute.

Better AI inference cost efficiency is a separate value proposition from training. Inference is the phase where a trained model answers requests, and it runs repeatedly after training is done. Bigger memory pools and higher bandwidth help keep more of the model on one accelerator instead of splitting it across multiple devices. The MI300X and MI325X numbers matter here: 192GB, 256GB, 5.3TB/s, and 6.0TB/s are the hardware facts that support the cost argument.

Open software and ecosystem flexibility is part of the selling point because customers want hardware that fits existing tools. Advanced Micro Devices, Inc. pushes ROCm as its software stack and relies on the x86 ecosystem that many enterprises already run. That lowers switching friction for buyers that want to move workloads without rebuilding every application layer. For academic work, this is important because it explains why the company can compete even when it is not the only vendor with high-end silicon.

Specialized silicon for AI, HPC, and sovereign workloads matters because not every buyer wants a generic cloud setup. Sovereign workloads often require local control, on-premises deployment, and tighter data handling. Advanced Micro Devices, Inc. expanded into adaptive compute with the $49B Xilinx acquisition and added AI systems capability through the $4.9B ZT Systems deal. Those numbers show that the value proposition goes beyond chips alone and into platform and systems delivery.

• Data center buyers want 192-core CPUs and accelerators with 192GB to 256GB of memory.
• AI buyers want inference density tied to 5.3TB/s and 6.0TB/s memory bandwidth.
• Software teams want ROCm and x86 compatibility instead of a closed stack.
• Government and regulated buyers want controlled deployments and system-level design options.
• OEM, console, and embedded customers want a supplier that spans four revenue segments.

Broad portfolio across data center to embedded is one of the clearest canvas-level value propositions. In 2024, Advanced Micro Devices, Inc. generated $12.577B from Data Center, $7.055B from Client, $2.603B from Gaming, and $3.549B from Embedded. That means Data Center represented 48.8% of total revenue, Client 27.4%, Gaming 10.1%, and Embedded 13.8%. The mix shows that the company sells compute from enterprise servers down to edge and embedded systems.

Client-side compute also supports the value proposition across notebooks and desktops. Advanced Micro Devices, Inc. has client processors with up to 50 AI TOPS, which gives it a way to sell local AI features into consumer and commercial PCs. That matters because the same company can sell into servers and personal computers, which broadens demand sources and reduces reliance on one market cycle.

• 192 cores in 5th Gen EPYC
• 192GB HBM3 in MI300X
• 256GB HBM3E in MI325X
• 5.3TB/s bandwidth in MI300X
• 6.0TB/s bandwidth in MI325X
• $49B Xilinx acquisition
• $4.9B ZT Systems acquisition
• $25.785B 2024 revenue
• 49% gross margin in 2024

Advanced Micro Devices, Inc. - Canvas Business Model: Customer Relationships

AMD's customer relationships are built on multi-year engineering, deployment, and support cycles. The clearest public proof is Frontier, which uses 9,408 compute nodes and 37,632 AMD Instinct MI250X GPUs.

Long-term enterprise and cloud co-development. AMD's strongest customer ties sit in systems that are built and validated over time. EPYC 9004 processors scale to 96 cores with 12 DDR5 memory channels and 128 PCIe 5.0 lanes, while MI300X carries 192GB of HBM3. Those numbers matter because cloud and enterprise customers buy for memory capacity, parallelism, and I/O headroom, not just chip count. Frontier shows the same model at scale: 9,408 nodes and 37,632 GPUs make the relationship a system program, not a single sale.

Direct strategic account management. AMD reported $22.68 billion in revenue in 2023 across 4 segments: Data Center, Client, Gaming, and Embedded. That structure means the customer relationship is not one-size-fits-all. Large accounts need platform reviews, technical validation, supply visibility, and road-map alignment because one deployment can affect orders across more than one segment. The business model depends on account-level attention where the buyer is deciding on processor generations, accelerator fit, and deployment timing.

Developer community engagement through ROCm. ROCm is the software layer that keeps developers tied to AMD hardware. The practical hook is hardware scale: EPYC 9004 reaches 96 cores, and MI300X offers 192GB of HBM3, which shapes how developers fit models, allocate memory, and tune performance. For AI and HPC users, these numbers are part of the software conversation because code has to match the hardware envelope before production deployment.

OEM and partner-led support. AMD uses partner channels through Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro. That matters because enterprise buyers usually want the system builder and the chip vendor aligned on qualification and service. The repeated numeric targets are the same across those channels: 96 cores on EPYC 9004, 12 memory channels, 128 PCIe 5.0 lanes, and 192GB on MI300X. Those specs make partner support more standardized across different server and accelerator configurations.

Multi-year supply and deployment programs. Frontier began deployment in 2022, and its scale of 9,408 nodes and 37,632 GPUs shows why the customer relationship stays active after the initial sale. The ongoing work includes software tuning, hardware support, and capacity planning. That is the same logic behind AMD's broader enterprise and cloud relationships: the customer does not just buy silicon, it commits to a platform that has to keep running for years.

• 9,408 Frontier compute nodes
• 37,632 AMD Instinct MI250X GPUs
• 96 EPYC 9004 cores
• 12 DDR5 memory channels
• 128 PCIe 5.0 lanes
• 192GB HBM3 on MI300X
• $22.68 billion AMD revenue in 2023
• 4 reporting segments

Advanced Micro Devices, Inc. - Canvas Business Model: Channels

AMD reported $22.7 billion in revenue in 2023, with $6.5 billion from Data Center, $4.4 billion from Client, $6.2 billion from Gaming, and $5.6 billion from Embedded. The channel mix spans 3 OEM names, 4 named cloud platforms, and 1 developer cloud.

Direct sales to hyperscalers

$6.5 billion Data Center revenue in 2023.

1 channel route for EPYC CPUs and Instinct accelerators into hyperscale buyers.

OEM channels via Dell, HP, Lenovo

3 OEM names: Dell, HP, Lenovo.

$4.4 billion Client revenue in 2023.

Cloud deployments through Azure, OCI, Google Cloud, Tencent Cloud

4 named cloud platforms.

Azure, OCI, Google Cloud, and Tencent Cloud.

AMD Developer Cloud

2024 launch year.

AMD Instinct MI300X access.

Consumer retail and system integrators

$6.2 billion Gaming revenue in 2023.

5.6 billion Embedded revenue in 2023.

• 22.7 billion total revenue in 2023
• 6.5 billion Data Center revenue in 2023
• 4.4 billion Client revenue in 2023
• 6.2 billion Gaming revenue in 2023
• 5.6 billion Embedded revenue in 2023
• 3 OEM names: Dell, HP, Lenovo
• 4 cloud platforms: Azure, OCI, Google Cloud, Tencent Cloud
• 1 AMD Developer Cloud

Advanced Micro Devices, Inc. - Canvas Business Model: Customer Segments

AMD's 2023 customer base mapped to $22.682B of revenue across $6.495B Data Center, $4.576B Client, $6.242B Gaming, and $5.369B Embedded.

That mix points to five core buyer groups: hyperscale cloud providers, enterprise AI and HPC customers, PC OEMs and commercial buyers, gaming and console partners, and industrial, automotive, and embedded customers.

Hyperscale cloud providers sit inside the $6.495B Data Center segment and represented 28.6% of AMD's 2023 revenue. The customer logic is simple: a small number of very large cloud operators buy high-volume server CPUs and accelerators, and a single public proof point for this demand is Frontier at 1.102 exaflops. The size of the segment matters because cloud buying ties to multi-rack deployments, not one-off unit sales.

Enterprise AI and HPC customers also sit inside the $6.495B Data Center segment, so the same revenue pool serves both cloud and non-cloud compute demand. The key numeric signal is 1.102 exaflops, which shows AMD silicon in exascale-class computing. For academic work, this segment is best treated as a mix of training, inference, simulation, and scientific computing demand, all tied to high-value system purchases rather than low-margin commodity volume.

PC OEMs and commercial buyers generated $4.576B of 2023 revenue in the Client segment, or 20.2% of AMD's total. This segment covers notebook and desktop OEM orders plus commercial PC demand. The number matters because PC demand is more cyclical than Data Center, so a $4.576B revenue base gives AMD a large but more volatile buyer group than cloud and embedded.

Gaming and console partners produced $6.242B of 2023 revenue, equal to 27.5% of total revenue. That makes Gaming one of AMD's largest customer pools. The segment is anchored by 2 major console ecosystems and by discrete graphics demand. The size of the segment matters because semi-custom console programs and graphics demand can move materially with hardware cycles and unit refresh timing.

Industrial, automotive, and embedded customers generated $5.369B in 2023, or 23.7% of AMD's total revenue. This segment is important because it combines industrial, automotive, networking, and communications demand in products that usually have longer lifecycle expectations than PC parts. The $5.369B figure shows that embedded is not a niche line for AMD; it is nearly as large as Client and close to Gaming.

• $6.495B Data Center revenue served both hyperscale cloud and enterprise AI/HPC buyers.
• $4.576B Client revenue served PC OEMs and commercial buyers.
• $6.242B Gaming revenue served console partners and graphics buyers.
• $5.369B Embedded revenue served industrial and automotive demand.
• $22.682B total 2023 revenue spread across 4 reportable segments.

Advanced Micro Devices, Inc. - Canvas Business Model: Cost Structure

Advanced Micro Devices, Inc. had a cost structure anchored by $6.456 billion of research and development in 2024, equal to 25.0% of $25.785 billion of revenue. Outsourced wafer, packaging, and memory costs sit inside cost of revenue rather than in owned-fab depreciation.

$6.456 billion of R&D is the clearest fixed-cost line. It covers CPU, GPU, adaptive computing, and software development, and it is large enough to explain why AMD needs scale in revenue to keep operating leverage positive.

TSMC wafer and advanced packaging charges are direct costs in AMD's model, but AMD does not publish a separate dollar amount for wafer starts, die packaging, or advanced packaging capacity. That means the financial effect shows up inside cost of revenue and helps explain why a 49% gross margin still leaves a large amount of revenue needed to cover non-manufacturing costs.

HBM and DRAM input costs matter most in AI accelerators. MI300X uses 192 GB of HBM3, and MI325X uses 256 GB of HBM3E. Higher memory content raises bill-of-materials cost and makes AI accelerators more expensive than client CPUs or lower-memory parts.

Sales, marketing, and partner support sit in SG&A, which means selling, general, and administrative expenses. AMD does not disclose separate public amounts for channel rebates, distributor incentives, hyperscaler support, or field engineering, so these costs remain embedded in the broader SG&A total.

Supply-chain and compliance costs are spread across procurement, logistics, quality, trade, legal, export control, and regulatory work. AMD's reliance on third-party manufacturing and memory suppliers means any constraint in wafer capacity, advanced packaging, or HBM supply affects both cost and delivery timing.

Advanced Micro Devices, Inc. - Canvas Business Model: Revenue Streams

Advanced Micro Devices, Inc. reported $25.8 billion of revenue in 2024 across 4 operating segments. Data Center was $12.6 billion, Client was $7.1 billion, Embedded was $3.6 billion, and Gaming was $2.6 billion.

Data Center CPUs and AI accelerators recorded $12.6 billion in 2024. That was $1.9 billion more than Client plus Embedded combined at $10.7 billion, and $10.0 billion more than Gaming at $2.6 billion.

• $12.6 billion Data Center revenue
• $10.7 billion Client plus Embedded combined
• $10.0 billion Data Center above Gaming

Client Ryzen processors generated $7.1 billion in 2024. That is $5.5 billion below Data Center and $4.5 billion above Gaming, with Client remaining above $7.0 billion inside the $25.8 billion total.

• $7.1 billion Client revenue
• $5.5 billion below Data Center
• $4.5 billion above Gaming

Gaming Radeon and semi-custom chips contributed $2.6 billion in 2024. That was the smallest of the 4 revenue streams, but it still added more than $2.0 billion to total revenue.

• $2.6 billion Gaming revenue
• 4 operating segments
• $2.0 billion plus contribution to total revenue

Embedded processors and SoCs produced $3.6 billion in 2024. That was $1.0 billion more than Gaming and $1.5 billion less than Client.

• $3.6 billion Embedded revenue
• $1.0 billion above Gaming
• $1.5 billion below Client

Licensing and strategic deployment-related sales were not disclosed as a separate revenue line in the 4-segment reporting. The separately disclosed amount is $0, while total reported revenue still equals $25.8 billion.

• 4 disclosed operating segments
• $0 separately disclosed licensing revenue
• $25.8 billion total reported revenue