CompanyGraph
Advanced Micro Devices, Inc.
AMD · United States
Designs x86 CPU and RDNA GPU architectures on a single die, translating the 2006 ATI acquisition into integrated APUs manufactured entirely through TSMC's advanced nodes.
Because all silicon — Ryzen, EPYC, and console APUs — is produced exclusively on TSMC's 7nm and 5nm nodes, the company's output in any period is capped by TSMC's wafer allocation decisions rather than by design readiness or customer demand, and no internal capital deployment can expand that ceiling. The same fabless structure that removes manufacturing cost also removes manufacturing control, so cryptocurrency-driven GPU demand swings and U.S. export restrictions on Chinese server and AI markets create demand volatility that cannot be absorbed by adjusting production. Binding x86 and RDNA development onto a single die means any schedule slip in either architecture propagates as a delay to the full APU, and console customers — Sony and Microsoft — absorb that combined latency at the one moment a competing supplier can credibly propose an alternative: the console generation switch. On the server side, the x86-64 instruction-set dependency cuts in the opposite direction, because switching from EPYC to Intel Xeon forces Linux kernel optimisation and application recompilation work on the customer, creating a technical migration barrier that protects existing design-wins even as wafer allocation constrains how many new ones can be supplied.
How does this company make money?
Chips are sold per unit to OEMs (original equipment manufacturers such as PC and server builders), system integrators, and distributors, with price tiers set by core count, clock speeds, and performance bins — the sorted grades assigned to chips after testing. Console APUs for PlayStation and Xbox generate a separate semi-custom revenue stream, where the design is built to specification for each platform holder and units are sold as those consoles are manufactured.
What makes this company hard to replace?
Sony PlayStation and Microsoft Xbox custom APU designs require 3–4 year development cycles with extensive software optimisation, which means the console generation switch is the primary — and narrow — window in which a competing supplier could be considered. On the server side, x86-64 customers running EPYC processors face Linux kernel optimisation work and application recompilation requirements when switching to Intel Xeon, creating a technical barrier to migration that is specific to the instruction-set transition.
What limits this company?
TSMC's advanced-node wafer starts — the physical initiation of each batch of chips at 7nm or 5nm — are rationed across all fabless customers at the same time, so the number of Ryzen and EPYC processors that can be manufactured in any given period is capped by allocation priority, not by how ready the designs are or how much demand exists. The company cannot raise this ceiling through its own capital investment because the constraint lives inside TSMC's fabrication facility, not inside the company's balance sheet.
What does this company depend on?
The mechanism depends on five named upstream inputs: TSMC foundry capacity at 7nm and 5nm nodes; the x86-64 instruction set licence from Intel; PCIe and DDR memory controller IP (the interface standards that allow the chip to communicate with other components in a system); Windows and Linux driver certification (software layers that allow operating systems to recognise and use the hardware); and the Sony PlayStation and Microsoft Xbox design wins that commit the APU to specific console platforms.
Who depends on this company?
Sony PlayStation 5 consoles depend on the custom APU directly — losing it would cause gaming performance degradation across the platform. Microsoft Xbox Series X and Series S would lose their custom silicon, forcing a platform redesign. Dell and HP server divisions rely on EPYC processors for their data centre product lines, and losing access would reduce their competitiveness against Intel Xeon-based configurations.
How does this company scale?
Once a chip design is validated, it replicates at effectively zero additional cost per copy, allowing the same Ryzen or RDNA architecture to serve multiple product SKUs across different market segments. The bottleneck that does not scale with demand is TSMC wafer allocation: advanced-node capacity is rationed among all fabless customers based on long-term commitments and strategic relationships, and the company cannot purchase its way to a larger share.
What external forces can significantly affect this company?
U.S. export controls restrict advanced chip sales to Chinese customers, including server and AI markets, limiting addressable demand in that geography. TSMC's location in Taiwan introduces geopolitical risk — any disruption to Taiwan's status as a stable manufacturing base would affect foundry access directly. Cryptocurrency mining demand causes unpredictable swings in GPU unit volumes, creating demand patterns that are difficult to plan production around.
Where is this company structurally vulnerable?
Binding CPU and GPU development onto the same die means a schedule slip in either the x86 or RDNA architecture propagates as a delay to the entire APU. Console customers — Sony PlayStation and Microsoft Xbox — have platform launch dates fixed by production and retail commitments, so they absorb the combined latency of both component readiness problems at the same time. This exposes the design-win relationship at the only moment a competitor can credibly propose an alternative: the console generation switch.