AMD Pushes Supply Chain Partners to Expand AI Chip Capacity

Reuters first reported that AMD asked suppliers and manufacturing partners to accelerate AI chip production as demand for CPUs and AI infrastructure continued to rise globally.

AMD CEO Lisa Su said during a Taipei media briefing that demand for AI-related processors had exceeded the company’s expectations, particularly as enterprises expanded inferencing and agentic AI workloads. Reuters reported that AMD expects supply capacity to increase every quarter through 2026, with larger expansion continuing into 2027 and beyond.

The comments add to growing signs that AI infrastructure demand is beginning to strain semiconductor manufacturing capacity beyond GPUs. Chip packaging, substrates, and advanced-node production have become increasingly constrained as cloud providers and enterprises scale AI systems.

AMD also said it plans to invest more than $10 billion across Taiwan’s AI ecosystem, including manufacturing and packaging infrastructure tied to AI chips and servers.

The company’s manufacturing ecosystem includes Taiwan Semiconductor Manufacturing Company (TSMC), ASE Technology, Siliconware Precision Industries (SPIL), Powertech Technology (PTI), and Amkor Technology.

AI infrastructure spending has increasingly spread beyond hyperscalers and model developers into semiconductor manufacturing systems, with equipment suppliers also reporting accelerated demand.

Packaging Capacity Emerges as a Bottleneck

The pressure on semiconductor packaging has become a central issue across the AI hardware industry.

Modern AI processors require advanced packaging technologies such as chiplet integration and high-bandwidth memory assembly. Those systems rely heavily on specialized packaging capacity that remains concentrated among a small group of suppliers.

Earlier this week, Reuters reported that Amkor Technology was working with AMD on advanced packaging after acquiring additional land in Arizona for a semiconductor packaging campus expected to begin production in 2028.

NVIDIA CEO Jensen Huang previously said advanced packaging demand around Blackwell and Hopper systems remained constrained despite TSMC increasing capacity.

Several semiconductor suppliers have also warned that AI infrastructure demand is beginning to stress packaging and manufacturing systems across the broader supply chain.

Broadcom said earlier this year that supply-chain bottlenecks tied to AI infrastructure had expanded beyond wafers into printed circuit boards, lasers, and manufacturing components.

AI Infrastructure Demand Expands Beyond GPUs

AMD’s comments also reflect a broader shift in AI infrastructure demand patterns.

The first phase of the generative AI boom centered heavily on GPU deployments for training large language models. Companies are now scaling inferencing infrastructure and agentic AI systems that require larger mixes of CPUs, GPUs, networking hardware, and memory systems.

AMD recently began production ramp-up for its next-generation EPYC Venice processors on TSMC’s 2-nanometer process technology.

At the same time, semiconductor suppliers are expanding manufacturing capacity across the U.S. and Taiwan as AI infrastructure demand increases pressure on packaging materials, substrates, and optical interconnect technologies.

TSMC said earlier this month that it expects the semiconductor market to reach $1.5 trillion by 2030, with AI and high-performance computing accounting for more than half of industry demand.