TSMC chip capacity is ‘fierce’ battleground for AI customers, amid record quarter

A semiconductor foundry is a manufacturing company that produces chips for customers that design them. Wafer capacity measures how many silicon wafers a factory network can process into chips over a period. It matters because every CPU, GPU, or AI accelerator needs manufacturing space before it can reach customers.

TSMC said it could make more than 17 million 12-inch-equivalent wafers last year. The figure combines production capacity using a standard wafer size, making different factories easier to compare. TSMC uses advanced processes to manufacture chips for Nvidia, AMD, Broadcom, and major technology companies.

Demand now exceeds available supply. TSMC is adding factories in Taiwan, Japan, and Arizona, while installing more equipment in existing plants. However, new facilities take years to open. Until then, wafer capacity remains a key bottleneck for AI companies trying to expand computing infrastructure.

TSMC operated capacity for more than 17 million 12-inch-equivalent wafers last year. That is an enormous manufacturing base, but it is not enough for the current AI expansion. The company says demand for its advanced chipmaking technologies remains exceptionally strong.

AI companies need processors from TSMC to build data centers and other computing systems. TSMC North American CEO Sajiv Dalal said he receives at least 10 calls each day about wafer capacity. TSMC also expects fourth-quarter revenue to rise 58% from a year earlier, driven by advanced-chip demand.

The article does not give a single total for all AI companies’ requested wafers. It does show the pressure clearly: customers compete fiercely for factory lines, while U.S. hyperscalers plan more than $700 billion in capital expenditures this year. Analysts expect that spending to increase in 2027.

Nvidia, AMD, and Broadcom rely on TSMC because their graphics processors and central processors are needed to run AI workloads. TSMC makes the vast majority of the world’s most advanced semiconductors, giving it an unusually central role in the AI supply chain.

The dependence is expanding beyond chip specialists. Apple, Google, Amazon, Microsoft, Meta, and Axiom also use TSMC to manufacture custom, in-house AI accelerators. These companies can design specialized processors, but they still need advanced factory lines to produce them at scale.

This concentration makes TSMC’s capacity especially valuable. Customers compete for access, negotiate years ahead, and sometimes create bespoke arrangements. If TSMC cannot add capacity fast enough, chipmakers and technology companies may struggle to match their AI infrastructure plans and aggressive revenue targets.

TSMC’s capacity shortage could become a limit on the entire AI industry. AI companies may have money and plans for new infrastructure, but those plans still require physical processors. Without enough advanced chips, construction and investment could move faster than the supply of hardware needed to operate them.

Analyst Jordan Klein said continued compute growth over the next 12 to 18 months could make TSMC constraints threaten massive infrastructure investments. He also warned about AI companies’ aggressive revenue growth targets. U.S. hyperscalers are already projecting more than $700 billion in capital expenditures this year.

The immediate risk is a mismatch between spending ambitions and chip availability. TSMC is adding factories and equipment, but new capacity takes years. If demand keeps accelerating, companies could face slower expansion, tighter supply, and greater competition for production slots before new fabs become available.

Large customers negotiate multiyear agreements because advanced logic chips require complex, expensive manufacturing. Production cannot be expanded instantly, especially when many companies want the same advanced processes. Securing capacity early helps companies protect future chip supply and plan their products.

Nvidia CEO Jensen Huang and AMD CEO Lisa Su have both visited Taiwan multiple times this year. D-Matrix CEO Sid Sheth said customers often reach Kevin Zhang, a TSMC senior vice president central to deciding which chipmakers get factory access. D-Matrix’s manufacturing vice president has visited TSMC since 1992.

Relationships can matter particularly when capacity is scarce. TSMC had 534 customers in 2025, but its top 10 produced 78% of net revenue. Startups therefore compete with much larger orders, while TSMC also has an incentive to support promising newcomers that could become major customers.

Samsung and Intel could provide alternatives because both are among the leading advanced chip manufacturers. A second source would reduce customers’ dependence on one supplier. It could also make the industry more resilient if TSMC’s factory lines remain full.

Dan Hutcheson of TechInsights said customers want another source to “twist TSMC’s arm” and keep pricing honest. Intel is the only U.S.-based advanced logic manufacturer, and Hutcheson said all the big chipmakers are now testing there. That testing suggests customers are exploring alternatives rather than relying on one path.

The change would not be immediate or guaranteed. Intel’s foundry business has struggled with delays and setbacks for years. Samsung and Intel would need to prove they can deliver the required advanced chips at scale. Still, capacity pressure gives both companies an opening to win customers from TSMC.

Logic chips are processors that perform calculations or control computing tasks. CPUs handle general processing, GPUs perform highly parallel work, and AI accelerators are specialized for machine-learning workloads. Their importance comes from the calculations needed to run software, data centers, and AI systems.

Manufacturing starts with silicon wafers, thin round substrates that hold many chip designs. A foundry repeatedly patterns extremely small circuit features onto the wafer, adds and removes materials, and creates the layered structures that form processors. The wafer is then separated into individual chips and packaged. The article highlights ASML’s extreme ultraviolet lithography machines as tools TSMC is adding to existing factories.

TSMC makes these advanced logic chips for Nvidia, AMD, Broadcom, and technology companies designing their own accelerators. Because production is complex and expensive, available wafer capacity is scarce. New fabs and additional tools should increase supply, but expansion will take years.