AI Demand Sends the Foundry Market Up 29%—and Keeps TSMC Far Ahead

AI is reshaping semiconductor manufacturing from the wafer fab outward. In Q2 2026, the worldwide pure-play foundry market grew 29% year over year as AI-related orders surged and manufacturers redirected capacity toward those programs. TSMC captured 73% of the market for the second consecutive quarter, while Samsung Foundry remained in second place at roughly 7%. 4 14

The central lesson for electronics manufacturers is that securing a processor is no longer enough. AI systems depend on a coordinated chain of leading-edge wafers, memory, substrates, advanced packaging and testing—so a shortage at any one stage can delay the entire product.

Why AI demand produced such strong foundry growth

AI accelerators are concentrated in advanced semiconductor manufacturing, where capacity is expensive and difficult to expand quickly. The same AI buildout also increases demand for supporting chips and memory, creating pressure beyond the most advanced logic nodes.

Counterpoint Research identified two principal forces behind Q2’s growth: a sharp increase in AI-related orders and the reallocation of production capacity. Together, they created supply-and-demand imbalances across both advanced and mature processes. 4

That matters because AI infrastructure uses more than the main compute die. Server systems also require connectivity, control, power-management and other components that may be produced on different process technologies. When foundries prioritize AI-related production, capacity can become constrained across multiple parts of the manufacturing portfolio rather than at only one leading-edge node. 4

How TSMC maintained a 73% share

TSMC’s lead reflects a combination of process availability and capacity breadth:

• 2nm mass production: TSMC’s 2nm manufacturing entered volume production, giving customers access to its newest process technology. 3 4
• 3nm expansion: The company continued ramping 3nm production as demand for advanced AI and high-performance computing chips remained strong. 3 4
• Tight mature-node supply: Capacity pressure was not limited to leading-edge production; shortages also affected mature 8-inch and 12-inch processes. 4
• Advanced-packaging constraints: TSMC’s position was reinforced by tight supply in advanced packaging, an increasingly important part of delivering complex AI systems. 4

The result was not simply a larger pool of wafer capacity. TSMC was positioned across several connected stages that AI-chip customers need, helping it preserve 73% share in both Q1 and Q2 2026. 4

Why Samsung stayed No. 2

Samsung Foundry remained the second-largest pure-play foundry in Q2, with its share broadly unchanged from the previous quarter. Counterpoint pointed to improving SF2 yields, stronger demand for SF4 and SF5, and wafer-price increases during the first half of the year as contributors to its performance. 14

Samsung also raised prices on some advanced contract-chip orders by as much as 15% in July, according to Reuters reporting. The increases varied by process and customer location; reported SF4 increases were highest for some customers in China and the United States. 1

The timing is important. Because those increases began in July, they occurred after the second quarter had closed. They therefore help explain pricing and customer-cost pressure in the second half of 2026, but should not be treated as a direct cause of the market’s Q2 growth. 1

Samsung’s challenge is consequently two-sided: it is benefiting from demand for established advanced nodes such as SF4 and SF5 while working to improve the manufacturing economics and yields of SF2. The company’s ability to convert that demand into sustained share gains will depend on execution as well as pricing.

Why advanced packaging has become a bottleneck

A finished AI accelerator requires more than a fabricated logic wafer. High-performance systems combine compute with high-bandwidth memory and complex interconnects, then move those components through specialized assembly, testing and thermal-management processes.

This makes advanced packaging a separate capacity constraint. It requires dedicated equipment, materials, process integration and yield management, and it cannot necessarily be expanded as quickly as wafer starts. Counterpoint specifically cited tight advanced-packaging supply alongside mature-node constraints as a factor in the foundry market. 4

The bottleneck has two consequences:

1. Finished wafers can wait for packaging. More front-end capacity does not automatically produce more shippable accelerators if back-end capacity is limited.
2. Customers must reserve a complete path. A procurement plan that covers only the foundry step can still fail if HBM, substrates, packaging or testing are unavailable.

The industry’s investment response shows how strategic packaging has become. SK hynix broke ground on its first U.S. HBM packaging base in West Lafayette, Indiana. The project represents an investment of more than $4 billion, with next-generation HBM mass production targeted for the second half of 2029.

Powertech Technology is also expanding fan-out panel-level packaging. Its capacity for the technology was reported as fully booked through 2030, with AMD and Broadcom identified as customers, and the company approved a $400 million Singapore joint venture with Broadcom focused on advanced panel-level assembly technology.

The available reporting supports those commitments and bookings, but it does not independently establish the $2.2 billion Powertech investment figure sometimes associated with the project. That figure should therefore not be used as a confirmed number.

Memory investment is expanding alongside foundry capacity

AI demand is also driving large investments in memory manufacturing. Kioxia and Sandisk plan to invest more than $31 billion in Japan through 2032 to expand semiconductor technology and production capacity, subject to government support. The program includes infrastructure connected with the companies’ Yokkaichi and Kitakami operations.

These projects address a different part of the AI supply chain from TSMC’s logic manufacturing, but the strategic connection is direct: AI systems need both advanced compute chips and large quantities of high-performance memory and storage. Expanding only one side of that equation does not remove the supply constraint.

What electronics manufacturers should do now

The Q2 market data points to a practical change in sourcing strategy. Manufacturers should treat AI-chip procurement as an integrated capacity-planning exercise rather than a series of independent component purchases.

Reserve capacity earlier

Longer lead times and high utilization make late spot purchasing increasingly risky. Companies with credible demand forecasts should discuss wafer, memory and packaging reservations earlier, using ranges rather than relying on a single overly precise forecast.

Track packaging and yield as supply metrics

Traditional semiconductor dashboards often emphasize wafer allocation and process-node availability. For AI products, procurement teams should also monitor advanced-packaging slots, HBM availability, substrate supply, test capacity and packaging yields.

Qualify alternatives where technically possible

A second foundry or packaging provider may not be a drop-in replacement, particularly for advanced AI designs. Still, early qualification of alternate suppliers can reduce dependence on one manufacturing route and expose integration problems before a production crisis.

Plan the entire manufacturing chain

A resilient plan should connect:

• logic wafer fabrication;
• HBM or other required memory;
• substrates, interposers and interconnects;
• advanced assembly and packaging;
• testing, thermal solutions and logistics.

The broader outlook

The 29% Q2 expansion shows how quickly AI demand is flowing through the semiconductor industry, but the market-share figures also show that growth is highly concentrated. TSMC’s 73% position underscores the value of having leading-edge process technology, mature-node capacity and advanced packaging available within a coordinated ecosystem. 4

Samsung remains the clear No. 2, supported by SF4 and SF5 demand and efforts to improve SF2 yields, but its price increases also show how tight capacity is becoming. 1 14

For buyers, the defining constraint is no longer just whether a chip can be designed or a wafer can be fabricated. It is whether every required manufacturing stage can be secured at the same time.