TSMC to prioritize AI orders as 3nm shortage deepens by end of Q3 2026

TSMC is preparing for a new round of 3nm shortages: by the end of the third quarter of 2026, free capacity may shrink further. The main cause is a sharp rise in orders for AI chips and the priority the company gives its most important customers.

Why the shortage is growing

Demand for TSMC's leading-edge 3nm process has outpaced supply for a long time, but the imbalance is now tilting toward cloud and AI orders. Large providers and accelerator designers need ever more power-efficient and powerful chips for servers, networking equipment and related infrastructure. As a result, a large share of the most valuable production slots now goes to projects tied directly to the rapid growth of the AI market.

At the same time, customers are growing anxious that advanced capacity is expanding more slowly than they would like. TSMC's U.S. sites cannot yet fully relieve the Taiwanese fabs: only one facility is fully operational, while the others are still in design or under construction. The pressure therefore remains on Taiwan, where far more companies want 3nm output than TSMC can serve within a reasonable time.

Who gets priority

With capacity tight, TSMC allocates access unevenly. It first protects the segments that bring in strategic revenue and guarantee long-term fab utilization. That means the winners are not simply large customers, but those closely tied to AI growth or those that locked in volumes long ago through long-term agreements. For other customers, the window is narrowing: they find it harder to secure the volumes they need, and their negotiating position on price and timing is weaker.

• AI component designers such as AMD, Nvidia, Broadcom and Marvell
• Customers with long-term contracts, including Apple and MediaTek
• Orders for server and cloud infrastructure
• Projects that need stable volumes and high margins

What customers are doing

Companies that do not make the priority list are already weighing workarounds. One obvious option is to shift some orders to Samsung, if the Korean foundry can offer acceptable lead times, pricing and yield on its new lines. But that is more of a hedge than a full replacement: moving a complex design to another process and another fab takes time, money and extra engineering resources.

More radical ideas are also emerging. Some players are considering not only diversifying suppliers but also taking partial control of their own manufacturing. Against this backdrop, there is discussion of Elon Musk's plans for a SpaceX and Tesla joint venture in Texas to produce 2nm chips for those companies' needs. Even if such projects move slowly, the trend itself is telling: the shortage of advanced capacity is pushing large customers toward tighter vertical integration.

Another important detail is that 3nm is needed for more than AI accelerators. It is also in demand for networking components and edge computing, where power efficiency and transistor density matter too. But with limited capacity, AI infrastructure gets the advantage. For other segments, that means higher prices, a risk of shipment delays and harder planning of product launches over the coming quarters.

What this means

The chip market is entering a phase in which the main shortage forms not across all advanced nodes at once, but around a specific 3nm window that is critical for AI infrastructure. For electronics makers and chip designers, this is bad news: a company without strategic status at TSMC will have to pay a premium, push back its schedule, or build an alternative supply chain in advance. By market participants' estimates, the problem will not disappear quickly and could last another one to two years.