TSMC to further prioritize AI orders amid a 3nm chip shortage by the end of the third quarter of 2026

TSMC is preparing for a new round of 3nm production shortages: by the end of the third quarter of 2026, available capacity could become even tighter. The main reason is a sharp increase in orders for AI chips and the priority the company gives to its most important customers.

Why Shortage Is Growing

Demand for TSMC's advanced 3nm process technology has long outpaced supply, but now the imbalance is shifting even more toward cloud and AI orders. Major cloud providers and accelerator developers need increasingly energy-efficient and high-performance chips for servers, networking equipment, and related infrastructure. As a result, a significant portion of the most valuable production slots is now going to projects directly tied to the rapid growth of the AI market.

In parallel, customers are growing anxious about the fact that the expansion of advanced capacity is not proceeding as quickly as desired. TSMC's US facilities are not yet able to fully relieve the load on Taiwan's fabs: only one plant has entered full operation, while others are still in the design or construction phase. Therefore, the center of tension remains in Taiwan, where the number of customers wanting 3nm output is noticeably greater than the company can serve within reasonable timeframes.

Who Gets Priority

In conditions of capacity shortage, TSMC distributes access unevenly. The company primarily protects segments that bring it strategic revenue and guarantee sustained factory utilization. This means that the winners are not just large customers, but those who are either closely tied to AI growth or have long secured volumes through long-term agreements. For other customers, the window of opportunity narrows: they find it harder to obtain needed volumes, and their negotiating position on price and timelines becomes weaker.

• Developers of AI components like AMD, Nvidia, Broadcom, and Marvell
• Customers with long-term contracts, including Apple and MediaTek
• Orders for server and cloud infrastructure
• Projects where stable volumes and high margins are important

What Customers Are Doing

Companies that are not among the priority customers are already considering workaround scenarios. One obvious option is to hand off some orders to Samsung, if the Korean contract manufacturer can offer acceptable timelines, price, and yield quality on its new lines. But this is more of a backup than a full replacement: transferring complex designs to a different process technology and different fab requires time, money, and additional engineering resources.

More radical ideas are also emerging in the market. Some players are thinking not only about diversifying suppliers, but also about partial control over their own manufacturing. Against this backdrop, Elon Musk's plans to create a joint venture between SpaceX and Tesla in Texas to produce 2nm chips for these companies' needs are being discussed.

Even if such projects move slowly, the trend itself is telling: the shortage of advanced capacity is pushing major customers toward tighter vertical integration. Another important detail is that 3nm process technology is needed not just for AI accelerators. It is in demand for networking components and edge computing segments, where energy efficiency and transistor density are also important.

But with limited capacity, AI infrastructure gets the advantage. For other segments, this means rising prices, risk of delivery delays, and more complex planning for product launches in the coming quarters.

What This Means

The chip market is entering a phase where the main shortage is formed not around all advanced nodes at once, but around a specific 3nm window critical for AI infrastructure. For electronics manufacturers and chip developers, this is bad news: if a company lacks strategic status with TSMC, it will have to either overpay, shift timelines, or build an alternative supply chain in advance. According to market participants' assessments, this problem will not disappear quickly and could extend for another one to two years.