Jensen Huang goes to Taiwan to "secure production capacity": TSMC's N3 capacity increase is limited, and supply is expected to remain highly strained through 2026.

Jensen Huang goes to Taiwan to "secure production capacity": TSMC's N3 capacity increase is limited, and supply is expected to remain highly strained through 2026.

Chip giant Nvidia’s thirst for production capacity is forcing its CEO Jensen Huang to personally go to the doorstep of key partner TSMC to “ask for supplies,” but the latest analysis shows that even the world’s largest semiconductor foundry is already nearing its limits to meet market demand.

WallstreetCN reports that recently Jensen Huang appeared at TSMC’s annual sports day and publicly requested increased chip supply. TSMC CEO C. C. Wei also confirmed receiving the expansion request. Jensen Huang bluntly stated, “Without TSMC, there is no Nvidia.” The market generally believes the main purpose of Huang’s visit was to secure sufficient N3 capacity for Nvidia’s next generation of AI chips (including the Rubin series).

However, according to ChaseWind Trading Desk, JPMorgan pointed out in a research report published on November 10 that, even though Nvidia is reportedly asking TSMC to expand N3 capacity to 160,000 wafers per month (wfpm), TSMC’s actual capacity by the end of 2026 may only reach 140,000 to 145,000 wafers. This forecast indicates that a supply gap for the world’s most advanced chips will persist over the next two years.

This supply-demand imbalance points to a dual impact. On one hand, chip companies relying on advanced processes may face growth bottlenecks; on the other hand, TSMC, which holds pricing power, is poised to see a significant improvement in its profit margins. JPMorgan estimates that the tight supply has already generated “hot-run” orders, which will help TSMC’s gross margin climb to the lower-to-mid 60% range in the first half of 2026.

Capacity “Extremely Tight”

Faced with urgent customer demand, TSMC has not opted to build new N3 fabs. According to JPMorgan’s supply chain research, TSMC prefers to reserve newly constructed fabs in Kaohsiung (Fab 22) and Hsinchu (Fab 20) for even more advanced nodes such as N2 and A16. Analyst Gokul Hariharan noted that TSMC’s strategy is to encourage customers to migrate to leading-edge nodes faster rather than investing heavily in new N3 capacity toward the latter stages of its lifecycle.

Thus, the main increase in N3 capacity in 2026 will come from production line conversions at the Tainan Fab 18 facility. JPMorgan forecasts that as some accelerator demand moves to N3, N4 process utilization will fall below 100% in the first half of 2026, at which point TSMC will be able to convert approximately 30,000 wafers/month of N4 capacity to about 25,000 wafers/month of N3 capacity.

However, this conversion also involves uncertainties: if Nvidia is permitted to ship GPUs like the B30A to the Chinese market, it could lead to tighter N4 demand, thus slowing the N4 to N3 conversion pace.

Multiple Approaches to “Squeeze Toothpaste”

Apart from production line conversion, TSMC is also trying other ways to “squeeze” out more capacity.

The report mentioned that TSMC is adopting “cross-fab collaboration”, utilizing idle N6/N7 capacity at its Fab 14 to handle some of the back-end (BEOL) processes in N3 production. JPMorgan believes that, although this method is not very efficient, given N6/N7 utilization at about 65% in 2026, this approach could provide an additional 5,000 to 10,000 wafers/month of capacity in the second half of 2026.

In addition, there’s “overseas support.” The much-watched Arizona, USA facility (Fab 21), is expected to begin equipment move-in during Q2 2026 in its second phase, but it won’t supply about 10,000 wafers/month of N3 chips until early 2027, and thus cannot alleviate the 2026 supply crunch. Overall, JPMorgan estimates TSMC’s N3 total capacity by end-2026 will be about 140,000 to 145,000 wafers/month, still falling short of the reported market demand of 160,000 wafers.

Clients “Urgently Adding Orders”

In the context of extremely tight capacity, major tech giants have already taken proactive steps, joining the “grab for capacity” race. The lineup of buyers is almost a full roster of top tech giants.

In high-performance computing (HPC), beyond Nvidia’s Rubin and Rubin CPX, there’s also Broadcom’s TPU v7, Amazon’s Trainium 3, as well as Meta’s Iris and Microsoft’s Maia 300 custom ASIC chips.

In consumer electronics, Apple’s C2 baseband chip and iPhone 17, as well as Qualcomm and MediaTek’s flagship Android SoCs, will all be moving en masse to N3. Since the major clients have already booked up all capacity, the report explicitly states that crypto mining machine customers will “basically not be able to have their demand met” in 2026.

TSMC Margins Expected to “Significantly Improve”

The scarcity of capacity is being directly translated into TSMC’s profitability. JPMorgan’s supply chain research shows that due to the extremely tight capacity, several customers are already executing or planning “hot-run” and “super-hot-run” orders, paying 50% to 100% higher wafer prices to secure earlier deliveries.

The report notes that even though such high-priced orders usually don’t exceed 10% of total capacity, they are sufficient to have a significant positive impact on TSMC’s financial performance.

JPMorgan predicts that if hot-run orders persist, and with stable exchange rates, TSMC’s gross margin in the first half of 2026 should reach the low-to-mid 60% range, surpassing current market forecasts. In addition, TSMC’s plan to raise prices for advanced processes by 6%-10% starting in Q1 2026 will further underpin its profitability.

 

~~~~~~~~~~~~~~~~~~~~~~~~

The above exciting content is from ChaseWind Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join the 【ChaseWind Trading Desk · Annual Membership】

Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute individual investment advice, nor does it take into account the special investment goals, financial status, or needs of any individual user. Users should consider whether any opinions, views, or conclusions in this article fit their specific situations. If you invest based on this article, you are responsible for the risks.