According to informed sources, Taiwan Semiconductor Manufacturing (TSM.US) plans to increase its 3nm process wafer foundry pricing again in the second half of 2026, with a potential hike of up to 15%. Further increases of 5% to 10% are possible in 2027. Chairman and CEO C.C. Wei is expected to provide a systematic explanation regarding AI demand, advanced process deployment, and overseas expansion progress at the shareholder meeting on June 4, a topic of significant market focus.
This shift in supply-demand dynamics does not mark TSMC's first recent price increase. In Q4 2025, the company had already notified some clients of price hikes for its 5nm/4nm processes. However, unlike previous cycles driven by smartphone SoCs, the underlying rationale for this 3nm price increase is fundamentally different. It signals that semiconductor pricing power is formally transitioning from the "consumer electronics cycle" era to the "AI infrastructure arms race" era, initiating a fundamental transformation in the supply-demand structure for advanced processes.
The direct driver for TSMC's current 3nm price hike is a "historic shift" in demand structure. In the past, the primary demand for the 3nm process came from smartphone SoCs, supported by a few major clients like Apple. Today, the AI server upgrade cycle is in full swing. Companies including Nvidia, AMD, Google, AWS, and various cloud service providers are accelerating their adoption of 3nm technology. Demand for AI accelerators and custom ASICs is pouring in simultaneously, rapidly heating up wafer starts.
The core of this price increase lies in the demand structure shifting from a "single-engine" to a "multi-engine" model. ASIC industry participants point out that major cloud providers have been actively developing in-house ASICs in recent years to reduce reliance on general-purpose GPUs, further broadening the sources of demand for 3nm. The utilization rate at TSMC's main Fab18 facility remains high, with no significant easing in customer queueing situations. In terms of wafer starts, monthly 3nm capacity was around 130,000 wafers at the beginning of the year and has gradually increased to between 160,000 and 175,000 wafers in the second quarter. However, even with ongoing capacity expansion, the growth rate of AI demand continues to far outpace market expectations, making it difficult to close the capacity gap in the short term. Analysts at UBS commented on this in a report, stating, "AI demand is not a cyclical bubble but a multi-year capacity build-out cycle, where chip companies are currently constrained by supply capabilities."
From the supply side, the tightness in 3nm capacity is not a temporary allocation issue but a systemic shortage expected to persist until 2027. In the latest earnings conference call, C.C. Wei stated that the company is investing significant capital to accelerate the expansion of existing fabs and build new capacity to meet the continuously rising demand. He also acknowledged that, influenced by major clients like Nvidia, AMD, and Apple continuing to renew wafer orders, the supply shortage situation will extend into 2027. Concurrently, TSMC's 3nm wafer fab in Taiwan, originally slated to achieve a monthly capacity target of 150,000 wafers by the end of 2026, has seen this target revised upward to 180,000 wafers per month, approximately 20% higher than the original plan. However, Deutsche Bank had already issued a warning in January of this year—TSMC's 3nm capacity for 2026 is fully booked for the entire year and extends into 2027, forcing the company to significantly increase its capital expenditure plans for 2026.
Facing 3nm capacity pressure, TSMC has had to postpone some new 3nm projects, instead guiding clients to shift their mass production plans for 2027-2028 forward toward the 2nm GAA process. The supply-demand imbalance has even pushed some urgent-order clients to pay higher premiums to secure capacity. Last year, there were reports that some clients were willing to pay up to 100% more for 3nm emergency orders. This phenomenon is redefining the pricing anchor for advanced processes.
Placing the 3nm price increase within the context of TSMC's latest quarterly financial performance provides a clearer understanding of its strategic necessity. In Q1 2026, TSMC delivered what could be described as its "sweetest ever" quarterly report: quarterly revenue of approximately NT$1.134 trillion (about $35.9 billion), a 35.1% year-over-year increase; net profit reached NT$572.5 billion, up 58%; and earnings per share (EPS) were NT$22.08, setting a new record for the best quarterly performance ever. On the profit side, the non-GAAP gross margin for the quarter was a remarkably high 66.2%, up 7.4 percentage points year-over-year and 3.9 percentage points sequentially, breaking through the upper end of the previously estimated range of 63% to 65% to reach a historical high. The core drivers for the soaring gross margin include: an increased proportion of advanced processes (3nm/5nm gross margins already exceed 70%), capacity utilization maintained at a high level above 95%, economies of scale diluting fixed costs, and favorable exchange rates further boosting profitability.
Looking at the business structure, revenue from the High-Performance Computing (HPC/AI) platform grew 20% sequentially in Q1, officially surpassing 60% to account for 61% of total revenue. In contrast, the smartphone business declined 11% sequentially, dropping to a 26% share. The weakness in high-end smartphone end-demand presents a stark contrast to the explosive growth in AI demand. Looking ahead, TSMC has provided very strong guidance: full-year 2026 dollar revenue is expected to grow over 30%, with capital expenditure moving toward the upper end of the $52 billion to $56 billion range; Q2 revenue is projected to be between $39 billion and $40.2 billion, with the midpoint representing approximately 10.3% sequential growth and about 32% year-over-year growth, while the gross margin is expected to remain in the high range of 65.5% to 67.5%. Based on the long-term high growth cycle for AI computing power, the company has signed long-term supply agreements with core North American clients to lock in future orders. The HPC/AI business is expected to maintain a high growth rate, and the overall high-growth trend for 2026 appears strongly assured.
Another layer of logic supporting this round of price increases lies in the structural upward shift in TSMC's own cost base. On one hand, overseas fab construction—including in Arizona, USA; Kumamoto, Japan; and Dresden, Germany—brings significant incremental capital expenditure and operational costs. On the other hand, depreciation pressure from advanced processes continues to increase, coupled with the initial yield ramp-up phase for the 2nm process. TSMC needs to support overall gross margin stability through reasonable pricing of its mature processes. Market analysis suggests that compared to the 2nm process, which is still in its early yield ramp-up phase, the 3nm process offers more advantages for AI clients in terms of mass production stability and cost control, making it the most mature and reliable advanced process option currently available. As the global AI computing power race intensifies, competition in advanced processes has evolved from pure technology iteration to a comprehensive contest involving capacity scale, yield levels, and supply chain integration capabilities. TSMC holds significant leading advantages across all these dimensions.
UBS further notes that TSMC's ability to raise prices on its most advanced nodes indicates the company can capture a larger share of the value created by AI computing demand. However, higher wafer costs will ultimately be passed on to chip buyers—Nvidia, AMD, and hyperscale cloud providers—which could compress their gross margins or drive up the price of AI hardware.
The timing of this price increase news coincides with a closely watched internal "bonus controversy" at TSMC. Reports indicate that TSMC recently considered potentially reducing the employee bonus allocation ratio, sparking backlash among the rank and file. Some employees reportedly called for actions like "turning off phones after work and not being on call" on social media platforms. Chairman C.C. Wei urgently issued a letter to all employees on the afternoon of May 26, canceling original business travel plans, opening the performance bonus inquiry system early, and personally taking the lead to hold an employee communication meeting on the 27th, making 41 lecture halls and meeting rooms available for employees to register online for participation. On Wednesday, Wei told employees that the average profit-sharing bonus this year would increase by over 30%, responding to concerns some employees expressed online about incentive plans. As a key pillar of the global AI infrastructure, TSMC is expanding employee incentive plans alongside soaring profits. According to informed sources, at the internal company-wide employee meeting on Wednesday, Wei expressed confidence that the per-employee profit-sharing bonus for Taiwan-based staff would see a year-over-year increase exceeding 30%, better than the growth level of the previous year.
Industry analysts believe that to maintain employee bonuses at high levels, a large-scale price increase for clients has become an inevitable choice for TSMC based on cost considerations, foreshadowing the 15% hike for the second half of the year.
With the release of the price increase signal, market focus is shifting to TSMC's upcoming annual shareholder meeting on June 4. C.C. Wei is expected to provide a systematic explanation on topics including AI market demand, advanced process deployment, and overseas expansion progress. This will be the market's first opportunity to directly obtain formal management positions on pricing strategy and technology roadmaps following the price hike rumors. Notably, just before the shareholder meeting, a "working dinner" between Nvidia CEO Jensen Huang and TSMC's senior executives has also attracted market attention. When asked if TSMC could meet Nvidia's capacity demands, Wei reportedly responded with a "zipper over the mouth" gesture, saying, "We are already working very hard." This interaction further reinforces market expectations of continued tightness in advanced process capacity. The June 4 shareholder meeting will not only provide a formal explanation of the price increase logic but will also serve as a key window for global semiconductor investors to assess the sustainability of AI demand, the advanced process supply landscape, and the boundaries of TSMC's long-term pricing power. For clients directly purchasing 3nm wafers, such as Nvidia and AMD, TSMC's price increase will raise the manufacturing cost of their AI chips—costs that may ultimately be passed on to end-customers, namely hyperscale cloud service providers like Google, Microsoft, and Amazon, through higher AI hardware prices. In a market where competition for computing power scale remains a core objective, the capital expenditure of cloud service providers may face even greater pressure.
Comments