TSMC to raise prices on 3nm chips by up to 15% due to AI boom

TSMC Plans to Raise Prices on 3nm Chips by Up to 15% Due to AI Boom

TSMC plans to raise prices on its 3nm process by up to 15% in the second half of the year, with an additional 5-10% increase next year, driven by growing demand for chips from NVIDIA, AMD, and cloud services for AI tasks.

---

Analytical Article: 15% for 3nm. Why TSMC is Pressing the Panic Button, Not the Greed Button

Author: Independent Analyst with Insider Perspective

Date: 2026-05-28

When news broke on May 27, 2026, that TSMC was raising prices on 3nm wafers by 15% in the second half of the year and preparing another 5-10% in 2027, everyone talked about the "greed of the Taiwanese monopolist." But if you think this is just a desire to squeeze an extra billion out of NVIDIA, you're wrong.

This is a cry for help disguised as a price hike.

TSMC has found itself trapped by its own success. Demand for 3nm from AI clients has exceeded all forecasts so much that the company can't ramp up capacity fast enough, even with capital expenditures of $520-560 billion this year. And no one in the mainstream media is mentioning that this news came out exactly one week before the annual shareholder meeting on June 4, amid an internal employee revolt over bonuses. It's all interconnected.

[The Gist]: What's Really Happening

The formal reason is a capacity shortage amid the AI boom. The real reason is a structural shift that TSMC cannot control.

Previously, 3nm relied on three or four major customers from the smartphone world (Apple, Qualcomm, MediaTek). Today, according to supply chain data, the list includes NVIDIA with the Vera Rubin platform, AMD with Instinct M1350/M1355, Google with TPU V8 and V7E, AWS with Trainium 3, Meta with MTIA-3, and even OpenAI with their Titan chip. That's 7-8 giants simultaneously, plus dozens of ASIC projects from hyperscalers. Each requires not thousands, but tens of thousands of wafers.

The problem is that TSMC has already ramped up Fab 18 capacity from 130,000 wafers per month at the start of the year to 160,000-175,000 in the second quarter. But even that is not enough. Clients are still waiting in line. And TSMC management has made a shocking decision: instead of further expanding 3nm (though the plan has already been raised from 150,000 to 180,000 wafers by year-end), they are starting to send clients to 2nm.

Yes, you heard that right. TSMC is raising 3nm prices by 15% to cool demand and shift attention to 2nm, which is expected to enter volume production in late 2026 or early 2027. But 2nm currently has low yields and high costs. Clients don't want to go there. So TSMC is forced to make the old node (3nm) so expensive that the new one (2nm) suddenly doesn't seem so scary.

My non-obvious insight, which most readers don't know: the reason for the price hike is not AI or inflation, but an internal employee revolt.

24 hours before the price leak, on May 26, 2026, a call spread across Taiwanese social media for TSMC employees to "turn off phones after work and not be on standby" in response to rumors of a reduced bonus pool contribution. CEO C.C. Wei (魏哲家) urgently canceled business trips, sent a company-wide email, and held an all-hands meeting on May 27, opening 41 conference rooms for online participation.

At that meeting, he promised that the average bonus per employee would increase by more than 30% year-over-year. Now do the math: TSMC's total compensation is tens of billions of dollars. A 30% increase means an additional $5-7 billion per year. Where will that come from? Only from raising prices on the most mass-market premium product—3nm. The 15% is not about margins; it's about plugging a hole in the salary budget.

Timeline and Context

Let me overlay timestamps to show this is no coincidence.

• November 2025 – February 2026: TSMC quietly raises prices on 5/4nm by 3-10%. The market barely notices.
• March 2026: TSMC announces 2026 capital expenditures of $520-560 billion—a record in company history.
• April 2026: In an earnings call, Wei says the capacity shortage will last into 2027 and beyond.
• May 25-26, 2026: Surge of employee discontent on social media, threats of a "silent strike."
• May 26, 2026 (evening): Leak to media about the 15% price hike.
• May 27, 2026: Wei holds emergency staff meeting, promises +30% bonuses.
• June 4, 2026 (expected): Annual shareholder meeting where Wei will officially confirm the new pricing policy.

See the sequence? First, worker revolt. Then, price leak. Then, bonus promise. Then, official confirmation. They couldn't announce bonus increases without a confirmed revenue source. The 3nm price hike is the financial backing for promises made to a frightened workforce.

Who Wins and Who Loses

Winners (obvious):

• TSMC employees in Taiwan. Their bonuses will rise 30% this year. That's hundreds of thousands of dollars for senior engineers.
• TSMC shareholders. The price hike directly boosts margins. UBS has already raised its target price to 3,000 Taiwanese dollars per share.

Winners (non-obvious):

• Samsung Electronics. Yes, the main competitor. Because TSMC, by raising prices, opens a "window of opportunity" for Samsung with their 3nm GAA (Gate-All-Around). Clients unwilling to pay TSMC's +15% might at least start negotiations with the Koreans. Samsung is desperately in need of major customers to recoup its $14 billion investment in the new Taylor, Texas plant.
• Intel. Intel's own 18A (equivalent to 2nm) isn't yet ready in volume, but their 3nm-compatible capacity (Intel 3) could get a "second wind" as a budget alternative.

Losers:

• NVIDIA and AMD. Their margins will shrink. Every 15% wafer price increase translates to a 2-3% hit to gross profit on the finished product, which they can't raise prices on instantly because contracts with cloud providers are locked in for quarters ahead.
• Hyperscalers (Google, AWS, Meta). They build their own chips (TPU, Trainium, MTIA) hoping to save money. But now those chips are suddenly getting more expensive at the production stage, before they've even gone into volume.

What the Media Isn't Saying

Taiwanese press writes about a "historic shift" and an "era of AI pricing." But there are two skeletons in the closet that go unmentioned.

First: TSMC physically cannot produce more 3nm wafers. Equipment for EUV lithography (ASML Twinscan NXE:3800E) is delayed. To expand Fab 18 to 180,000 wafers per month, they need 40-50 new scanners. ASML produces about 60 per year globally. TSMC competes for them with Intel and Samsung. Supply delays are inevitable.

Second omission: Prices have risen, but volumes have not. This increase is not about "selling more"; it's about "selling the same amount, but for more money." Fab 18 is already at 100% utilization. TSMC has no spare capacity. So the price hike simply redistributes the existing pie: clients who can pay stay; those who can't move to 5nm or competitors. This is not business growth; it's segmentation by ability to pay.

Forecast: Next 30 Days and 90 Days

Next 30 days (by end of June 2026):

Expect official confirmation from Wei at the shareholder meeting on June 4. But the key will be NVIDIA's reaction. If on June 3-5, Jensen Huang (CEO of NVIDIA) announces price increases on their B300 or Rubin GPUs, it means TSMC has pushed through its plan. If Huang stays silent, there's tough backroom bargaining, and the final price could be reduced to 10-12%. We'll see an interim result 7-10 days after the meeting.

Next 90 days (by end of August 2026):

The first 2nm contracts with "take it or leave it" terms will start appearing. TSMC will aggressively promote 2nm, offering discounts to the first three clients that sign long-term agreements. Likely candidates: Apple (always first on a new node) and one of the hyperscalers (Google or AWS).

But for most of the market, this means 3nm will become a "premium tier," and the mid-range AI accelerator segment will shift back to 5nm and 4nm, which will also rise in price following 3nm (domino effect).

However, my main forecast: TSMC's internal problems won't go away. Bonuses have increased, but engineers in Taiwan are burning out. The turnover rate at Fab 18, according to unofficial data, has reached 15-18% annually. Raising salaries is a temporary painkiller. If TSMC doesn't solve the overtime and "always on" culture, by the end of 2026 we'll see not just price hikes, but real supply disruptions. And then 15% will seem like small potatoes.

— Editorial Team