SWIN

3.2800.00%

AI Build-Out Has ‘Broken the Old Demand Cycle,’ Chip Veteran Says as TSMC Posts Record Profit

LongbridgeAII'm LongbridgeAI, I can summarize articles.

TSMC reported record Q3 profit, driven by AI demand exceeding supply. Industry veteran Steve Williams argues that AI has broken the traditional semiconductor cyclical demand pattern, suggesting current market fears are misplaced. TSMC's 2nm process revenue share jumped significantly, indicating strong adoption. Despite a recent sector selloff, Williams expects sustained chip demand due to constrained capacity in leading-edge manufacturing and packaging, with equipment makers potentially becoming the next bottleneck.

TSMC (NYSE:TSM) posted record profit Thursday and said artificial intelligence demand still exceeds supply, even as semiconductor stocks extend a sharp selloff.

The Philadelphia Semiconductor Index fell 4.3% Wednesday and is down about 11% over six sessions, according to Reuters.

Semiconductor industry veteran Steve Williams, who spent more than 25 years at Applied Materials Inc. (NASDAQ:AMAT), says investors may be applying an outdated playbook.

"They’re still in this fear of the old cyclical demand cycle," Williams told Benzinga. "And the current technology and AI build-out has broken the old demand cycle."

2nm Makes a ‘Big, Big Jump’

TSMC’s third-quarter net profit rose 69%. The company guided to $52.3 billion to $54.3 billion in fourth-quarter revenue, roughly 58% year-over-year growth at the midpoint.

But one number stood out to Williams.

TSMC’s 2-nanometer process jumped from 3% of wafer revenue in the second quarter to 8% in the third. 2nm is TSMC’s newest leading-edge process, and the rapid increase likely points to strong early adoption of its latest technology.

"I expected an increase, but that’s a big, big jump," Williams said.

He expects that share to keep rising as leading-edge customers compete for limited capacity.

TSMC kept 2026 capex at $60 billion to $64 billion, but Williams said the unchanged figure does not signal weaker demand. Equipment arriving this year was ordered months ago, with some larger tools carrying lead times of six months or more.

The Bottleneck Keeps Moving

Williams told Benzinga last week that advanced packaging was TSMC’s immediate bottleneck.

TSMC has signed a $2 billion manufacturing agreement with GlobalFoundries to add U.S. silicon-interposer capacity for its CoWoS packaging ecosystem.

But solving packaging would simply move the constraint, Williams said, most likely back to TSMC’s ability to manufacture enough leading-edge 2nm chips.

Push expansion further and equipment makers themselves may become the bottleneck.

"They can only go so far before ASML can’t make what they need," Williams said.

ASML (NASDAQ:ASML) said this week its planned supply will not fully satisfy customer demand in 2027 or 2028.

Demand May Outlast the Building Boom

Williams sees chip demand persisting even if the current data-center construction wave slows.

The growing installed base of data centers will still require chips as computing equipment is replaced and upgraded, he said.

"Even if it slows down, the chip demand will be pretty solid," Williams said.

Williams also argues the industry currently has several overlapping sources of demand while leading-edge wafers, packaging and equipment remain constrained.

Polymarket traders put roughly a 6% chance on an AI-bubble burst by Dec. 31, and a 19% chance by June 30, 2027.

"There’s still a lot of stuff not being built today on the chip side that if there was supply, it would be built," Williams said. "It’s massive."

Image: Shutterstock

Read Also: AMD Challenges Nvidia's AI Inference Lead, SemiAnalysis Finds

Login to unlock2,054characters for free

Due to copyright restrictions, please log in to your Longbridge account to view this content.
Thank you for your understanding and support of licensed content.