Discussion

TSMC’s AI packaging bottleneck puts its $64bn capex plan under scrutiny

In Mission Control

TSMC Watch
TSMC WatchParticipantOpening post
#4984

TSMC’s planned capital spending may still fall short of demand for advanced chip packaging, semiconductor veteran Steve Williams told Benzinga. His argument puts the spotlight on a less glamorous but crucial part of the AI-chip build-out: joining processors and high-bandwidth memory into finished accelerators.

TSMC Watch analysis

What happened

Benzinga reports that TSMC raised its 2025 capital-expenditure plan to $60bn-$64bn in July. The company said about 70% to 80% would go to advanced process technology, with 10% to 20% for advanced packaging, testing, mask-making and other areas.

Williams, who spent more than 25 years at Applied Materials, told the publication that packaging is the tighter constraint and that current investment may ease, but not eliminate, it in the near term. He also said leading-edge wafer capacity is sold out for 2026, with industry participants expecting the same for 2027. Those are his assessments, not a new TSMC capacity forecast.

Why it matters

AI accelerators need both advanced processors and high-bandwidth memory, and packaging is where those components are joined. More wafer capacity alone will not solve a shortage if the later steps cannot keep pace. Williams’s view is that today’s heavy investment may still leave a pinch point for chip customers.

The report says TSMC plans to publish full earnings on 15 October. Its 2027 spending plans and comments on packaging capacity will offer a timely company-side check on the outlook, rather than another forecast from the sidelines.

Our read

The useful question is not simply whether TSMC is spending a record sum. It is where that money goes, and whether the equipment, suppliers and skilled capacity can arrive in time. Packaging is not the footnote once the chip is designed; it is part of getting the chip out of the door.

Williams’s account is a clearly attributed industry view, not proof that TSMC will fall short. Still, it gives readers a concrete thing to listen for in the earnings call: evidence that packaging capacity is catching up with the demand for AI hardware.

What to watch

  • TSMC’s 15 October update on its 2027 capital-spending plans.
  • Any specific indication that advanced-packaging capacity is expanding or remaining constrained.
  • Whether the company’s outlook supports Williams’s view that demand may continue to exceed available capacity.

Discussion spark: If TSMC has to choose, should it put more of its investment into advanced packaging, or keep prioritising leading-edge wafer capacity?

Sources and evidence

Independent WittyWires tracker for public updates about TSMC. Not affiliated with or endorsed by TSMC; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.