The Circuit Ep 149: TSMC Q4 25 Earnings, OpenAI Needs more Compute and Monetization
36 snips
Jan 19, 2026 The hosts dive into TSMC's recent earnings, unpacking the complexities of semiconductor cyclicality and the challenges posed by advanced packaging. They explore TSMC's cautious CapEx strategies amid rising AI demand and discuss the intricate relationship between wafer demand and manufacturing capacity. The conversation then shifts to OpenAI's partnership with Cerebras, examining new monetization efforts like ads and subscription tiers to address compute costs. They also analyze the fluctuating GPU pricing landscape and the implications for hyperscalers and enterprise clients.
AI Snips
Chapters
Transcript
Episode notes
TSMC Is Not A Reliable End‑Market Signal
- TSMC is a poor end-market bellwether because it's a constrained, high‑fixed‑cost foundry with limited upside per capacity unit.
- Its capital intensity and cyclicality make it slow to detect demand downturns and risky to overbuild ahead of real demand.
Huge CapEx Raises Stakes On AI Demand
- TSMC expects AI accelerator demand to grow ~50% and is massively increasing CapEx, creating large downside if AI demand fades.
- They warned that heavy CapEx makes them vulnerable if the AI boom proves transitory.
Packaging Is The New Capacity Bottleneck
- Advanced packaging (CoWoS) is a major bottleneck because it requires manual, complex steps that are hard to scale and automate.
- Large AI chips consume many wafers and packaging capacity, amplifying foundry constraints beyond pure lithography limits.