TSMC's position is structurally simple: every major AI accelerator design in our analysis is fabricated by TSMC. The substitution arc within the AI compute market (Nvidia vs Broadcom ASICs vs AMD vs Google TPU) is largely silent on TSMC because all the named players use TSMC capacity. Whichever silicon "wins" share, TSMC captures the foundry revenue.
The Nvidia Blackwell ramp. Blackwell volumes are TSMC CoWoS-packaged on 4nm. The B200 / GB200 / GB300 ramp curve directly translates to advanced-node revenue at TSMC. The May 19 weekly's discussion of B200 demand-pull scarcity implicitly reflects packaging capacity constraints at TSMC as much as Nvidia's own demand.
The custom-silicon trajectory. Broadcom's $20bn 2025 → $100bn+ 2026 AI revenue trajectory rests on multi-hyperscaler ASIC programmes (Google TPU, Meta MTIA, Anthropic, OpenAI) — all fabricated by TSMC. The growth in custom-silicon volume is additive to TSMC's revenue base on top of Nvidia's continued growth.
The substitution arc neutrality. Per the May 17 brief's implicit framing: if Nvidia retains share, TSMC captures the foundry revenue. If Broadcom's ASICs gain share, TSMC captures the foundry revenue. If AMD's MI450 wins training capacity at Meta and OpenAI, TSMC still captures the foundry revenue. The only scenario where TSMC loses meaningfully is a credible shift to a competing foundry (Samsung, Intel Foundry) — neither of which currently has the same advanced-node capability at scale.
The geopolitical / Taiwan risk. Not directly covered in our recent briefs, but worth flagging: TSMC's Taiwan concentration introduces a non-trivial tail risk that is partially mitigated by the Arizona Fab 21 build-out and Japan / Germany expansions. Coverage in our pricing tape doesn't reach this dimension directly; news / SEC filings carry it.
Coverage depth note. TSMC has appeared in only 4 paragraphs across our recent briefs — much less than the upstream / downstream names it supplies. Future briefs that surface TSMC-specific data (CoWoS capacity, 2nm node ramp, advanced-packaging utilisation) would deepen the standalone TSMC narrative beyond the inferred picks-and-shovels framing.
The composite picture: a name with structurally-positive exposure to every AI silicon thread in Signwl's analysis, with light direct coverage in our briefs but consistent indirect exposure through every accelerator-related thesis. The single best "all of the above" position in the AI accelerator value chain.