The only pure-play foundry at the leading edge at scale — it manufactures the chips it does not design for NVIDIA, AMD, Broadcom and Apple alike, and owns the CoWoS packaging that governs how many AI accelerators ship. The chokepoint the entire complex depends on.
Our Position
Research
Primary sources
Research Trail
Covered in the Knowledge Base
Hardware & Computing
In the Atlas
TSMC in the frontier-stack Knowledge Base
The balance sheet is fixed but the business is not — at ~4x EV/sales with a -27% GAAP gross margin, a commoditizing substrate core, and a ~35% burned-strategic overhang (Renesas), WOLF is priced for an AI-datacenter / 200mm-utilization inflection the P&L will not confirm before FY27; WATCHING until gross margin crosses zero.
The purest listed play on the Big-3's DDR4 exit — a real supply-shock oligopoly windfall (GM 30%→53% in six quarters, sold out through 2027), but +890% in a year prices a transient legacy-DRAM squeeze as if it were a structural moat; WATCHING, not chasing, at 52× trailing peak-cycle earnings.
Source documents — open to read in full
This is a narrow pre-earnings refresh, not a re-run of the 14-lens deep-dive battery. It exists to do three things ahead of Thursday's Q2 2026 print (per _index.json/filings, TSMC guided the release for 2026-07-16; some public calendars list it 2026-07-16 US time — reconfirm the exact print time before Thursday):
deep-dive-2026-07-10.md, which does carry a verdict + Position-seed section). That omission is intentional, not an oversight.Full 14-lens re-run (management commentary, forensic checks, comps, valuation ladder) should wait for the actual Q2 print — this file will be superseded by a real post-earnings refresh once Thursday's numbers are out.
All four filed as Form 6-K (TSMC is a foreign private issuer — no 10-Q/10-K), fetched directly from SEC EDGAR (data.sec.gov / www.sec.gov, not blocked for direct fetch — WebFetch to sec.gov 403s, but the repo's scripts/research/ingest-filing.ts / raw Node fetch do not):
| File | Accession | Filed | Report period | Content |
|---|---|---|---|---|
filings/6-k-2026-q2.md | 0001046179-26-000447 | 2026-07-13 | 2026-06-30 | June 2026 monthly revenue report (the filing this refresh was ordered to ingest) — June + May monthly, Jan–Jun cumulative |
filings/6-k-2026-04-monthly-revenue.md | 0001046179-26-000213 | 2026-05-08 | 2026-04-30 | April 2026 monthly revenue, Jan–Apr cumulative |
filings/6-k-2026-03-monthly-revenue.md | 0001046179-26-000136 | 2026-04-10 | 2026-03-31 | March 2026 monthly revenue, Jan–Mar (Q1) cumulative |
filings/6-k-2026-q1-earnings-release.md | 0001046179-26-000199 (Ex-99.1) | 2026-04-16 | 2026-03-31 | Q1 2026 earnings press release — Q1 actuals (revenue, GM, op margin, net margin, EPS) and the original Q2'26 guidance (revenue $39.0–40.2B; GM 65.5–67.5%; op margin 56.5–58.5%; guidance FX assumption NT$31.7/US$1.00) |
Note: TSMC's June monthly report was itself delayed from its normal ~10-day-after-month-end cadence to July 13, attributed to a Taiwan typhoon (per the prior dossier's Lens 5 note) — worth remembering if Thursday's print timing also slips.
Carried claim (Monday S1 v4, summary-derived): "Q2 revenue ~$39.6B effectively printed via monthly 6-Ks (guidance $39.0–40.2B)."
This pass's primary recomputation: Summing the three monthly 6-Ks (or equivalently, subtracting the Q1 cumulative from the H1 cumulative — both methods cross-check to the same NT$ figure):
filings/6-k-2026-03-monthly-revenue.md]filings/6-k-2026-q2.md]filings/6-k-2026-04-monthly-revenue.md + filings/6-k-2026-q2.md] — the two independent computations agree to within NT$0.16M (rounding only).Converting to USD (TSMC's own P&L translation methodology; no primary Q2 actual-average-FX figure exists yet — that only appears in Thursday's release):
filings/6-k-2026-q1-earnings-release.md]: Q2 revenue ≈ US$40.08Bfilings/6-k-2026-q1-earnings-release.md]: Q2 revenue ≈ US$40.21BFlagging the discrepancy rather than silently overwriting it: both FX-conversion methods land Q2 revenue at ~$40.1–40.2B — at or effectively at the top of TSMC's own $39.0–40.2B guided range, not the ~$39.6B (closer to mid-band) figure carried from Monday's pass. I cannot fully reconcile why the Monday figure came in lower — possibilities: a different (weaker-USD) FX assumption, a partial-month data cut at the time of that pass (before the June 6-K existed), or the Motley Fool prediction-piece source cited in the shorts state file used its own estimate rather than TSMC's own cumulative arithmetic. Either way, the primary-source math this pass supports a high-end-of-guidance print, arguably the single most decision-relevant delta from this refresh. This is a data finding, not a directional call — what it implies for Thursday is left to the scenario table below and to Connor's own read.
Carried claim (Monday S1 v4): "Gross-margin staircase +740bps; Q2 GM guidance band 65.5–67.5%."
This pass: Re-derived directly from the Q1'26 earnings release's own side-by-side 1Q26/1Q25 table [primary: filings/6-k-2026-q1-earnings-release.md]:
Carried claim (Monday S1 v4): "Capex $52–56B guided, tracking high end."
This pass: No primary-filing capex actual exists for Q1/Q2 2026 in a text-parseable SEC document (TSMC's quarterly capex actuals surface on the earnings call / slide deck, not the brief press-release exhibit, and the Q1 presentation exhibit is image-only JPGs, not machine-readable text). Corroborated via web aggregation of the Q1'26 earnings call: Q1 2026 capex actual ≈ $11.1B, with management guidance language described as "towards the high end" of the $52–56B full-year range [summary-derived: MLQ.ai / DataCenterDynamics / Yahoo Finance Q1 FY2026 earnings-call coverage, pulled 2026-07-14]. Status: carried forward, unchanged, still summary-derived (not independently primary-verified this pass — a genuine gap, not a confirmation).
Carried claim (Monday S1 v4): "CoWoS sold out through 2027."
This pass: Not re-verified against a new primary source this pass (no primary filing text discloses forward CoWoS booking in the documents I ingested — that level of detail lives on earnings calls/analyst days, not the monthly-revenue 6-Ks). Consistent with the standing dossier's own Lens 2/3 sourcing (CoWoS 130–150k wpm capacity by end-2026; NVIDIA >60% of 2026 packaging + >half of 2026–27 expansion). Status: carried forward, unchanged, summary-derived.
Carried claim (Monday S1 v4): "HPC 61% of revenue mix."
This pass: Corroborated via web search of TSMC's own Q1'26 earnings-call/slide disclosure (not independently retrievable as parseable primary text — the Q1 presentation exhibit on EDGAR is JPG images only): HPC revenue +20% QoQ to 61% of Q1 2026 revenue-mix, smartphone 26%, IoT 6%, automotive 4%, DCE 1% [summary-derived: GuruFocus/Investing.com Q1 2026 earnings-call-highlights coverage, pulled 2026-07-14]. This is the Q1 2026 mix (not FY2025, which the standing dossier has at HPC 58% [primary: 20-F]) — the two figures are not in conflict, they're sequential quarters showing the mix continuing to shift toward HPC. Status: carried forward, confirmed via independent web corroboration, still summary-derived (no primary machine-readable text source found).
Q2 2026 print expected 2026-07-16 (reconfirm exact date/time before Thursday). Three GM outcome bands against the guided 65.5–67.5% range, and what each mechanically implies for (a) the margin-staircase narrative and (b) consensus, stated as mechanical facts only:
| Scenario | GM outcome | Mechanical implication for the margin-staircase narrative | Mechanical implication for consensus |
|---|---|---|---|
| Beat | GM > 67.5% (above guided high end) | Staircase continues past the guided band — a 5th consecutive quarter of QoQ or YoY GM expansion (1Q26 was 66.25%; anything >67.5% is a fresh sequential high). Extends the +740bps YoY trend already confirmed above rather than merely meeting it. | Full-year 2026 GM run-rate assumptions get revised upward; any model anchored to the 65.5–67.5% guided band understates FY26 GM if this band is the new floor, not the ceiling. Analyst full-year EPS estimates mechanically require an upward revision unless offset by revenue or opex assumptions moving the other way. |
| In-band | GM 65.5–67.5% (within guided range) | Staircase narrative holds exactly as guided — no data point requiring the thesis to be re-underwritten in either direction. Consistent with management's own Q1 forward-guidance credibility (Q1 guidance was also hit/beaten per the primary release). | Consensus, which per the standing dossier's Lens 5 is built off the guided band, requires no mechanical revision. The revenue print (see scenario-independent finding above: implied ~$40.1–40.2B via primary math) is the more consequential number in this outcome, not GM. |
| Miss | GM < 65.5% (below guided low end) | First GM deceleration since the FY2025 20-F documented a "clean 3-year climb" (54.4% → 56.1% → 59.9% → 66.2% Q1'26). Would be the first data point in over a year inconsistent with the "margins expand even into the heavy-capex year" framing carried across the last two dossier refreshes. Mechanically it would NOT by itself validate the bear case laid out in the standing dossier's Lens 10/13 (D&A wave, overseas dilution) — management already guides 2–3pt N2 dilution + 2–4pt overseas dilution within the 65.5–67.5% band, so a below-band print would need to exceed even that disclosed dilution math to be a genuine surprise rather than a guided-for outcome landing at the guided floor. | Requires distinguishing "missed own guidance" (a credibility event — TSMC has not missed its own quarterly guidance band in the data on hand) from "guided-for margin compression materializing on schedule." Consensus models that already bake in the 2–3pt/2–4pt disclosed dilution factors would see limited surprise; models that extrapolated the +740bps YoY run-rate flatly forward would see the larger miss. |
Scenario-independent finding to weigh alongside GM, from this pass's primary recomputation above: the revenue side has, per the primary monthly-6-K arithmetic, effectively already printed at ~$40.1–40.2B — at or near the top of the $39.0–40.2B guided range — regardless of which GM scenario plays out. GM and revenue are the two separate axes Thursday's print will move; this table only addresses the GM axis per the task scope.
deep-dive-2026-07-10.md, web-sourced, not re-verified this pass]; HPC 61% of Q1'26 mix [GuruFocus, Investing.com, pulled 2026-07-14].research/calendars/earnings.json or TSMC's IR calendar this pass — worth a 30-second check before Thursday.fetch proved sufficient and more authoritative); if a future pass needs the presentation slide images (for CoWoS/HPC-mix primary confirmation), TSMC's own IR PDF URLs were located via web search (e.g. investor.tsmc.com/english/encrypt/files/...1Q26%20Presentation%20(E).pdf) and may be fetchable directly, unlike sec.gov via the WebFetch tool.A textbook semiconductor cyclical wearing a secular-growth costume — the 2024–25 CIS recovery is real and automotive is a genuine share-taking engine, but 2026 consensus EPS has already been cut BELOW 2025's actual and Q1-2026 net profit fell ~42% YoY while the stock still holds ~35x trailing; the automotive ramp must out-run mobile normalization and domestic price competition to justify the multiple. WATCHING — constructive only on a reset toward ~25x or hard evidence auto mix is structurally l