high convictionactive · updated 2026-07-24T00:00:00.000Z
TSMC saturation → 4-year price hikes → marginal buyers shift → Intel anchor stack → re-rate
TSMC leading-edge capacity is structurally 3x short of customer plans. 2nm sold out through 2026, 3nm booked through 2028. TSMC notified customers of price increases for four consecutive years. Marginal buyers (AWS, Microsoft, Apple-preliminary, Terafab) commit to Intel 18A/14A as the alternative. Government brokered Apple-Intel. Picks-and-shovels (ASML, AMAT, KLA, LRCX) capture capex from both build-outs.
The chain
1
TSMC leading-edge capacity is ~3x short of customer plans. 2nm fully booked through 2026; 3nm booked through 2028.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "TSMC CEO C.C. Wei stated capacity is 'about three times short' of customer plans (SIA Awards Nov 20 2025)"
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "2nm sold out through 2026; combined Taiwan capacity ~90,000–100,000 wafers/month in early 2026; both plants fully booked for the year"
2
TSMC announced 2nm wafer price increases for four consecutive years starting 2026 — a sustained margin tailwind for TSM and a sustained cost push on customers.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "Customers notified that they will bear price increases on 2nm wafers for four consecutive years starting 2026... a direct multi-year tailwind for TSMC's margins and a direct push for customers to find alternatives"
3
TSMC's response capacity arrives multi-year late. Arizona 2nd-fab 3nm in 2H27; Kumamoto in 2028. CEO admits shortages drag into 2027 and beyond.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "Arizona 2nd fab 3nm volume production targeted for 2H27; Kumamoto in 2028. TSMC is also pouring $56B into new fabs but Wei admitted shortages will drag into 2027 and beyond"
4
Marginal buyers commit to Intel 18A/14A as the alternative — AWS (multi-billion 18A), Microsoft (Maia 2 on 18A/18A-P), Apple-preliminary (M-class 2027, iPhone 2028), Terafab (Tesla/xAI/SpaceX on 14A). Trump personally pushed Apple-Intel.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "Intel has stacked four anchor foundry customers (AWS, Microsoft, Apple-preliminary, Musk's Terafab) against a backdrop of TSMC capped 3x short of AI demand"
lip-bu-tan in 2026-04-23-earnings-intc-q1-2026: "Long-term agreements signed — including with Google, typically spanning **3-5 years**." — Primary-source CEO confirmation that Google has a **3-5 year LTA** for server CPUs and ASICs, making Google a named anchor customer at step 4. This corroborates and strengthens the anchor-stack claim with primary-source Q1 2026 earnings call testimony.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "Apple CEO Tim Cook stated on the Q2 2026 earnings call that 'availability of advanced nodes' is Apple's 'greatest supply chain constraint'"
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "the US government played a key role in the Apple deal, with President Trump personally advocating for Intel during a White House meeting with Cook"
From 2026-05-11-autoresearch-apple-intel-deal-may-2026-update: "Intel's Q1 2026 earnings (April 23) revealed 18A yields are improving 7–8% per month, and the yield target has been advanced by six months to mid-2026 — roughly two quarters ahead of the original year-end 2026 plan"
From 2026-05-11-autoresearch-us-fab-bottleneck-anchor-update-may-2026: "Intel boosted equipment orders 50%+ YoY (TrendForce, April 20) signaling genuine capacity expansion commitment; Intel is expanding its High-NA EUV allocation from 1 to 2 units"
From 2026-05-15-autoresearch-exec-capex-statements-may-12-15: **Tesla $3B Austin semiconductor R&D fab on Intel 14A** (Tesla Q1 earnings, April 22, 2026) — distinct from Terafab ($25B+ consortium). Produces "a few thousand wafers per month" at R&D/pilot scale. Two separate Tesla-ecosystem commitments to Intel 14A now on record: the $3B R&D fab (operational/near-term) and Terafab (production-scale, multi-year). This is additional 14A customer validation beyond the Terafab commitment and ahead of Terafab's production ramp.
**⚠ Counter-evidence (2026-07-23 Q2 print).** From 2026-07-23-earnings-intc-q2-fy2026 + 2026-07-24-autoresearch-intel-foundry-18a-14a-external-customers-q2-2026: at the Q2 print Intel **named no external 18A/14A customer** despite raising FY26 capex to >$20B; external foundry revenue was only **~$293M** on a $5.8B (mostly-internal) Foundry segment, and 14A still shows only "two unnamed prospects" (firm decisions H2 2026→H1 2027). The one *fresh* name — Fortinet (07-21) — is on **Intel 4** (a security ASIC), not leading-edge. Under direct analyst pressure (david-zinsner: capex is "gated by concrete customer commitments … long-term agreements … prepays"; lip-bu-tan deflected to PDK/yield), management asserted but would not *name or quantify* an external anchor. **Read:** Step 4's *disclosed, verifiable* demand remains **internal** (record DCAI +59% YoY on Xeon 6); the external-anchor leg is management-asserted, not print-confirmed — Step 4 stays `partial`. **Falsifier to watch:** if capex keeps rising while external foundry revenue stays ~$0.3B/qtr, the "speculative build" reading strengthens over the "anchor-customer" one. (Internal-demand strength + the david-zinsner "most severe supply constraints in industry history across leading-edge logic, wafers, memory, substrates" line independently corroborate hbm-cowos-as-binding-bottleneck and memory-shock-to-oem-value-transfer.)
gavin-baker in 2026-05-20-podcast-invest-like-the-best-gavin-baker-watts-and-wafers-invest-like-the-best: **"If Taiwan Semi did what Jensen wanted, I think Nvidia could sell $2 trillion of GPUs in 26 or 27, maybe 2.5 trillion, maybe 3 trillion. But there is a limit where consumers would consume so much they probably would be in an overbuild. So Taiwan Semi. If we don't get a bubble, we need to throw a party for them because they will have single handedly prevented a bubble."** Reframes TSMC's saturation discipline as the *active bubble-prevention mechanism* — not just a passive bottleneck but a deliberate constraint that keeps the cycle from over-shooting demand. Watch TSMC capacity-expansion decisions as the leading indicator of whether the cycle stays disciplined or breaks into a 2000-style buildout.
gavin-baker in 2026-05-20-podcast-invest-like-the-best-gavin-baker-watts-and-wafers-invest-like-the-best: **"The history of markets is I don't know who but one of intel and Samsung. They're not going to stay disciplined. They will break and then at some level that will force everyone else to break."** Adds a falsification leg to the saturation thesis: if Intel or Samsung breaks discipline first and ramps aggressively, it would force TSMC to follow → loss of supply discipline → overbuild risk. Baker's framing: "there's a Goldilocks zone where they expand enough they make it hard for intel or Samsung to really truly emerge as a at scale second source with something well north of 30% market share. And yet they also keep this fundamental constraint on wafers that helps us avoid a bubble." Implication for the us-fab-capacity-bottleneck thesis: a *modest* INTC re-rate is consistent with the saturation chain, but a *full* multi-foundry-equal scenario implies the saturation chain breaks.
gavin-baker in 2026-05-20-podcast-invest-like-the-best-gavin-baker-watts-and-wafers-invest-like-the-best: On Terafab specifically: **"Joint venture to build the world's largest fab here in America. I think they're going to be successful. One they have a partnership with intel which is very important because they're getting access to 50 years of institutional knowledge. That's just nine months, a few quarters 12 months, three to five quarters behind the front. That's an advantage. It's also an advantage that I believe that Terrafab is going to get attention from the A teams, all the Semi cap equipment companies. One big reason Taiwan Semi caught up is ASML and KLA and LAM Research and Applied Materials. They wanted them to catch up. They don't like having a monopsony. The A teams were in Taiwan working Intel made some mistakes and presto. So the A teams will be here because of Elon's reputation in hardware engineering."** Endorses terafab as a credible third-foundry leg specifically because (a) Intel partnership gives 50 years of foundry know-how compressed into the build, only ~9-15 months behind frontier; (b) the WFE A-teams from ASML/KLA/LAM/AMAT will physically move to Texas, mirroring the TSMC-catch-up dynamic of the 2010s. This is the *operational shape* of how Terafab catches up, not just a thesis statement.
5
Picks-and-shovels (ASML, AMAT, KLA, LRCX) capture capex from BOTH TSMC's continued expansion AND Intel's new build. ASML overtook AMAT as #1 WFE in early 2026 on the High-NA cycle gating Intel 14A.
From 2026-05-08-autoresearch-intel-foundry-anchor-customers: "ASML overtook Applied Materials as #1 WFE provider by revenue in early 2026, driven by the High-NA cycle. The EXE:5000-class High-NA systems are priced at ~$380M each and shipping to Intel and TSMC"
From 2026-05-11-autoresearch-picks-shovels-semicap-update-may-2026: "Applied Materials reported record Q1 FY2026 results (revenue $7.01B, EPS $2.38 vs. $2.20 estimate, shares +11%) and guided Q2 to $7.65B. Morgan Stanley models 2027 WFE at $185B+"
From 2026-05-15-autoresearch-amat-q2-fy2026-earnings-results-guidance: **AMAT Q2 FY2026 beat-and-raise (May 14)**: $7.91B actual vs $7.65B guide (+3.4%); Q3 guidance $8.95B ± $500M (13% sequential step-up); CEO Dickerson: AI demand "durable and multi-year," customer conversations extend into 2027-2028; AMAT forecasts 30%+ WFE business growth CY2026 and 2027 as another record year. EPIC Center founding partners: TSMC + Samsung + Micron + SK Hynix — AMAT benefits from all foundry capex simultaneously, validating the picks-and-shovels capture at Step 5.
**cc-wei in 2026-07-16-earnings-tsm-q2-fy2026**, asked how TSMC would react to EMIB-T gaining traction: *"our packaging capacity is so tight that now it's limiting my customers' growth. **We welcome that additional flexibility in the market.** That will help TSMC's front-end wafer business growth, which is a majority part of TSMC's business. The technology looks good, according to the newspaper. **We hope they will be successful, that share some of the loading from TSMC.**"*
**And again, when pressed on whether back-end share loss becomes a front-end gateway** — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: *"The front end's wafer business and the back end's business are two different things... If they are the same, you can expect ASE become the front-end competitor also... since our capacity in the back end is so in shortage mode, the gap is bigger. **I welcome that the competitor offers some of the flexibility to my customer** so that their front-end wafer can be put into the package, and that help TSMC's front-end wafer business. That's our attitude."*
**Wei's moat claim, unchanged** — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "there is no shortcuts... the technology, manufacturing, and customer trust. These three fundamental never change." And on switching costs: "choosing a technology, ramping it up, is not buying a milk from 7-Eleven... You don't like it, you go to another store. No." He also notes Intel's "very strong U.S. government support" while adding "We also got the government support, by the way, although we don't announce it."
What would falsify this
- Step 1: TSMC reports leading-edge utilization below 100% on a quarterly call (signals shortage breaking).
- Step 2: TSMC walks back the 4-year price-hike cadence — would weaken both the TSM-margin tailwind and the alternative-foundry push.
- Step 4: Apple publicly walks back the Intel deal, OR an Intel anchor (AWS, MSFT, Apple, Terafab) names Samsung Taylor as primary instead.
- Step 5: WFE capex growth flattens — would invalidate the picks-and-shovels capture step (low probability given $830B hyperscaler capex).
Contradictions / tensions
- Apple deal remains preliminary — no purchase orders. Risk that the deal stays paper for years.
- 18A yield numbers are KeyBanc-modeled (~7%/mo improvement from ~55% mid-2025). Q1 2026 commentary suggests trajectory holds but yields aren't Intel-disclosed.
- Nvidia $5B should be re-classified as equity/co-design, not foundry production order. The Nvidia foundry test stopped moving forward.
- **China-fab context (does NOT threaten Step 4, but bounds the moat):** SemiAnalysis's STEEL N+3 teardown (2026-06-14-feed-semianalysis-smic-n3-metal-pitch-vs-intel-18a, ⚠ partial) finds smic N+3 reaches a 32.5 nm metal pitch (~10% *tighter* than Intel 18A's 36 nm) but only ~113.4 MTr/mm² logic density (~TSMC N6-class, vs Intel ~238 / TSMC ~313 at the true leading edge) — via DUV SAQP multi-patterning, no EUV, which "increases complexity, reduces process margins, and limits further scaling." So China is *not* a leading-edge alternative that erodes Intel 18A's "US-domiciled leading-edge" pitch — but its trailing-edge capability floor (and the Huawei 3D "LogicFolding" / domestic-EDA pivot) is higher than a pure-EUV-denial reading implies. The export-control moat binds on the *ceiling*, not the current node.
Implications
- Primary beneficiaries: INTC (capacity buyer of last resort with anchor customers), TSM (still earns multi-year price hikes on the existing book), ASML, AMAT, KLAC, LRCX (capex flows from both sides).
- Re-rating logic for INTC: 4 anchor customers + Trump as advocate + Q1 yields ahead of schedule = the 'TSMC alternative' story is no longer a hope.
- TSMC's expansion is partial mitigation that doesn't break the chain — the shortage drags into 2027+ per CEO.
- Samsung as a third leg threatens to split the order flow rather than concentrate it on Intel (see samsung-as-third-foundry-leg).
Companies
Concepts
US fab capacity bottleneck → Intel as the alternative foundryTSMC capacity shortfall and 4-year pricing powerPicks-and-shovels: leading-edge fab buildout
Open questions
none