TSMC

Taiwan Semiconductor Manufacturing Company makes chips that other companies design. It has traded on the Taiwan Stock Exchange as 2330 since 1994 and on the NYSE as an American Depositary Share, ticker TSM, since 1997 — one ADS for five ordinary shares. TEXXR’s own wire copy treats the dual listing as routine, tagging the company “TSMC (2330.TW)” out of habit; the ticker isn’t the story here.

The business itself just posted a record quarter: Q2 2026 revenue up 36% year over year to about $39.45 billion, net income up 77.4% to about $21.9 billion, and chips built on processes of 7 nanometers or smaller — the leading edge — making up 77% of wafer revenue. Full-year capex guidance moved twice in that same release, from $52–56 billion to $60–64 billion, credited in TSMC’s own language to “the AI megatrend”.

None of that is a scoop. TSMC is among the most covered, most modeled companies in tech media, and this page doesn’t try to out-explain wafer pricing or node roadmaps that already get covered well elsewhere. What TEXXR holds instead is the record: every edge the graph has extracted tying TSMC to the rest of the AI-capex buildout, dated — so a claim as broad as “TSMC is the industry’s chokepoint” can be checked against when the coverage actually started making it.

Before 2023, TEXXR’s TSMC record splits into two stories, neither about artificial intelligence. One is Apple’s foundry: a customer relationship the record prices at $2 billion a year as early as 2014, Fab 18 turning out roughly a quintillion Apple-bound transistors every six months by 2023. The other is chip-shortage geopolitics: a $12 billion Arizona fab announced in May 2020, framed explicitly as Washington trying to “cut dependence on Asia”; that investment tripling to $40 billion by December 2022; and Nancy Pelosi meeting TSMC chairman Mark Liu in Taipei that August to discuss the CHIPS and Science Act, on the same trip that rattled markets over Taiwan.

The turn shows up in mid-2023, in TSMC’s own numbers before it shows up in anyone’s framing of them. Bloomberg’s account of TSMC’s Q2 2023 earnings — revenue down in absolute terms — credits “a boom in artificial intelligence” for beating a lowered bar. Two weeks later, Reuters reports TSMC committing $2.9 billion to an advanced packaging plant “driven by AI demand”. By that November, Nvidia’s quarterly revenue passes TSMC’s for the first time in the record — $18.12 billion to $17.28 billion — putting a customer, not the foundry, atop the chipmaker-by-revenue ranking. TEXXR’s edge-arc tracking shows the same turn mechanically: TSMC’s leading co-entities in 2024’s first quarter are Morris Chang, Japan, and smartphone chips; a year later they’re Nvidia, Taiwan, Intel, and the United States. By July 2026, TSMC’s own capex language has caught up with a shift it once merely benefited from.

TSMC’s most recent 500-edge batch in TEXXR’s graph runs financial (32.2%), partnership (15.6%), and launch (13.6%) — a company still mostly reporting numbers and signing deals. The categories worth reading closely are the small ones: competitor edges (9.2%) and regulatory edges (8.6%) barely existed in TSMC’s 2024 record at all.

The quarterly view shows how fast that record has grown and turned over. TSMC ran 36 tracked edges in 2024’s first quarter; by 2026’s first quarter, 133 — nearly a fourfold increase in the density of relationships the graph extracts about the company each quarter. Across those same two quarters, regulatory edges grew 22 percentage points and competitor edges 11, while financial edges fell 17 points as a share of the total. The co-entity network turned over almost completely in the same stretch — a similarity score of 1.00 between the two quarters’ entity sets means essentially none of the names TSMC was tied to in early 2024 recur two years later. That’s the signature of a junction node picking up new spokes every quarter, not one relationship deepening.

Those spokes name names. Apple, TSMC’s oldest large customer, still carried a plain partnership edge as recently as July 2025 (“TSMC became Apple’s principal supplier”). The graph’s most recent Apple edges, from January and May 2026, use different predicates — “supply,” then “competitor”: Apple, per sourced reporting, now has to fight other customers for TSMC capacity, the same foundry it once had close to exclusive claim on, because — as one report put it, TSMC’s resources keep tilting toward AI. The dollar figure behind that shift is the sharpest number in the record: Apple’s annual TSMC spending rose from $2 billion in 2014 to $24 billion in 2025, twelvefold, and that still accounts for only a quarter of TSMC’s total revenue. AMD sits on the other side of the same capacity story: its CTO told the record in February 2026 that AMD is TSMC’s first customer for the 2-nanometer process, the node after the one Apple and Nvidia are currently contesting. Nvidia’s own edges show the graph’s texture up close — eleven near-identical edges recording “Nvidia secures early allocation of TSMC’s N3 process capacity” across two days in March 2026, one underlying claim, from a single podcast interview, refracted through outlet after outlet picking it up.

The rest of the graph shows the industry hedging against the centrality this page is describing. A January 2026 Stratechery piece — its own edge in the graph — argued TSMC’s conservative capital spending was itself producing supply shortfalls, and that hyperscalers “must build up Samsung or Intel as rivals to TSMC”. A year earlier, Reuters reported TSMC had pitched Nvidia, AMD, Broadcom, and Qualcomm on taking stakes in a joint venture to run Intel’s foundry — TSMC’s most credible rival, floated as a project TSMC itself might help fund. Both edges are still tagged “rumored.” TSMC’s upstream side runs the same dynamic in reverse: as of mid-July 2026, the graph records TSMC resisting a planned price increase from ASML, whose EUV lithography systems are the one piece of equipment no advanced fab can do without.

The China-Taiwan risk thread runs through the same graph, not around it. A January 2026 Wall Street Journal piece, logged in the record, describes TSMC as having “provided a ‘Silicon Shield’” for Taiwan for decades — the same month Taiwan committed more than $250 billion in US investment as part of a broader trade deal, with TSMC’s own US pledge growing toward $265 billion across four planned fabs by that July. On export controls, the graph shows TSMC telling Chinese customers in November 2024 that it would stop manufacturing AI chips at 7-nanometer nodes or smaller for them, to align with US restrictions — one edge sitting, unremarked, next to that quarter’s otherwise ordinary run of earnings and capacity news.

Positioning: how much of TSM is sold short, on the record brokers file twice a month. It lags by design — read it as where the crowd stood, not where it stands.

Signals in the coverage record worth tracking, not trade ideas:

• The competitor-predicate share. It ran 9.2% in the latest batch against roughly zero as recently as 2024. Whether it keeps climbing as more customers get flagged competing for the same capacity, rather than partnering for it, is the graph’s cleanest read on the capacity crunch.
• The ASML price-resistance edge, still rumored as of July 2026. A flip to confirmed, or a new financial or controversy edge attached, is the thing to watch.
• The $265 billion US pledge. Whether it gets confirmed fab by fab in later coverage, or stays a rumored total attached to one trade-deal headline.
• The Samsung/Intel-as-rival threads — the Stratechery argument, the Intel-foundry joint-venture pitch. Whether either produces a confirmed edge, or both stay rumored and drop out of the record, is the cleanest test of whether TSMC’s position gets diluted or entrenched.
• Regulatory-edge volume. It grew faster than any other predicate between 2024 and 2026. Whether that keeps pace with the capex increases, or the record shifts back toward ordinary earnings and launch edges, says something about whether TSMC’s story is settling or still reorganizing.

TSMC lives or dies by the same constraint as these names — small equal-weighted groups, one constraint each. Not investment products: see all categories.