A supply-landscape judgment built on 199 A-share and HK-listed companies. It answers one question directly: where China’s semiconductor localization is strong and where it is stuck.
China’s listed semiconductor supply is heavily concentrated in design; the real localization constraints sit in key materials and advanced nodes — not in the design segment itself.
① Design firms are 6.1× the combined foundry + OSAT; ② the 8 scarce segments total just 31 firms, fewer than logic SoC alone; ③ 7 of them carry independent barrier and external-dependence evidence — real substitution constraints.
Semiconductors are the core chokepoint of US–China tech competition, and localization (self-reliance) is national strategy. Yet the question of “where strong, where lacking” has long rested on impressionistic descriptions of a few dozen leaders — far too thin to support resource allocation or investment judgment.
This report expands the sample from a few dozen leaders to 199 listed entities, then uses three layers of evidence — entity count → capital barrier → external dependence — to strictly separate two concepts that are often conflated:
The distribution of listed entities — a quantifiable, reproducible structural fact.
A “localization rate” — the report claims no localization rate and does not equate “supply-thin” with “chokepoint.”
Conclusions hold only for listed entities; unlisted startups and overseas entities are outside the sample.
The supply structure visible to capital markets is markedly unbalanced — manufacturing and materials density is far below design. This is a fact, not a conclusion.
The old “leader whitelist” inherently excluded small-caps; only after widening the sample did 51 small-cap entities surface — the value of a full-coverage judgment.
Photoresist, specialty gas, targets, wet chemicals, and CMP — plus foundry, CIS, and sensors — total 31 firms, still fewer than logic SoC alone at 38.
Color key: ■ red = scarce and constrained; ■ amber = scarce but with insufficient evidence to call it a constraint.
This report does not equate “few entities” with “chokepoint.” Instead it uses a three-layer judgment, where each layer requires independent evidence to advance — conclusions are written only as far as the data supports.
Few listed entities. A quantifiable statistical fact, and nothing more — no strategic meaning attached.
On top of thinness, independent evidence of capital, technical, or certification barriers turns it into a “high-barrier scarce segment.”
Add import dependence, local share, customer certification, or technology-gap evidence to classify it as a “real localization constraint.”
The barrier is capital formation — the enormous capex of building fabs and production lines — not R&D spending. None of the 8 scarce segments has higher R&D intensity than design.
The vertical line is the design benchmark (1.0×). Sensors fall below it — the only segment that is both thin and without a capital barrier, so it must not be lumped with the other 7.
Eight scarce segments, checked one by one: 7 hit independent external-dependence evidence and constitute real substitution constraints; sensors are the exception — thin, but with neither a capital barrier nor external-dependence evidence.
CIS’s constraint is “insufficient value-side substitution,” not an overall supply shortage — volume substitution is ahead, value substitution behind.
| Metric | China mainland | TSMC |
|---|---|---|
| Most advanced mass-production node | 12nm / 7nm risk | 3nm / 2nm |
| Sub-7nm revenue share | Hua Hong mostly ≥40nm | ≈70% |
| Core bottleneck | No EUV access | EUV available |
| ASP per wafer | ≈RMB 7,100 | ≈RMB 27,600 |
Mature-node capacity is ample: SMIC ~1.02m wafers/month at 93% utilization; Hua Hong 106%; Nexchip >103%. China’s 8-inch share is ~22% globally.
For targets, CMP, and wet chemicals, the barrier is building production lines — the focus should be effective capacity ramp and yield, not R&D spending alone.
Photoresist and specialty gas carry 2–3-year certification cycles; passing certification locks in the relationship — substitution hinges on certification progress and customer adoption, not current share.
The foundry constraint is in advanced nodes, rooted in upstream EUV lithography limits — watch equipment access and domestic equipment breakthroughs, not total capacity.
Sensor entities number in the single digits, yet their capital intensity is below benchmark with no external-dependence evidence — so they cannot be classified as a constraint. This proves the method can tell apart segments without barriers, rather than promoting every thin segment.
Chinese smartphone CMOS unit share exceeds 55%, but on a revenue basis overseas players still hold over 75% globally. A single “localization rate” cannot capture this “wins on volume” structural gap.
China’s mature nodes are already a major global supply base (8-inch at ~22% globally), with ample supply. The constraint is precisely the external equipment dependence of advanced nodes.
Unlisted startups, non-listed IDMs, and overseas foundries are outside the sample, so conclusions hold only for listed entities and do not claim to represent the entire Chinese semiconductor industry.
Covering design, equipment, materials, foundry, and OSAT across 21 sub-segments, with sub-segment classification auto-verified against 2026 interim revenue composition.
Capital formation is measured as cash spent on fixed assets over revenue, and R&D as R&D expense over revenue (2025 annual-report basis), splitting “high barrier” into two measurable axes.
External evidence is collected per segment and verified against original sources; unverified estimates are never presented as high-precision figures.
Industry totals come from SIA, CSIA, TrendForce, SEMI, and the National Bureau of Statistics; entity dimensions come from company periodic reports.
Scope: this report measures the distribution of listed entities, not a “localization rate”; “supply-thin” is a count fact and does not equal “chokepoint.”
Sample: listed entities only; unlisted startups, non-listed IDMs, and overseas entities are systematically absent, so conclusions cannot be generalized.
Timing: market-cap tiers are a snapshot at retrieval and fluctuate with prices; the RMB 100b / 10b thresholds are researcher-defined, not regulatory.
Local share: mostly third-party estimates, single-source and single-year with wide divergence; uncross-checked figures are not presented at high precision.
Statistical nature: all distributions are cross-sectional snapshots with no time series, so no trend judgment is made.