SMIC

Phase A — Understand the business

Company Overview

SMIC (Semiconductor Manufacturing International Corporation) is China's largest semiconductor foundry — a pure-play contract manufacturer that fabricates chips designed by others. Founded April 2000 in Shanghai by Richard Chang (Zhang Rujing), a TSMC/Worldwide Semiconductor veteran. It ranks #1 among mainland-China foundries and, by Chinese industry accounting, roughly the world's #2–#3 pure-play foundry by sales.

The business model in plain terms: customers (fabless chip designers) hand SMIC a design; SMIC turns silicon wafers into finished chips on its process lines and bills per wafer. Revenue = wafers shipped × blended wafer price. It is a capital-intensity machine: SMIC spent $8.1B of capex in 2025 (record, +10.5% YoY) to add capacity, and depreciation of that fab equipment is the single biggest swing factor in its margins.

Scale (FY2025): revenue $9.33B (RMB 67.3B, +16.5% YoY); net profit attributable RMB 5.04B (~$700M, +36.3% YoY); wafer shipments ~9.7M 8-inch-equivalents; average capacity utilization 93.5% (+8pp YoY).

What it makes: overwhelmingly mature and mid-mature nodes (28nm and above — power management, display drivers, MCUs, CIS, RF, BCD analog) plus a strategically outsized advanced-node franchise (7nm "N+2" and an emerging 5nm "N+3") that no other Chinese foundry can match. By wafer size, 12-inch ≈ 77% of revenue, 8-inch ≈ 23% (Q3 2025). By application, consumer electronics ≈ 47% of Q4 2025 sales.

Key customers: almost entirely Chinese fabless designers. The marquee (and most fraught) is Huawei's HiSilicon — SMIC fabricates the Kirin smartphone SoCs and the Ascend AI accelerators that are China's answer to NVIDIA. HiSilicon revenue reportedly doubled in 2024 on the back of this relationship.

Contract structure: conventional foundry — per-wafer pricing, no take-or-pay disclosed, capacity allocation negotiated. Pricing power is bifurcated: SMIC raised select mature-node prices ~10% in 2026 on AI-driven capacity tightness while simultaneously fighting a domestic mature-node price war.

Supply Chain

Map: equipment + materials → SMIC fabs → fabless customer → end device, with the whole point being that SMIC sits inside a deliberately de-Americanizing chain because it is barred from the best Western tools.

Upstream — process equipment (the chokepoint layer):

• Lithography: ASML is off-limits at the leading edge (no EUV; DUV increasingly restricted). SMIC runs sub-7nm on DUV multipatterning, which is the root of its cost and yield penalty. The only domestic scanner maker, SMEE (Shanghai Micro Electronics Equipment), ships a ~90nm-class ArF tool — 10–15 years / ~10 process generations behind ASML High-NA; litho localization is <5%. This is the binding national constraint.
• Etch: AMEC (Advanced Micro-Fabrication) — FY revenue $1.74B (+36.6%), 5nm etch tools reportedly in validation even on TSMC lines.
• Deposition / furnaces: NAURA — now >60% of the equipment on SMIC's 28nm lines.
• Clean: ACM Research (Shanghai). NAURA + AMEC + ACM = the domestic "Big Three," ~45–50% combined domestic share; overall Chinese fabs now source ~35% of equipment domestically (up from ~25% a year earlier).
• Still import-dependent chokepoints: RF power (~20% domestic), VAT-type precision valves (<10%), quartz (30–40%), precision ceramics (20–30%).

Midstream — SMIC's own fabs: Shanghai, Beijing (SMIC North / SMNC — SMIC bought out the minority for $5.79B in Jan 2026 to consolidate control), Shenzhen, Tianjin.

Downstream — customer → device → the HBM wall: SMIC → HiSilicon → Huawei Ascend 910C → Chinese AI datacenters. Critical single-point dependency for the growth story: SMIC can fab die for >1M Ascend chips/year, but CXMT (China's DRAM champion) can supply only ~2M HBM stacks in 2026 ≈ 250k–300k Ascend-910C packages — so HBM, not SMIC wafers, is the 2026 binding constraint on China's AI-chip output. Separately, the illicitly-stockpiled TSMC die bank that padded Ascend supply is now exhausted — future Ascend depends entirely on SMIC + domestic packaging + CXMT HBMHigh-bandwidth memoryMemory stacked in layers beside a processor so data reaches it faster. The usual bottleneck in AI chips.. A wildcard: SiCarrier (Huawei-spun equipment maker, raised $2.8B) is building Huawei-staffed fabs — a potential future in-house complement/rival to SMIC.

Names present, chokepoints marked — this lens is grounded, not generic.

Competitive Advantages (moats)

SMIC's moat is not the classic foundry moat (process leadership + ecosystem). On the merits it is a structurally disadvantaged manufacturer: no EUV, low yields, industry-lowest margins. Its durable advantages are almost entirely non-market:

1. Sole-source national champion (the real moat). SMIC is the only Chinese entity that can produce 7nm-class logic domestically. Under a hard-tech-sovereignty mandate, that makes it the default recipient of state capital, guaranteed demand (Huawei, the "Big Fund"), and captive customers who cannot go to TSMC for restricted parts. Switching cost for HiSilicon's restricted chips ≈ infinite — there is no alternative.
2. Scale within China. Largest domestic capacity, first mover on every domestic node — a self-reinforcing lead over Hua Hong, Nexchip, and other mainland foundries.
3. Learning curve on DUV multipatterning. SMIC has more real-world reps pushing sub-7nm without EUV than anyone on earth — a genuine (if perverse) accumulated know-how.

Bargaining power: high over domestic customers (they have nowhere else to go for advanced or sanction-safe supply — hence the 2026 mature-node price hikes), low over suppliers of the tools it actually needs (ASML/AMAT/LAM/TEL hold the cards it is barred from). Against global peers, SMIC has near-zero pricing power at the leading edge because its cost structure is uncompetitive — it wins Chinese share on access and policy, not on price/performance. The commercial-layer positioning matrix lists TSMC / Samsung / Intel as the leading-edge foundry set and does not even seat SMIC at that table — an accurate read of where it stands on merit.

Segments

No compiled our figures (empty), so this is web-reconstructed and labeled.

By wafer size (Q3 2025): 12-inch 77% / 8-inch 23% — the mix has been trending toward 12-inch as advanced-node and 12-inch analog capacity ramps.

By application (Q4 2025): consumer electronics ~47% (the largest and fastest-growing bucket, +10% QoQ on stockpiling of small-form-factor devices), with the balance across smartphone, industrial/automotive (a rising automotive-grade BCD platform), and computing/AI.

By geography: heavily China-concentrated and getting more so — ~85% of 2024 sales, ~89% in Q1 2026. The residual US/Eurasia revenue is legacy mature-node work that sanctions steadily erode.

By node: SMIC does not publish a clean node split. Directionally: the vast majority of revenue is 28nm-and-above mature/mid-mature; the strategically critical 7nm/5nm advanced tranche is small in revenue but is the entire equity story. Advanced-node (7nm & below) capacity: ~45k wafers/month end-2025 → ~60k in 2026 → ~80k in 2027, with 7nm capacity doubling in 2026.

Trend & cause: revenue is accelerating (2025 +16.5%, Q2-2026 guided +14–16% QoQ) driven by (a) capacity additions coming online, (b) AI-adjacent demand for power-management and data-transmission chips, and (c) selective price hikes. But earnings are decelerating relative to revenue because depreciation on the capex wave outruns gross-profit growth — the segment mix is shifting toward volume, not margin.

---

Phase B — Measure performance

Earnings Result (latest print: Q1 2026, reported ~May 2026)

• Revenue: $2,505M, +0.7% QoQ, +8.1% YoY.
• Gross margin: 20.1%, +0.9pp QoQ — driven by "steady and rising prices for certain products in advantageous segmented markets," with blended wafer prices +2.5% QoQ offsetting a −0.2% shipment dip.
• Operating profit $248M; net profit attributable $197M (+0.4% YoY — essentially flat).
• Utilization 93.1% (−2.6pp QoQ off Q4's 95.7%) — but management framed it as supply-constrained, not demand-soft: Dr. Zhao said SMIC "could not fully meet demand for BCD, logic, and embedded memory".
• Guidance (Q2 2026): revenue +14–16% QoQ (~$2.86–2.91B), gross margin 20–22% — a sharp acceleration.
• Balance-sheet / capital-allocation flag: management declined to distribute 2025 profits because free cash flowFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. was negative after capex — all cash is going into fab build-out. This is the tell: revenue records are being bought with cash the equity holder never sees.
• Unusual vs. own history: the YoY profit line went from +36% (FY2025) to flat (Q1 2026) even as revenue rose — the depreciation wave has arrived (see Lens 11). Net income attributable growth stalling while revenue accelerates is the single most important datapoint in this print.

Earnings Calls (sentiment trend)

Tracking the last ~4 calls (Q2 2025 → Q1 2026):

• Consistent management focus: capacity expansion ("one new fab per year"), utilization defense, and a repeated insistence that SMIC will "grow faster than the industry/peers".
• Tone shift — from margin defense to demand confidence. Through 2025 (Q2/Q3) the refrain was margin pressure, geopolitical/cyclical headwinds, and a warned "softer Q4." By Q1 2026 the language flipped to supply-can't-keep-up ("could not fully meet demand"), AI tailwinds ("AI boom has increased demand for power-management and data-transmission chips"), and pricing power in "advantageous segmented markets."
• What they started saying: "AI," "power management," "automotive-grade BCD," "price increases." What they keep saying but investors should discount: "grow faster than peers" (true on revenue, false on earnings). What they stopped emphasizing: leading-edge/7nm specifics — SMIC has gone conspicuously quiet on advanced-node detail since the Mate 60 sanctions blowback, a deliberate low-profile posture.
• Depreciation is the honest note in the bullish tone: the co-CEO guided 2026 depreciation +~30% — management itself is flagging the margin drag.

Comps (provenance-critical — multiples are or n/a)

Foundry peer table. Multiples as of ~2026-07 unless noted; sources per cell.

Cells sourced: SMIC; TSMC; GFS; UMC; Hua Hong. 5-yr avg ROE n/a for the peer set (not fabricated).

† The A/H premium is the whole story. SMIC's Shanghai STAR A-shares (688981) trade at roughly 2–3× the Hong Kong H-share price, so "market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth." ranges from ~$73B (everything valued at the H-price) to ~$120B+ (A-shares valued at their own price). The mainland A-share multiple is even more extreme than the ~117x H-share P/E. ‡ One web source cited UMC mkt cap ~$45.7B, which looks overstated vs UMC's revenue scale — left n/a rather than pass through a figure I can't verify.

The finding (falsifiable): SMIC trades at ~67x forward / ~117x trailing earnings and ~30x EV/EBITDA — a premium to every global foundry, including TSMC (29x P/E, 18.6x EV/EBITDA) — while earning the lowest gross margin (~20% vs TSMC ~62%) and a ~2.6% ROE in the peer group. There is no fundamental multiple bridge to that price. SMIC is valued as a strategic/policy option on Chinese semiconductor sovereignty, not as a foundry. GlobalFoundries is the only peer even in the same expensive zip code (40x fwd), and it earns far better returns.

Stock-Price Catalysts (>5% moves, last ~5 years — mostly )

The tape reveals a stock that reacts to geopolitics first, earnings second:

• Dec 2020 — added to the US Entity List: shares fell on the sub-10nm "presumption of denial".
• Aug 30, 2023 — Huawei Mate 60 Pro surprise launch (SMIC 7nm Kirin 9000s inside): chip stocks including SMIC surged ~5% on the "China cracked 7nm" narrative.
• Sept 7, 2023 — sanctions backlash: after US congressmen called for tighter restrictions, SMIC fell 8.3% (Shanghai) / 7.6% (Hong Kong). The two-week round-trip captures the entire SMIC risk-reward: a technology-sovereignty win instantly becomes a sanctions-escalation risk.
• 2024–2026 AI/China-tech re-rating: the stock roughly +78% over the trailing 52 weeks, 52-week range HK$37.00 → HK$93.50 — a ~150% peak-to-trough band driven by China AI-sovereignty enthusiasm (Ascend, DeepSeek-adjacent domestic-compute demand).
• May 2026 — Huawei "chip breakthrough to shorten the gap with TSMC": re-energized the domestic-advanced-node narrative.
• Feb 2026 — FY2025 print / 7nm-yield & depreciation worries: record revenue but margin/expansion anxiety.

Pattern: the market prices SMIC on policy headlines (entity-list actions, Huawei product reveals, export-rule changes) and China-AI-sovereignty sentiment far more than on the P&L. Earnings set the floor; geopolitics sets the swing.

---

Phase C — Judge people & books

Management

Structure: an unusual and historically unstable co-CEO model — Dr. Zhao Haijun (30+ yrs semi ops/tech) and Dr. Liang Mong Song (40+ yrs; ex-TSMC and ex-Samsung; 450+ patents; the technologist credited with SMIC's node progress), both appointed co-CEO/Executive Director in Oct 2017. Chairman lineage runs through Liu Xunfeng, with more recent references to Gao Yonggang — chair turnover is itself a signal.

1. Track record — genuinely delivered the impossible-on-paper: under Zhao/Liang, SMIC produced the first commercial 7nm-class logic without EUV (Kirin 9000s, 2023) and is now piloting 5nm (Kirin 9030 shows N+3 progress). On the core national mandate, this team has executed.
2. Tenure & skin in the game: long-tenured, but insider economic ownership is small relative to the state — the aligned "owner" is Beijing, not management. No our figures on the shelf; personal stakes n/a.
3. Capital allocation — reinvest at all costs: ROE ~2.6%, and management withheld the 2025 dividend to fund capex. Returns-on-capital are being sacrificed for capacity and sovereignty. Judged as a commercial capital allocator, the record is poor; judged as a policy instrument, it is on-mission.
4. Red flags — governance instability is chronic. The Dec 2020 blow-up: Liang Mong Song threatened to resign over the board's appointment of ex-TSMC veteran Chiang Shang-yi as vice-chairman without consulting him ("I deeply felt I was no longer respected and trusted") — shares fell ~9.7%; he stayed; SMIC bought him a $3.4M living unit. Repeated executive/director departures ("management shakeup — yet again," EE Times). A two-technologist ego structure over a state-controlled board is a standing key-man/governance risk.
5. Founder vs. professional-manager archetype: the founder (Richard Chang) was forced out in 2009 after the TSMC trade-secret loss (Lens 10/13). Today's team are professional managers operating under state direction — capable technologists, but not owner-operators with independent strategic latitude. Implication: strategy is set as much in Beijing as in the boardroom.

Forensic Red Flags

No EDGAR filings and empty financial CSVs, so this is a web-grounded forensic read — treat all figures as unaudited-per-public-sources.

Accounting/quality-of-earnings watch-items:

• Depreciation timing is the whole game. SMIC guided 2026 depreciation +~30%. Foundries have wide latitude on useful-life assumptions; any lengthening of equipment useful lives would flatter near-term margins — worth scrutinizing against a P&L that already leans on non-cash timing. Cash flow diverges hard from the revenue record: FCF was negative post-capex in 2025 (hence no dividend), so GAAP-ish "record revenue/profit" masks a business consuming cash.
• State subsidies and government grants materially support Chinese fabs; the extent baked into "other income"/margins is not transparently disclosed — a real quality-of-earnings caveat on every margin print.
• Related-party density is high by design: the customer base, the "Big Fund" shareholder, and state-directed suppliers overlap — a web of state-linked related parties that a Western forensic lens would flag even absent wrongdoing.
• Segment reporting is thin (no clean node/segment disclosure), limiting outside verification.
• Dual A/H listing with a large price gap complicates any single valuation and invites governance-arbitrage optics.

Regulatory findings (required sub-section):

• SEC (EDGAR EFTS — LR + AAER): zero findings. Per regulatory/regulatory-findings.md (fetched 2026-07-10), SMIC has no CIK and is not an SEC filer, so no EDGAR enforcement search is possible.
• Non-SEC / national-security enforcement (the material category here): SMIC was added to the US BIS Entity List in December 2020 citing China's military-civil-fusion doctrine and links to the military-industrial complex; items "uniquely required" to produce ≤10nm carry a presumption of denial. Enforcement is live: Applied Materials committed 56 violations (Nov-2020→Jul-2022) re-exporting ~$126M of US equipment to SMIC and affiliates via South Korea. The Sept 29, 2025 BIS "50% Affiliates Rule" extends SMIC's restrictions to entities it owns ≥50% (targeting the shell-company workaround), with a compliance deadline of Nov 9, 2026 (rule partially suspended for a year).
• Company's own legal-proceedings disclosure (10-K Item 3 equivalent): n/a — no SEC filing; HK/STAR disclosures not on the shelf.
• Founding-era IP litigation (material to the risk profile): TSMC sued SMIC for trade-secret misappropriation (2003, refiled 2006); a 2009 Oakland jury found SMIC liable on 61 of 65 claims, settled for $200M cash + 1.79B SMIC shares + warrants for 695.9M shares to TSMC, and founder Richard Chang resigned. SMIC's technology base carries a documented IP-appropriation origin — directly relevant to Lens 13.

Summary: no securities-fraud findings exist or are searchable, but SMIC sits under an active US national-security enforcement regime and carries a founding-era adjudicated IP-theft judgment — the "red flags" here are geopolitical and provenance-of-technology, not classic accounting fraud. Verified via the regulatory-findings file, web search, and public litigation record as of 2026-07-10.

---

Phase D — Project & stress-test

Forward Projection (FY2026–FY2028 EPS —, arithmetic shown)

Built bottom-up from FY2025 actuals ($9.33B revenue, ~$700M net income attributable, ~20% GM, ~7.5% net margin) and Q1/Q2-2026 guidance. Shares ~8.0B (A+H). No our model logged (unattended watchlist run).

Revenue path: Q1 2026 $2.51B + Q2 guide ~$2.88B (+15% QoQ) → H1 ~$5.4B; assume H2 ~flat-to-up on capacity ramp → FY2026E revenue ~$11.5B (+23%). Extend at ~15% → FY2027E ~$13.2B, FY2028E ~$15.0B.

The margin fork (why revenue growth ≠ EPS growth): 2025 net margin ~7.5%; depreciation +~30% in 2026 is a direct ~2–3pp drag on net margin unless price/mix fully offsets.

. The base case says the most important thing: even with +23% revenue, 2026 EPS is roughly flat because depreciation absorbs the growth. Against a ~67x forward P/E, that is the entire bear thesis in one line. FY27–28E EPS recovers only if the advanced-node (7nm/5nm) revenue share rises faster than depreciation compounds — plausible but unproven, and gated by HBM (Lens 2), not by SMIC's own fab.

Forward consensus EPS: n/a — not cleanly sourced (analyst consensus HK$86.73 12-mo target, high 137 / low 25 — a >5x dispersion that itself signals the market can't agree what this is worth).

Bull vs Bear

Bull case. SMIC is the irreplaceable manufacturing node of Chinese technology sovereignty — a government-guaranteed monopoly on domestic advanced logic with a customer (Huawei) that legally cannot buy elsewhere. As China force-marches toward self-sufficiency, SMIC captures (a) a mature-node volume wave with rising prices as AI tightens capacity, (b) the entire domestic advanced-node ramp (7nm doubling in 2026, 5nm piloting), and (c) an equipment-localization flywheel that steadily de-risks its supply chain. State capital removes the financing constraint that would break a Western foundry. If China's AI-compute buildout is even half as large as the bulls claim, SMIC's wafers are the scarce asset. Contrarian read the market may be missing: the mature-node price war narrative is inverting into a mature-node price hike as AI-driven demand soaks up 12-inch capacity — a margin catalyst hiding under the overcapacity fear.

Bear case (2–3 permanent-impairment risks). (1) The depreciation sink is structural, not cyclical — $8B/yr of capex on cost-disadvantaged DUV tools means the fixed-cost base compounds faster than gross profit; ROE ~2.6% may be the ceiling, not the trough, and revenue records keep producing flat earnings. (2) The growth engine is throttled outside SMIC's control — Ascend volume is capped by CXMT HBM (~250–300k packages in 2026), so SMIC's 7nm-doubling is partly building capacity for demand HBM can't feed. (3) Sanctions convexity — a single BIS action (DUV tool cutoff, the Nov-2026 50%-rule enforcement, a spares/consumables squeeze) can impair the advanced-node franchise overnight; the tape already shows SMIC dropping 8% on a congressional letter. Pre-mortem (18 months out, thesis broke): depreciation compressed net margin below 5%, the mature-node price war reignited as Chinese overcapacity flooded 28nm, HBM starved the Ascend ramp, and a fresh US tool-restriction froze 5nm — leaving a stock at 67x forward earning less than it did in 2025. Multiples verdict: yes, too high on any fundamental basis — only defensible as a policy option.

Devil's Advocate (short-seller)

Dismantling the bull case. What structurally breaks the money machine: SMIC's economics are permanently handicapped — no EUV means every advanced wafer costs more and yields less (7nm yields reported 20–40%), so the more "advanced" its mix becomes, the worse its unit economics get until domestic litho closes a 10–15-year gap that is nowhere close. Revenue concentration: ~89% China, with an outsized, loss-leading dependence on HiSilicon — SMIC reportedly earns thin profit on costly Huawei 7nm precisely because those chips are hard to yield. If Huawei in-sources via SiCarrier fabs (Huawei-staffed, $2.8B raised), SMIC's crown-jewel customer becomes a competitor. Why the moat is weaker than bulls think: it is a policy moat, not a capability moat — remove the sanctions cage and SMIC loses to TSMC/UMC on every commercial axis; the moat only exists because Beijing built the walls, and Beijing can also redirect the Big Fund to a rival. Most dangerous underestimated competitor: not TSMC — it's Hua Hong + Nexchip + other subsidized mainland fabs flooding mature nodes, plus SiCarrier/Huawei vertical integration at the top. Worst capital-allocation / governance: the adjudicated TSMC trade-secret theft in its DNA, chronic co-CEO instability (the $3.4M apartment to keep Liang from quitting), zero dividend, and ROE below its cost of capital. What must hold for today's price: that investors keep paying 67–117x for a ~2.6%-ROE manufacturer purely on sovereignty optionality. If growth disappoints 20–30%: at a 67x forward multiple, a revenue/margin miss re-rates violently — the 52-week chart (HK$37→93) shows the stock can halve. Single permanent-impairment scenario (and plausibility — moderate): a comprehensive US/allied cutoff of DUV tools, spares, and consumables (the ASML-service jugular) that freezes advanced-node output and forces yield regression — the one move that turns the policy option into a stranded asset.

Management Questions (ordered by information value)

1. Excluding government grants and subsidies, what is SMIC's gross and operating margin on a purely commercial basis — and how much of reported profit is state support?
2. What is the actual 7nm/N+2 yield today, and what yield is required for the advanced-node business to be gross-margin positive after depreciation?
3. Given CXMT's HBM ceiling (~250–300k Ascend packages in 2026), what utilization can you realistically achieve on the 7nm capacity you are doubling — are you building for demand that can be memory-fed?
4. Walk through the 2026 depreciation +30% guide: what useful-life assumptions underpin it, and at what revenue level does incremental capex become EPS-accretive rather than dilutiveDilutionIssuing new shares, so each existing share owns a smaller slice of the same company.?
5. If Huawei/SiCarrier bring captive fabs online, does HiSilicon volume stay with SMIC or migrate in-house — what share of your advanced-node revenue is at risk?
6. What is your contingency if BIS cuts off DUV tool servicing, spares, or consumables — how many months of advanced-node production could you sustain?
7. How do you allocate scarce advanced-node capacity between Huawei's Ascend, its Kirin phones, and third-party Chinese AI-chip designers — and who decides?
8. What is the path to 5nm/N+3 volume without EUV, what does it cost per wafer vs 7nm, and when does it reach commercial yield?
9. On the co-CEO structure: how are technology and operational decisions divided, and what governance changes followed the 2020 board dispute?
10. Given negative post-capex FCF, when (if ever) does SMIC intend to return capital, and what leverage/liquidity guardrails bound the fab-per-year plan?
11. How dependent is the domestic equipment roadmap (SMEE litho especially) on milestones outside your control, and what is your realistic litho-localization timeline?
12. What mature-node pricing do you assume for 2026–2027 given both AI-driven tightness and Chinese overcapacity — hikes or a renewed price war?
13. How exposed is the customer base to the Nov-2026 BIS 50% Affiliates Rule enforcement, and what revenue could it touch?
14. What is the automotive/BCD and power-management strategy — is this a durable margin pool or a cyclical bridge?
15. Where do you expect SMIC to be positioned vs TSMC/Samsung on a 5-year horizon, and which single external variable most determines that gap?

---