SMIC: Strategic Scarcity Is Proven, but Has the Capital Cycle Started Paying Its Way?

Revenue has grown rapidly. From 2021 to 2025 it increased from US$5.44bn to US$9.33bn, a compound rate of about 14.4%. Yet the cash burden grew faster. Capital expenditure on property, plant and equipment increased from about US$4.12bn in 2021 to US$8.40bn in 2025. In 2025 alone, capex equaled roughly 90% of revenue. Operating cash flow was US$3.19bn, leaving conventional free cash flow about negative US$5.21bn. SMIC has now produced negative conventional free cash flow in every year from 2021 through 2025.

Accounting earnings understate cash generation before capex but overstate the cash economically available to owners after keeping the manufacturing base competitive. Five-year cumulative operating cash flow was about 2.65 times cumulative net income, largely because depreciation is huge. But subtracting capex reverses the picture. Depreciation and amortization rose from US$1.87bn in 2021 to US$3.81bn in 2025. Management has indicated another large depreciation increase as new fabs enter service.

The moat and the return on that moat must be kept apart. SMIC has real advantages: China’s largest installed foundry footprint, customer qualifications accumulated over years, process breadth, unusually high utilization, access to domestic policy capital and a strategic position that becomes more valuable when Chinese chip designers have fewer foreign manufacturing options. The company’s largest customer represented only 8% of 2025 revenue and its five largest customers 35.8%, so the revenue base is not hostage to a single disclosed customer.

But the frontier is constrained. U.S. export controls structurally restrict SMIC’s access to advanced equipment, and independent teardown evidence suggests those restrictions have slowed cost-effective progression beyond 7nm-class processes. The same sanctions also create the domestic-substitution demand that fills SMIC’s fabs. On the present evidence, I think the demand benefit dominates over the next 12–24 months, while the equipment constraint becomes more important over three to five years because multi-patterning, lower yields and an inferior equipment set raise the amount of capital required for each incremental technology step.

That creates an unusual competitive position. Against Hua Hong, SMIC has substantially more scale, broader process coverage and materially better margins. Against UMC and GlobalFoundries, however, SMIC runs higher utilization but still earns lower gross margins. Tower Semiconductor occupies higher-value specialty niches and has recently produced sharply higher profitability, albeit at a very demanding market valuation. SMIC is the strongest strategic asset among mainland Chinese foundries, not the strongest economic foundry in the mature-node peer set.

The A/H market gives a stark illustration of how much “strategic value” can differ from economic value. The Shanghai line closed at CNY128.50 on August 7; at CNY1 = HK$1.1618, that is approximately HK$149.29 per share. Against the Hong Kong close of HK$66.90, the A share trades at about a 123% premium. The securities rank economically pari passu, so the enormous gap is a market-segmentation price, reflecting different investor pools, domestic policy enthusiasm, capital-account frictions and differing scarcity of semiconductor exposure rather than two different claims on SMIC’s operating assets.

That premium is analytically important but must not contaminate the valuation. Every valuation range later in this report applies solely to 0981.HK. Averaging the A and H prices would convert a capital-market anomaly into a false estimate of intrinsic value.

The refresh also changes how the previous house work should be read. The supplied June 5 report used HK$79.55 as its anchor. A verifiable daily market history, however, shows the actual June 5 close at HK$75.65 after a 7.2% drop that day. I retain HK$79.55 only as the prior report’s stated anchor and use HK$75.65 for market-to-market attribution. From the house anchor to HK$66.90, the decline is 15.9%; from the verified June 5 close it is 11.6%.

The previous bull case was early operationally, but too optimistic as a price case. Q1 beat its margin guidance and Q2 guidance became materially stronger, so the operating thesis has not been disproved. Yet the old HK$82–95 bull range assumed the market would continue paying an unusually high policy/AI multiple before SMIC proved that new capacity could earn acceptable returns. The stock instead de-rated with the semiconductor complex. July saw repeated global semiconductor selloffs as investors questioned elevated AI valuations and financing expectations. With no Q2 earnings disappointment yet available to explain SMIC’s fall, multiple compression and sector risk appetite are the cleaner explanation than an earnings collapse.

My qualitative portrait, then, is “company in transition.” Revenue growth and domestic strategic scarcity are real. So are the utilization rate and near-term pricing tailwinds. The unresolved question is whether a business spending US$8bn-plus a year to generate less than US$1bn of annual net income can eventually lift returns on capital faster than depreciation and the next expansion cycle consume them. That question, not whether China needs more domestic semiconductor capacity, is the central investment argument.

Vertical history, financial review, and capital-market narrative

SMIC was created at the beginning of the 2000s to solve a strategic and commercial gap: China was becoming a major electronics manufacturing base while lacking a globally credible pure-play wafer foundry. Richard Chang, an experienced semiconductor manufacturing executive, built the company around the Taiwanese foundry model: manufacture chips designed by other companies rather than compete with customers through a vertically integrated product portfolio. Shanghai provided industrial policy support, infrastructure and proximity to a rapidly expanding electronics ecosystem. The operating model remains recognizable today, although the geopolitical meaning of domestic manufacturing has changed enormously.

The 2004 Hong Kong/NYSE IPO was intended to finance scale at extraordinary speed. The U.S. prospectus priced ADSs at US$17.50, equivalent to roughly HK$2.73 per ordinary share at the prospectus exchange rate; Hong Kong market records generally cite a final local offer price around HK$2.69. The offering raised roughly US$1.6bn. Investors were being asked to finance a Chinese challenger to an industry whose economics rewarded scale, process execution and relentless capital spending.

The first stage, from founding through 2009, proved that SMIC could build fabs quickly but also exposed the cost of technological dependence. Litigation with TSMC over trade secrets and intellectual property culminated in a 2009 settlement under which SMIC agreed to pay US$200m and issue an equity stake to TSMC; founder Richard Chang resigned. The settlement mattered far beyond the cash payment. It interrupted management continuity and illustrated how difficult it was for a new foundry to acquire process know-how without running into the incumbent’s intellectual-property perimeter.

The second stage, during much of the 2010s, was less glamorous and more important. SMIC rebuilt operating credibility, widened its mature-process offering and moved gradually into smaller geometries. The company ceased to be primarily a startup-financing story and became an industrial-scale manufacturing story. Customers had more reasons to qualify SMIC for production that did not require the world’s most advanced process, while China’s electronics sector provided a large local demand base.

A third stage began as semiconductor self-sufficiency became explicit national industrial policy. SMIC voluntarily delisted its thinly traded NYSE ADSs in 2019, citing low trading volume and administrative costs, while retaining Hong Kong as its principal international trading venue. In 2020 it listed on Shanghai’s STAR Market and ultimately raised roughly RMB53bn. The transaction gave SMIC access to a domestic capital pool willing to value semiconductor manufacturing partly as strategic infrastructure.

The same year created the opposite force. The U.S. placed SMIC and related entities on the Entity List, materially restricting access to certain U.S.-origin semiconductor manufacturing equipment and technology. Restrictions on the most advanced manufacturing equipment became a permanent planning variable. This was the defining turn in SMIC’s modern history: the company gained a captive strategic demand pool in China while losing unrestricted access to the best global equipment.

The fourth stage, from 2021 through 2024, became a capital race. Revenue rose sharply in the pandemic shortage, gross margins expanded, then the semiconductor inventory correction exposed the fixed-cost burden. SMIC kept investing through the downturn rather than maximize free cash flow. In parallel, third-party teardown work showed that the company had reached 7nm-class production without EUV, an important technical result but one achieved through a manufacturing path that is likely more expensive and yield-sensitive than EUV-based production at the same nominal class.

The fifth stage, which is still unfolding in 2025–2026, is the test of that investment. Utilization has climbed toward the mid-90s, 12-inch capacity has become the core of revenue, domestic customer share has risen and management is attempting targeted pricing increases. At the same time, each new fab now feeds a rapidly growing depreciation charge. The operational question has shifted from “can SMIC build capacity?” to “can it earn enough from the capacity after depreciation and maintenance capital?”

The financial record captures the turn better than any strategic slogan.

Sources: SMIC 2025 annual report and Q1 2026 quarterly report. Conventional free cash flow is operating cash flow less PP&E acquisition, calculated from reported figures.

The business reason for the gross-margin path is straightforward. The 2021–22 shortage gave foundries pricing and utilization leverage before the 2023 inventory correction hit. Since then, SMIC has restored volumes and utilization, but the cost base has changed. Depreciation and amortization increased from US$1.87bn in 2021 to US$3.81bn in 2025, equal to about 41% of 2025 revenue. EBITDA reached US$5.26bn in 2025, yet operating profit was only US$1.11bn. The difference is the economic weight of the fab base coming through the income statement.

The balance sheet is healthier than the free-cash-flow line alone suggests. At the end of 2025, SMIC reported US$12.60bn of debt against a broad pool of cash, bank deposits and liquid financial assets, leaving only about US$660m of reported net debt. Q1 2026 similarly showed roughly US$13.85bn of cash-on-hand-type resources against US$14.51bn of debt. The funding model includes operating cash, bank lending, equity issues and capital contributions from minority shareholders. Financial distress is therefore not the central risk today; low returns on enormous incremental capital are.

That distinction matters for ROIC. A rough 2025 pre-tax operating-return proxy, operating profit divided by year-end equity plus net debt, is only about 3%. It is not a textbook ROIC because SMIC has large construction-in-progress balances, non-controlling interests and government support, but it captures the economic order of magnitude: the current asset base is generating low-single-digit operating returns. Government funding is also material relative to earnings; the 2025 report classified US$222.6m of government funding within non-recurring items, while broader government-related operating income was larger.

Management has made the capital-allocation priority explicit. SMIC paid no cumulative cash dividend and made no cancelled-share repurchases over the last three financial years. Its 2025 annual report says 2026 capital spending is expected to exceed 20% of latest audited net assets and that retained earnings are being reserved for capacity expansion and development of the core business. This is internally coherent. It also means shareholders should value the company as a reinvestment vehicle, because there is no distribution yield to compensate for poor incremental returns.

Governance has institutionalized rather than founder-led characteristics. Liu Xunfeng is chairman and executive director. Zhao Haijun and Liang Mong Song serve as co-chief executives; Zhao brings more than three decades of semiconductor operations experience, while Liang has more than four decades in the industry and has been central to SMIC’s technology push. Wu Junfeng is the senior executive responsible for finance. The chairman and co-CEO roles are separated.

The current capital structure also reflects consolidation of strategic assets. The July monthly return shows 6.014bn Hong Kong shares and 2.547bn Shanghai A shares outstanding, or about 8.561bn ordinary shares in total. Those two pools carry the same underlying economic rights even though their market prices differ radically.

The share-price history has become a history of changing labels. SMIC traded for years as a lagging, capital-intensive foundry. It was re-rated after the STAR listing and U.S. sanctions into a strategic self-sufficiency asset. The market later attached an AI-adjacent narrative as mature-node utilization recovered. Public price histories show gains of roughly 80% in 2024 and 127% in 2025 before 2026’s pullback. The current HK$66.90 remains far above the pre-re-rating levels despite being about 28% below the 52-week high around HK$93.50.

This history tells investors what has been proven and what has not. SMIC has proven an ability to raise capital, build fabs, operate them at high utilization and keep advancing technologically under restrictions. It has not yet proven that the post-2020 capital cycle can produce a through-cycle return on invested capital commensurate with the cost and geopolitical risk of the assets.

Business model, industry cycle, and horizontal peers

SMIC’s economic machine begins with installed wafer capacity. The foundry incurs enormous fixed costs before a wafer is shipped: fabrication facilities, lithography, deposition, etch, metrology, clean rooms, utilities, process-development teams and depreciation. Once that capacity exists, utilization becomes one of the most powerful variables in the income statement. A wafer sold into an underutilized fab carries a disproportionate share of fixed costs; a wafer sold when a fab is near full utilization can carry much better incremental economics.

So Q1’s 93.1% utilization matters more than the headline 11.5% revenue growth. Monthly capacity had reached 1.078m eight-inch-equivalent wafers by the end of Q1, up from 1.059m in Q4. Shipments increased 9.5% year on year even though they slipped 0.2% sequentially. Capacity is still growing while fabs remain heavily utilized.

The revenue mix also explains why “AI foundry” is an imprecise label. Nearly half of Q1 revenue came from consumer electronics. AI can tighten the surrounding semiconductor ecosystem, especially power management and connectivity, without SMIC becoming a direct analogue of TSMC’s GPU and accelerator business. That makes TSMC a useful technology and cycle reference but a poor direct valuation peer. TSMC’s leading-edge node economics, EUV access and customer mix are structurally different.

The mature-node pricing engine looks real but selective. TrendForce’s April 2026 work described foundries attempting increases of up to around 10% on some mature-node quotations, helped by reductions in 8-inch capacity elsewhere and AI-related power-management demand. The same analysis said broad-based price increases were unlikely because inventories remained high and mainland Chinese competition continued to add supply. For SMIC, the implication is a favorable 2026 price/mix environment rather than permanent industry-wide pricing power.

The advanced-node engine has a different economic character. Domestic fabless companies have strategic reasons to qualify SMIC even if a foreign foundry could theoretically offer a more efficient process. Export controls make that customer stickiness stronger. But repeated DUV patterning can turn a technically functioning 7nm-class process into an expensive manufacturing proposition. The relevant moat is “usable domestic advanced capacity under constraint,” not global process leadership. Teardowns establish technical capability; they do not establish attractive gross margin.

The main real moats are scale, customer qualification, capital access and domestic scarcity. Scale matters because process learning and fab utilization improve with volume. Qualification matters because customers do not casually move an automotive, MCU or power-management design between foundries after process validation. Few companies can fund US$8bn-plus annual investment through a downturn; that is the capital-access moat. Domestic scarcity matters because geopolitical constraints limit Chinese customers’ alternatives.

The weaker moat is advanced equipment access. U.S. Entity List restrictions remain structural, and Taiwan added SMIC to its strategic high-tech export-control framework in 2025. Washington has continued considering tighter constraints on equipment and servicing, while 2026 policy discussions have kept advanced semiconductor controls live. At the same time, domestic Chinese equipment is improving: Reuters reported in August that Samsung and SK Hynix were evaluating equipment from AMEC for Chinese fabs, illustrating that local tool suppliers are gradually broadening their addressable process steps. Lithography and some metrology remain much harder gaps to fill.

This gives export control its two-sided character. A tighter rule hurts SMIC’s equipment flexibility and advanced-node yield, but also reduces foreign competition for Chinese customers. I expect the demand-side benefit to dominate the next year because SMIC still has usable capacity and process platforms to sell. Over a five-year horizon, equipment access becomes the heavier variable because technology progression gets exponentially more difficult as the node shrinks.

The mature-node peer numbers reveal where SMIC sits economically.

† Peer guidance definitions are not perfectly comparable.

Sources: company filings/releases. SMIC and Hua Hong had not reported Q2 as of August 8; UMC, GlobalFoundries and Tower had.

Hua Hong has become the high-utilization Chinese specialty foundry. Its Q1 utilization was 99.7%, and 12-inch revenue was growing quickly, but gross margin remained only 13%. That makes Hua Hong the most direct proof that a full fab is not automatically a high-return fab. New-capacity depreciation and ramp costs can absorb the benefit. Its Q2 guidance of US$690–700m revenue and 14–16% gross margin implies improvement, but still leaves profitability well below SMIC.

UMC sets the mature-node discipline benchmark. Its 22/28nm business is now a large part of revenue, and Q2 utilization improved to 85% while gross margin reached 32.5%. Customers choose UMC for a mature, well-qualified specialty and logic platform with less geopolitical equipment risk than SMIC. The contrast is uncomfortable for SMIC: UMC is operating at lower utilization yet generating a gross margin more than 12 percentage points higher. Part of that gap is mix and accounting; much is capital efficiency.

Then there is GlobalFoundries, the Western strategic-specialty foundry. Its business is concentrated in differentiated rather than frontier logic, with U.S. and European government support and long-lived customer programs. Q2 revenue was US$1.786bn and IFRS/GAAP gross margin 28.3%, materially above SMIC’s current level. Its strategic premium resembles SMIC’s in one respect: both are valued partly because customers and governments want geographically secure capacity. GF converts that position into better current margin economics.

Tower plays the high-value specialty niche. Its silicon photonics, SiGe, RF and analog platforms expose it directly to data-center optical demand. Q2 2026 revenue rose 24% year on year to a record US$460m and gross profit rose 72% to US$138m, producing a 30% gross margin. Investors have re-rated Tower sharply on AI optical-connectivity expectations. Its resulting valuation is extreme enough that it is more useful as evidence of the market’s willingness to pay for differentiated specialty growth than as a sensible absolute valuation anchor for SMIC.

Valuation confirms that mature-node peers are no longer uniformly “cheap cyclicals.” GlobalFoundries trades around 12–13 times trailing EV/EBITDA. Tower’s valuation screens at roughly 15 times sales on some current datasets and has a triple-digit trailing earnings multiple. Hua Hong screens around 12 times trailing sales after an extraordinary domestic-semiconductor re-rating. By comparison, 0981.HK’s H-price-implied total equity value is about US$73.1bn; adding 2025 net debt gives roughly US$73.8bn of enterprise value, or about 14 times 2025 EBITDA and around 7.6 times estimated trailing revenue. SMIC is cheaper than the most speculative specialty names but more expensive than the economic quality of its cash flow would suggest.

That is why I do not benchmark SMIC’s fair value to Hua Hong’s current multiple. A peer can be more expensive because the peer is overvalued.

The A/H gap reinforces this point. At August 7 closes, 688981.SH at CNY128.50 converts to approximately HK$149.29 at CNY/HKD 1.1618. That is about 2.23 times the HK$66.90 H-share price, a 123% A-share premium. SMIC’s July return records 2.547bn A shares and 6.014bn H shares. Applying their respective market prices gives roughly HK$380bn of market value to the smaller A-share pool and HK$402bn to the much larger H-share pool.

The A/H premium is a market-structure premium, not an operating-fundamentals premium. Domestic investors have a smaller pool of pure semiconductor manufacturing assets, the A share sits directly inside the self-sufficiency trade, capital-market access is segmented and the shares are not freely arbitraged into one price. Because the two lines ultimately participate in the same company economics, I give the Shanghai valuation no weight in determining the fair-value bands for 0981.HK.

Current fundamentals and price-move attribution

The last five reported quarters show a business that came through the 2025 trough without a revenue relapse.

Sources: SMIC annual and quarterly disclosures; Q3 utilization from contemporaneous reporting.

Revenue growth was not primarily a single-customer event. In 2025 wafer shipments increased about 21% while reported average selling price per standard eight-inch-equivalent wafer slipped from about US$933 to US$907. The company entered 2026 after a volume-led recovery, not after broad pricing inflation. Q1 then showed a sequential shipment decline of only 0.2%, while management and call reporting indicated better price/mix. That is what makes Q2’s guided double-digit revenue step significant: it would mark a transition from pure utilization recovery toward price/mix plus volume.

Q1’s cost line is the warning embedded in the beat. Operating expenses rose 42.5% sequentially and 30.3% year on year to US$256m, partly because R&D increased. Total Q1 depreciation and amortization was approximately US$1.09bn, up about 26% year on year. Gross margin held at 20.1% because utilization and pricing absorbed those costs, but the bar keeps rising as more capacity enters depreciation.

Management’s demand signal is nonetheless stronger than it was at the previous refresh. Q2 revenue guidance of +14–16% sequentially implies US$2.856–2.906bn, and the 20–22% margin guide suggests the company expects the extra revenue to absorb new depreciation rather than simply fill capacity at low incremental margin. Reuters reported that management linked the demand to AI-related capacity tightness, customers shifting some orders toward Chinese foundries and targeted price adjustments.

The central uncertainty is durability. A targeted price increase on a constrained process can last several quarters. It does not prove industry-wide scarcity. TrendForce’s mature-node work explicitly warns that Chinese capacity growth and inventory overhang remain constraints on generalized price hikes. I model the Q2 step-up as genuine near-term earnings improvement but do not extrapolate 15% sequential growth or high-single-digit pricing indefinitely.

The share-price decline since the last house report needs to be decomposed with similar discipline. The first complication is the anchor. The prior house report supplied by the operator records HK$79.55 on June 5. Public daily market records show June 5 opened at HK$80.80 but closed at HK$75.65 after a 7.18% fall. I treat HK$79.55 as the prior research anchor and HK$75.65 as the verified trading close. At HK$66.90, the share is down 15.9% from the former and 11.6% from the latter.

Earnings deterioration explains little of that move. There has been no Q2 result yet, no post-Q1 profit warning and no withdrawal of the strong Q2 guidance. Q1 itself reached the top of management’s margin range. The evidence points away from realized earnings cuts as the primary cause.

Multiple compression explains more. Global semiconductor shares experienced several sharp July de-risking episodes as investors questioned the financing and valuation of AI infrastructure. Reuters reported a third consecutive decline in global chip names on July 17 as investors reduced AI exposures, and another broad Asian semiconductor selloff later in the month over lofty valuations, financing concerns and Chinese competition. SMIC had entered that period with a large policy/AI premium after huge gains in 2024–25, making its multiple unusually exposed to changing risk appetite.

Company-specific factors account for another, smaller piece. SMIC is still issuing shares through employee plans and completed changes in its A-share capital structure associated with bringing strategic manufacturing interests more fully inside the listed group. More important economically, investors are increasingly confronting the depreciation bill from the 2023–25 capex surge. That is a fundamental reason for some multiple de-rating even when revenue estimates rise.

Policy risk contributed to the risk premium rather than producing a new SMIC-specific earnings shock. The company remains on U.S. restricted lists; 2026 U.S. proposals have sought tighter equipment and servicing limits, and Commerce officials have continued signaling additional semiconductor measures. China’s domestic equipment base is improving, including reported progress in locally produced immersion DUV equipment and growing adoption of domestic etch/deposition tools, but replacement is incomplete.

My attribution is therefore primarily a valuation event. I would describe the June–August decline as roughly two-thirds multiple and semiconductor-risk-premium compression, with the remainder split between capex/depreciation recognition and recurring geopolitical risk. The earnings-revision contribution appears small to mildly positive because the latest formal operating update was stronger, not weaker. That is an analytical attribution rather than a statistical factor decomposition, but it fits the observable evidence.

This is also the answer to whether the market is “wrong.” The share decline does not tell us that Q2 demand weakened. It tells us the market became less willing to pre-pay for future semiconductor scarcity. I agree with that re-pricing. SMIC deserves a strategic premium to a commodity mature-node foundry, but the premium should be earned against a cash-flow denominator, not justified solely by China’s desire for semiconductor independence.

Valuation, risks, catalysts, and tracking dashboard

The valuation has to start with cash-flow passthrough rather than a headline earnings multiple.

Across 2021–25, cumulative operating cash flow was about US$18.09bn against cumulative net income of about US$6.82bn, giving an OCF/net-income ratio of roughly 2.65 times. That superficially looks excellent. The reason is largely non-cash depreciation: cumulative D&A over the same five years was enormous and rose every year.

Capex changes the conclusion. PP&E acquisitions totaled almost US$34bn over 2021–25 and exceeded operating cash flow in each of those years. Conventional free cash flow was negative US$1.11bn in 2021, negative US$0.82bn in 2022, negative US$4.27bn in 2023, negative US$4.49bn in 2024 and negative US$5.21bn in 2025.

SMIC says most recent capex is being used for capacity expansion, so deducting the entire US$8.40bn from earnings would understate normalized owner economics. Maintenance capex is not separately disclosed. I estimate it at roughly US$3.5–4.2bn for 2025, centered near the US$3.81bn D&A charge. That makes 2025 owner earnings, defined here as operating cash flow minus estimated maintenance capex, roughly negative US$0.3bn to negative US$1.0bn. The range is necessarily an estimate.

At HK$66.90, the H-price-implied total equity value is approximately US$73.1bn. On 2025 attributable profit of US$685m, that is over 100 times earnings. On the owner-earnings estimate, the yield is around zero or negative, so an owner-earnings P/E is not meaningful. The gap between accounting P/E and owner-earnings economics is much greater than 30%, so in the scenarios below I default to normalized EV/EBITDA and asset/cash-return economics rather than headline P/E.

Historical valuation offers only limited comfort. The stock is well below its 52-week high, but its current H-line valuation still implies roughly 14 times 2025 EBITDA and around 7.6 times trailing sales. Those are premium multiples for a foundry with a 20% gross margin and negative owner earnings. The center of valuation has structurally shifted upward since 2020 because domestic manufacturing now has strategic scarcity value. I accept part of that shift; I do not assume the Shanghai-line policy premium will migrate completely into Hong Kong.

For the absolute valuation I use a one-year-forward framework based on normalized 2027 revenue and EBITDA. This allows Q2’s current acceleration to enter the estimates without treating one quarter as a permanent run rate. The model uses the current 8.561bn share count and US$1 = HK$7.834.

These scenarios are valuation analysis within a research framework, not investment advice. Inputs use company-reported revenue/capacity data, current share count and current FX; multiples are my assumptions informed by SMIC’s current roughly 14-times trailing EBITDA valuation and GlobalFoundries’ roughly 12–13-times EV/EBITDA reference point.

The conservative case assumes Q2’s strength largely represents a cyclical tightness phase and that China mature-node expansion caps pricing thereafter. The 10-times multiple still gives SMIC credit for domestic strategic scarcity; a commodity foundry with structurally low returns could deserve less. The base case assumes 2026–27 utilization remains around 90% or higher, gross margin stabilizes near or just above 20% despite rising depreciation, and domestic substitution sustains moderate growth. The optimistic case requires both pricing and mix to improve while new capacity reaches acceptable yields quickly. That is possible, but it asks the investor to accept a high multiple before owner earnings normalize.

The expectation gap at the August 13 print concentrates in three numbers. Revenue needs to land near or above the midpoint of the +14–16% guidance. Gross margin needs to be at least around 21% to show that incremental revenue is absorbing depreciation. Utilization at the next disclosure needs to remain in the 90s. A revenue beat with a 19% margin would be much less valuable than the headline suggests because it would imply new capacity is buying sales without generating sufficient incremental return.

Margin-of-safety verdict: none. The current HK$66.90 is about 23% above my HK$54 conservative value. The base scenario’s most fragile assumption is the 12-times EBITDA multiple: cutting that assumption to 70%, or 8.4 times, drops base fair value to about HK$51. That is below the current price even before assuming an earnings miss.

The flat-earnings test is even harsher. With no dividend and no buyback, three years of flat economic earnings plus an unchanged multiple produces approximately zero annualized shareholder return before any eventual distribution. That does not clear a positive sovereign bond yield. Under the framework specified by the research brief, there is no margin of safety at this buy price. SMIC can still compound operating capacity during those years; unless returns on that capacity rise, capacity growth is not the same thing as shareholder compounding.

The primary permanent-loss risks are concrete.

Export-control tightening has high probability and high potential impact because restrictions are already structural, rather than hypothetical. The indicator is a new BIS rule that restricts additional classes of DUV, metrology, etch/deposition tools, spare parts or servicing. The transmission path runs from slower equipment delivery to lower advanced-node yield, higher capital cost per wafer and eventually a lower valuation multiple. The offset is stronger domestic customer captivity.

Mature-node overcapacity is medium-to-high probability with high economic impact. SMIC and Hua Hong are both adding 12-inch capacity while other Chinese projects are doing the same. The observable warning would be utilization falling through 85% while industry price increases fail to stick. That combination would push gross margin back into the mid-to-high teens even while depreciation keeps rising. TrendForce’s current caution on broad price hikes is the relevant early signal.

Capital-return erosion has high probability but a slower transmission path. Capex exceeded US$8bn in 2025 and free cash flow was deeply negative. If revenue continues growing but a rough operating return on capital remains below 5% through 2027, the market can rationally lower the multiple because the company would be using ever more capital to produce each dollar of sustainable owner earnings.

Advanced-node economics carry medium probability and high impact. The technology itself exists at 7nm-class and probably beyond according to third-party teardowns. Yield and cost are undisclosed. A design win produced through repeated DUV patterning at poor yield could add strategic prestige while destroying economic value. The indicator to watch is any credible disclosure on advanced-node wafer volume, yield, customer concentration or margin rather than another teardown establishing that a chip can be made.

Valuation compression remains medium probability. The H share is much cheaper than the A share, but “cheaper than a 123% premium” is not an absolute valuation argument. A retreat from roughly 14-times trailing EBITDA toward 9–10 times would overwhelm several years of moderate earnings growth.

Near-term positive catalysts are the August 13 Q2 print reaching or beating the guidance midpoint, gross margin above 21%, utilization returning toward 95%, evidence that targeted mature-node price increases stick into Q3, and domestic tool progress that lowers the risk of equipment bottlenecks. A benign U.S. policy interval would also reduce the geopolitical discount, although that should never be a base-case assumption.

Negative catalysts are a Q2 margin result below 20%, Q3 guidance implying that Q2 represented order pulling rather than sustained demand, utilization below 90%, a larger-than-expected depreciation increase, another US$8bn-plus capex year without better operating returns, or new equipment/service restrictions. The most damaging print would be high revenue growth accompanied by weak margin because that would directly challenge the economics of the expansion.

Sources: SMIC filings, current market data and valuation calculations described above.

Gross margin and utilization should be tracked together. A margin decline at stable 95% utilization means pricing, mix or depreciation is deteriorating. A margin decline alongside utilization below 85% means the cycle itself is turning. Capex and D&A should be tracked against revenue because a foundry can hide poor incremental economics behind rapid top-line growth for several years before depreciation fully catches up.

The A/H premium is useful as a sentiment indicator, not a valuation input. A widening premium with unchanged operating estimates indicates increasing domestic policy enthusiasm. A rapid narrowing could mean either H-share catch-up or an A-share thematic unwind; the direction of both prices has to be examined before drawing a conclusion.

Cross-synthesis, final conclusion, research uncertainties, and sources

Vertically, SMIC has proven something difficult. It survived a damaging intellectual-property dispute, founder departure, semiconductor downturns, the loss of unrestricted access to advanced U.S. manufacturing technology and repeated capital cycles. It expanded from a new Chinese foundry into a manufacturing network capable of more than one million eight-inch-equivalent wafers of monthly capacity. It has kept customers filling that network at utilization above 90% and has produced 7nm-class silicon despite lacking EUV. Those are capabilities, not narratives.

The source of that success has changed. Early SMIC depended heavily on foreign technical knowledge and abundant capital. The 2010s rewarded operational normalization. Since 2020, Chinese industrial policy, domestic fabless growth and geopolitical separation have become powerful tailwinds. Management execution matters because fabs still have to yield functioning wafers at scale, but SMIC’s present strategic value cannot be separated from the era in which it operates.

Those tailwinds remain. China is not about to decide that semiconductor self-sufficiency matters less. Export restrictions make a domestically controllable foundry more valuable to Chinese customers. The A-share premium is a market expression of that strategic scarcity. Yet the same environment makes equipment harder to acquire and raises the cost of technological catch-up. The tailwind and headwind come from the same policy conflict.

Horizontally, SMIC’s clearest advantage over Hua Hong is scale and breadth. Its advantage over UMC and GlobalFoundries is access to the deepest captive domestic substitution pool. Its weakness against UMC and GF is current economic quality: those companies are achieving substantially higher gross margins without running fabs as full as SMIC. Tower illustrates a third path, using specialized technologies to capture high incremental value rather than winning primarily through scale.

This matters because foundry investment analysis can easily confuse utilization with moat. A 99% utilized fab that earns a 13% gross margin may be strategically valuable and economically mediocre. Hua Hong currently provides that example. SMIC at 93.1% utilization and 20.1% gross margin is better, but still well below UMC’s 32.5% margin at 85% utilization. The gap says SMIC has pricing and cost work left to do before domestic scarcity turns into peer-leading returns.

The next year is about whether Q2 establishes a new earnings level. The midpoint revenue guide of roughly US$2.88bn is strong enough that a clean realization with a 21–22% gross margin would show genuine operating leverage. A result close to US$2.9bn with sub-20% margin would tell a different story: demand would be excellent, but depreciation and new-fab economics would be consuming the benefit. The market should care more about the second outcome than the first headline.

The next three years are about returns on the 2023–26 investment cycle. Revenue can plausibly exceed US$12bn and still produce disappointing shareholder returns if another US$20bn-plus of capital is required to sustain the process roadmap. The denominator matters. A 5% return on US$50bn of operating capital creates less economic value than a 15% return on US$20bn, even when the first company reports much more revenue.

Five years out, equipment localization determines how much of SMIC’s strategic scarcity can become economic advantage. Chinese deposition, etch and related equipment suppliers are improving, and domestic immersion DUV progress is strategically important. But replacing the complete equipment, metrology, materials and software ecosystem needed for competitive advanced-node manufacturing is a much larger undertaking. The long-term bull case needs more than sanctions: it needs domestic tools to reduce the cost of operating under sanctions.

The current price is no longer anchored in the euphoria that produced the 2025 highs, but it does continue to pre-spend some future success. At HK$66.90, the H line implies approximately US$73bn of equity value against only US$685m of 2025 attributable profit and negative estimated owner earnings. A near-term revenue acceleration can close part of that gap. It cannot make capex disappear.

The market may currently be underestimating the strength of Q2 demand while still overestimating how much of that demand becomes free cash flow. Those two statements can coexist. A foundry can beat quarterly revenue expectations and remain fully valued because each extra dollar of revenue requires an exceptionally large installed capital base.

The old bull case deserves a split verdict. The operating argument was early: Q1 execution and Q2 guidance strengthened after the June report. The valuation argument was too generous: the old HK$82–95 bull band assumed the market would capitalize policy scarcity at a high multiple before SMIC proved better capital returns. I would not reproduce that framework today. My optimistic operating case can support a value near HK$96; a price materially above about HK$105 would already be paying more than 10% above that optimistic value.

The previous ideal-buy level around HK$42 happens to fall near my independently derived range, but for a different and stricter reason. My range comes from requiring at least a 20% discount to a freshly derived HK$54 conservative value, rather than carrying the old number forward. The arithmetic produces an upper buy threshold around HK$43.4.

The A-share price must be kept entirely outside that exercise. At the same date, the Shanghai market values the same economic share at roughly HK$149 equivalent. That does not make HK$66.90 cheap. It makes Shanghai extraordinarily expensive relative to Hong Kong. An A/H arbitrage gap can persist for years when capital pools and investor narratives are segmented.

For an existing H-share holder, HK$66.90 is close enough to my HK$74 base value to justify patience while Q2 is confirmed. For a new buyer demanding a margin of safety, it is not low enough. There is no dividend while waiting, conventional free cash flow remains negative, and the next five days contain an unusually information-rich catalyst in the August 13 results.

Bull reasons, each tied to the earlier analysis:

• Q2 guidance implies roughly US$2.86–2.91bn revenue, a 14–16% sequential jump, while gross margin is guided to 20–22% despite a rapidly rising depreciation base.
• Q1 utilization remained 93.1%, and China represented 88.9% of revenue, giving SMIC unusually direct exposure to domestic semiconductor substitution.
• Third-party teardowns confirm that SMIC can manufacture 7nm-class and newer internally named process variants despite advanced-equipment restrictions, preserving strategic scarcity.
• The balance sheet carries only modest net debt relative to roughly US$35bn of total equity, reducing the probability that the capex program becomes a liquidity crisis.

Bear reasons:

• PP&E capex reached US$8.40bn in 2025 against US$3.19bn of operating cash flow, leaving conventional free cash flow negative by more than US$5bn.
• Gross margin of 20.1% remains materially below UMC’s 32.5% and GlobalFoundries’ 28.3%, despite SMIC operating at higher utilization than UMC.
• Advanced-node economics remain opaque: teardowns establish capability but do not establish commercial yield or return on the capital required for multi-patterned DUV production.
• At HK$66.90 the H line still implies about 14 times 2025 EBITDA and over 100 times 2025 attributable earnings, with no shareholder distribution yield.

The first pre-mortem is an overcapacity script. By 2027, SMIC, Hua Hong and other mainland fabs complete the current mature-node expansion while AI-adjacent demand normalizes. SMIC utilization falls from above 90% to 78–82%, targeted price increases reverse and gross margin falls from about 20–22% to 13–15%. Depreciation approaches US$5–6bn annually because the new fabs are already in service. Normalized EBITDA falls toward US$5bn and the H-share multiple compresses from around 14 times trailing EBITDA to 7–8 times. With modest net debt, that combination can produce an H-share value around HK$30–35, roughly half today’s price.

The second is an advanced-node equipment script. During 2027–28, U.S. rules further restrict servicing or delivery of critical DUV, metrology or deposition tools. SMIC can still manufacture 7nm-class chips but cannot improve yields economically on the next process generation. Domestic customers remain captive, so revenue does not collapse; instead capex stays above US$8bn, gross margin remains below 20% and ROIC stays around 3–4%. The market stops valuing advanced-node activity as future profit and treats it as policy-mandated capital expenditure. A move toward 8-times EBITDA on US$4.5–5bn of normalized EBITDA would again put the share around HK$27–35.

My final judgment follows from those asymmetries. SMIC is a strategically scarce manufacturing asset whose near-term operating outlook is improving. Q2 guidance is strong enough that the current cycle should not be described as deteriorating. The balance sheet can finance the expansion, customer demand is broad, and export restrictions strengthen the domestic substitution channel.

The unresolved weakness is the conversion of strategic importance into shareholder economics. Five consecutive years of negative conventional free cash flow, rising depreciation, low-single-digit operating returns on an enormous capital base and zero dividends leave little room for a valuation mistake. HK$66.90 is far more defensible than the prices near the 2025 high, but it is around my base-value zone rather than a distressed valuation. Waiting for either a lower price or proof that the new capacity earns materially better returns is preferable to paying today for that proof in advance.

【Company-profile scores】

• Fundamental quality: medium
• Growth: medium
• Moat: medium
• Financial soundness: strong
• Management credibility: medium
• Valuation attractiveness: medium
• Risk level: high
• Suitable investor type: cyclical

【Investment rating】

• Rating: Hold
• One-line thesis: Near-full utilization and strong Q2 guidance support earnings, but negative owner earnings and export-control risk leave little margin of safety at HK$66.90.
• Current-price classification: acceptable hold
• Whether to wait for a better price: yes. A new-money entry becomes compelling below HK$43 if gross margin remains at least 20%, utilization remains above 90%, and no major new equipment restriction alters the process roadmap. The opportunity cost is missing a Q2-driven or policy-driven re-rating.
• Target holding horizon: 3–5 years, with a mandatory reassessment after the August 13 Q2 release.
• Expected annualized return: on a 12-month mark-to-fair-value basis, approximately -19% conservative, +10% base and +43% optimistic. If those values take three years to realize instead, the corresponding annualized price returns are roughly -6.7%, +3.3% and +12.8%, before distributions.
• Max-loss risk: roughly 50–60% in a combined mature-node overcapacity/export-control scenario that pushes gross margin into the low teens and the EV/EBITDA multiple toward 7–8 times.
• Reassessment-trigger signals: gross margin below 19% for two consecutive quarters; utilization below 85%; annual capex above US$9bn without a corresponding increase in ROIC; a new export-control rule materially restricting DUV/metrology/tool servicing; or sustained gross margin above 25% with improving owner earnings, which would invalidate the conservative return assumptions.

【Ideal Buy Price】38–43 HKD

Basis: the range lies at least 20% below the independently derived conservative fair value of about HK$54. The upper edge is approximately HK$43.4 on the stated margin-of-safety rule.

【Valuation Range】

• current: 66.90 HKD (close as of 2026-08-07)
• bear (conservative · ideal buy zone): [38, 43] HKD
• base (fair · acceptable hold zone): [63, 85] HKD
• bull (optimistic · above the clearly-overvalued line): [106, 115] HKD

The rating differs modestly from the supplied June 5 house conclusion. The earlier report said Watch. My independent work lands on Hold because two things have moved in the holder’s favor: the H-share price is lower and the latest formal operating evidence, especially Q1 execution and Q2 guidance, is stronger. It does not move to Cautious Buy because owner earnings remain approximately zero or negative after a reasonable maintenance-capex allowance and the current price remains above conservative value.

The research has four important blind spots. First, Q2 is five days away, making this report unusually exposed to imminent new information; every Q2 number here is guidance. Second, SMIC does not publicly disclose enough advanced-node volume, yield or margin data to value that business independently, so third-party teardown evidence receives a discount. Third, maintenance versus growth capex is not disclosed, so the owner-earnings estimate uses depreciation as the central maintenance proxy. Fourth, I did not have the underlying texts of the cited in-house TSMC, Hua Hong, GlobalFoundries, UMC, Tower, NAURA, AMEC and Tokyo Electron sibling reports, only the identifiers and selected conclusions provided in the research brief, so I cannot audit their detailed assumptions line by line.

The source hierarchy for this refresh is SMIC’s 2025 annual report, Q1 2026 quarterly report, July 30 board-meeting notice and July share-capital return first; peer company filings and investor releases second; BIS/HKEX and other regulatory materials for policy; and Reuters, TrendForce, TechInsights-derived reporting and market-data services for matters the company does not disclose. The valuation assumptions, A/H premium calculation, owner-earnings estimate and price attribution are my calculations from those inputs.

Other tickers mentioned

• 01347.HK — Hua Hong Semiconductor is the closest mainland-China specialty-foundry comparison, with near-full utilization but materially lower gross margin.
• UMC.US — United Microelectronics is the mature-node capital-discipline benchmark, with much higher gross margin despite lower utilization.
• GFS.US — GlobalFoundries shows how a strategically protected specialty foundry can monetize geographic and customer scarcity at higher margins.
• TSEM.US — Tower Semiconductor is the specialty-niche comparison, especially for silicon photonics, analog and AI optical-connectivity exposure.
• 2330.TW — TSMC is referenced as a technology and industry-cycle benchmark but deliberately excluded from the direct valuation peer set.
• 002371.SHE — NAURA is relevant to the localization of Chinese semiconductor manufacturing equipment under export controls.
• 688012.SHG — AMEC is relevant to China’s improving domestic etch/deposition tool ecosystem and the gradual substitution of imported equipment.
• 8035.TSE — Tokyo Electron is part of the global wafer-fab-equipment context against which SMIC’s equipment-access constraint must be understood.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.