SigmaStar Technology (301536.SZ) Investment Report

Rating: Cautiously Bearish | Target Price: RMB 75–90 | Current Price: RMB 122.31 (Close 2026-08-14) | Margin of Safety: Approx. −39% | Time Horizon: 6–12 Months

Valuation Verdict: Overvalued. Probability-weighted fair value across three scenarios is approximately RMB 86, roughly 29% below the current price.

---

Key Conclusions Summary

SigmaStar Technology is the global leader in edge-side visual AI SoC shipments (26.7% market share in 2024; 41.2% in the security surveillance segment), genuinely positioned at the inflection point of edge AI demand—a quality growth stock with a real moat. However, the current price of RMB 122.31 (market cap ~RMB 51.6 billion) already prices in "extended memory super-cycle tailwinds + sustained above-industry growth + peak earnings × high multiple" simultaneously: roughly two-thirds of the 2026 net profit surge (interim forecast RMB 820–900 million, +583%~650% YoY) comes from memory price increases transmitted through the "SoC + embedded memory" model and inventory value appreciation—a cyclical tailwind that cannot be linearly extrapolated. Even under the market consensus "memory prices plateau at high levels" scenario, net profit attributable to shareholders will decline from a peak of ~RMB 1.7 billion in 2026 to RMB 900 million–1.4 billion in 2027 — the high growth rate in the reporting period represents the cyclical peak of "volume and price rising together," not sustainable steady-state earnings. Combined with the lock-up expiry of approximately 234 million shares on 2027-03-29 (55.6% of total share capital), high customer concentration (top five account for 88%), and inventory devaluation risk on RMB 1.22 billion of high-cost inventory, the current price lacks a margin of safety. We recommend avoiding the stock and reassessing after the earnings center normalizes and lock-up expiry risk is released.

---

Investment Thesis

C1 — SigmaStar is the global leader in edge-side visual AI SoCs, positioned at a genuine demand inflection point (Confidence 0.72)

Key Evidence

• Source: Frost & Sullivan (as cited in SigmaStar's HK IPO prospectus) / Locator: Market share and industry space—2024 global visual AI SoC shipment market share of 26.7% (No. 1), security surveillance SoC with AI compute shipments of 41.2% (No. 1), robotics visual AI SoC shipments ranked No. 2 globally in 2025H1 (23%); global visual AI SoC shipments of approximately 246 million units in 2024, projected to reach 954 million units by 2029, CAGR of approximately 31.1%.
• Source: SigmaStar Technology 2026 Interim Results Announcement / Locator: Earnings guidance range—2026H1 revenue of RMB 2.62–2.72 billion, +86.74%~93.86% YoY; net profit attributable to shareholders of RMB 820–900 million, +583.72%~650.42% YoY.

The company has accumulated 600+ products across its full-stack visual IP (ISP+NPU+audio/video codec), with cumulative AI-compute SoC shipments exceeding 550 million units. Customer validation cycles and automotive-grade certifications constitute entry barriers. However, two definitional caveats: first, "31.1%" is the CAGR for shipment volume, not "penetration rate" growth (the five-year CAGR of visual SoC AI penetration from 39.1%→84.8% is only ~16.8%); second, 41.2% refers to "security surveillance SoCs with AI compute" and 23% refers to "2025H1 shipment volume"—neither represents total security surveillance SoC market share. Additionally, industry penetration gains are an industry-wide tailwind; HiSilicon (high-end surveillance/automotive), Ambarella (7nm automotive), and Rockchip (AIoT platform) are all competing for share—SigmaStar does not have this market to itself.

C2 — Approximately two-thirds of the 2026 earnings surge comes from cyclical memory price tailwinds and cannot be linearly extrapolated (Confidence 0.60)

Key Evidence

• Source: SigmaStar Technology 2026 First Quarter Report / Locator: Explanation of major accounting data changes—"Upstream memory supply is tight with prices continuously rising; the company reasonably passed through costs and continuously optimized product and customer structure... earnings growth is the result of simultaneous volume and price increases driven by price pass-through combined with shipment growth"; Q1 gross margin of 46.05%, +12.77pp YoY, a record high since 2022 (FY2025 full year was only 34.16%).
• Source: Times Weekly, 2026-03-19 / Locator: Gross margin—Smart security surveillance gross margin from 2021–2025: 46.40%→38.64%→34.41%→33.42%→32.38%, declining for four consecutive years; industry price wars are a long-term reality.

The company adopts a "SoC chip + embedded memory integrated solution + turnkey service" model, with memory accounting for the bulk of raw materials. When memory prices rise, the company builds inventory at lower costs and passes through higher prices with its products, capturing "inventory appreciation" excess gross profit—a one-time tailwind: the incremental jump in Q1 gross margin from the ~33% norm to 46% comes predominantly from memory price increases rather than product mix upgrades. Key judgment: regardless of whether memory prices plateau at high levels or decline in 2027, as long as prices stop rising, gross margin will revert toward the 34%–38% norm, with a certain decline between 2026 peak earnings (~RMB 1.7 billion) and 2027 normalized earnings (RMB 900 million–1.4 billion under plateau; RMB 450–650 million under reversal). The shipment volume (unit count) and ASP breakdown in the 8/27 interim report will be key evidence for validating the respective contributions of "volume growth" and "price growth."

C3 — The current price is priced for the bull case; valuation is significantly above fair value baseline (Confidence 0.65)

Key Evidence

• Source: This report's own calculations / Locator: Valuation multiples—PE(TTM) of approximately 108x (post-interim report TTM basis will drop to ~49x, so this is context only); PB of approximately 16x, at a high percentile of the ~2.4-year window since listing (ROE only ~10% and declining in recent years).
• Source: This report's valuation model / Locator: Three scenarios and implied expectations—Based on 2026 peak net profit attributable to shareholders of ~RMB 1.725 billion, the implied PE is ~30x; based on 2027 normalized net profit attributable to shareholders of ~RMB 1.05 billion, the implied PE is ~49x (PEG of ~1.5–2x, higher than Rockchip's ~1.5x, near the lower bound of Amlogic's ~2.8x).

The current price implies market expectations of sustainable normalized earnings of RMB 1.47–1.72 billion by 2028–2029 (three-year CAGR of ~39%). However, extrapolating from industry demand mathematics (shipment CAGR of 31.1% × normalized net margin of 11%–12% × company share), 2029 normalized earnings would be only ~RMB 1.08 billion—a gap of ~35%–60% that would require "extended memory tailwinds + simultaneous share and margin expansion" to close, which is precisely the condition for the bull case (≈current price), not the base case. The substance of the valuation debate is not "30x on reported earnings vs. 85x on normalized earnings," but rather what multiple normalized earnings should command—we believe a discount is warranted for memory volatility, the 2026→2027 earnings decline, and lock-up expiry supply, warranting 33–37x (below Amlogic's 41x and Rockchip's 60x), corresponding to a fair value baseline of RMB 75–90.

C4 — Clear valuation-compression catalysts with specific timing, compounded by earnings quality and positioning risks (Confidence 0.58)

Key Evidence

• Source: TrendForce 2026-07 / Locator: Contract prices—2026Q3 DRAM contract prices expected to rise 13%–18% QoQ, NAND 10%–15% QoQ, significantly narrowed from Q2's 55%–63%; Bernstein's 2026-07 tracking notes "price increases approaching ceiling, but supply tightness extends into 2027."
• Source: Lock-up expiry schedule (CNFOL/Sohu Securities) / Locator: 2027-03-29—IPO original shareholder lock-up expiry of approximately 234 million shares, representing 55.6% of total share capital and ~125% of current float (SigmaStar 121 million shares, Elite Star 35.48 million shares, etc.).
• Source: SigmaStar Technology 2025 Annual Report / Locator: Major sales customers—Top five customers account for 88.16% of total annual sales; distribution channel share approximately 93%.

Three risks compound: ①Memory contract price increases have confirmed narrowing, with 2027H2–2028 judged by TrendForce/36Kr as the cyclical peak; when the tailwind recedes, earnings decline will coincide with devaluation risk on RMB 1.22 billion of high-cost inventory (end of 2026Q1); ②The 2027-03-29 lock-up expiry is approximately 1.25x the current float, with stronger sell-side incentives at elevated valuations (precedent: Qunkun Capital reduced ~4.5 million shares after the first lock-up expiry in 2025-09); ③Top five customers account for 88%, with the single largest at ~35%, and ~90% of revenue flows through distributors, making end-customer inventory opaque; one end customer previously suspended orders due to changes in its operating environment. Lock-up expiry does not necessarily mean selling (SigmaStar is a MediaTek-affiliated strategic shareholder with a precedent of disciplined reduction), but the resonance of supply shock and elevated valuation constitutes a clear valuation-compression risk within 6–12 months.

---

Key Financial Data

Reasons for Metric Changes (YoY ≥±20% with Management Explanations)

• 2025 Revenue +26.28%: Earlier R&D projects entered the harvest phase, with new growth curves in automotive and robotics scaling up; combined with tight memory supply in 2H, the company leveraged strategic reserves + the "SoC + embedded memory" integrated solution to consolidate competitiveness (2025 Annual Report MD&A).
• 2025 Non-GAAP Net Profit +39.20%: Non-recurring items were only RMB 56 million (investment interest RMB 42 million + government grants RMB 25 million), indicating enhanced core profitability (2025 Annual Report).
• 2026Q1 Revenue +49.35%, Net Profit Attrib. to Parent +330.29%, Non-GAAP +426.70%: Edge/end-side AI application explosion + memory price pass-through + product and customer structure optimization, with volume and price rising together; gross margin rose to 46.05% (2026 Q1 Report).
• 2025 Selling Expenses +49.39%: Increased headcount compensation from Security Expo, developer conference, and new marketing personnel.
• 2025 Financial Expenses +1825%: Increased interest expense from new borrowings; reduced exchange gains from USD/RMB depreciation.
• End of 2025 Accounts Receivable +110.6% (RMB 115M→242M): Increased revenue and higher period-end receivables with credit terms (explained in Q3 report; not separately noted in the annual report's asset change section).
• 2026Q1 Inventory +30.8% (RMB 930M→1.217B): Strong downstream demand + memory shortage; proactive strategic stockpiling to secure order fulfillment.
• 2026Q1 Operating Cash Flow −111.4% to negative (−RMB 16M): Proactive increase in raw material stockpiling, with significantly higher procurement cash outflows.

---

Latest Results Commentary (2026H1 Earnings Guidance)

The company released its interim earnings guidance on 2026-07-18: Net profit attributable to shareholders of RMB 820–900 million (+583.72%~650.42% YoY), revenue of RMB 2.62–2.72 billion (+86.74%~93.86% YoY), significantly exceeding prior market consensus (pre-guidance TUSHARE consensus 2026E EPS attributable to parent was only RMB 1.50, already surpassed by the single H1 period's RMB 820–900 million).

Implied Q2 single-quarter breakdown: revenue of RMB 1.626–1.726 billion, net profit attributable to shareholders of RMB 600–680 million, corresponding to a single-quarter net margin of 36.9%–39.4%—another step up from Q1's 22.2%, +173%~+209% QoQ. This net margin level far exceeds the company's FY2025 norm of 10.4%. The driver is the resonance of "memory price pass-through + high-end SSR670 and other product ramps," of which the inventory value appreciation from memory price increases is a one-time tailwind and not sustainable.

Judging whether results "beat expectations" requires two levels: On a reported basis, RMB 820–900 million is indeed a significant beat (prior sell-side 2026E full-year forecasts were only RMB 458–630 million; half a year alone exceeded the full-year forecast); however, on a quality basis, the "quality" of the beat is questionable—approximately two-thirds of the surge is contributed by memory price pass-through, representing a cyclical tailwind rather than steady-state earnings. The 8/27 official interim report's segment shipment volumes, gross margin QoQ trends, and 2H guidance are key to judging whether the "tailwind can persist" and whether "the market has fully priced this in." There is currently no reliable sell-side consensus to serve as a safety cushion (all existing research reports were published before the earnings guidance and are severely outdated), so we do not conclude that "the market has priced this in"; we merely state management's characterization ("quarter-over-quarter improvement, high growth sustainable," qualitative without numbers) and the company's historical growth rate.

---

I. Business Model and Earnings Quality

Business Model Overview: Fabless, asset-light SoC design; visual AI edge/end-side chips; design and R&D with distribution-led sales; wafer manufacturing and packaging/testing fully outsourced. Revenue is primarily one-time chip sales with no subscription-based recurring revenue; downstream is primarily distributors (~93%); weak bargaining power upstream against wafer foundries and memory suppliers; moderate-to-weak bargaining power downstream due to channel concentration.

Cash Content of Earnings Test: OCF/Net profit attributable to parent declined from 2.12 in 2023 → 1.63 in 2024 → 1.45 in 2025 (still >1, healthy), turning sharply negative in 2026Q1 due to proactive stockpiling (−RMB 16M, OCF/net profit attributable to parent of ~−0.07); FCF/net profit attributable to parent was 1.10/0.86/1.15 for 2023–2025, with extremely low capex (only RMB 92M in 2025)—free cash flow quality is good. Warning point: Working capital absorption has increased significantly—inventory +30.8% QoQ in 2026Q1, accounts receivable +110.6% at end of 2025; high growth is funded by "stockpiling + extended credit terms," and the Q1 cash flow turning negative is a real signal.

Recurring Earnings Test: 2025 non-recurring items were RMB 56 million, representing 18.3% of net profit attributable to parent (investment interest RMB 42 million + government grants RMB 25 million); non-GAAP net profit of RMB 252 million < reported net profit attributable to parent of RMB 308 million—approximately 18% of book profit comes from one-time/investment items; core earnings should be viewed on a non-GAAP basis. The company does not self-report a Non-GAAP measure; A-share practice uses non-GAAP (deducting non-recurring items) as the standard.

Return on Capital: 2025 ROIC of approximately 14.4% (non-GAAP net profit of RMB 252M / (interest-bearing debt + equity attributable to parent − cash-like assets)), above WACC (~10%) but not at a strong-moat level (>15% would signal a true moat), and ROE has declined from 32.5% in 2022 to approximately 9.9% in 2025.

Maintenance Capex Scrutiny: The Fabless model has extremely low capital intensity; 2025 capex was only RMB 92 million, ~20% of OCF; fixed assets were only RMB 268 million (mostly office buildings). There is no "reinvesting earnings to maintain capacity" issue—a typical asset-light model.

Moat / Red Flags: The moat comes from proprietary full-stack visual IP, the preferred high market share among global security surveillance brand customers, and a 600+ product matrix. Red flags are concentrated in—①Extremely high customer concentration: top five customers at 88.16%, single largest at 35.24% (2025), distribution share at 93%; large customer/channel dependence is the core vulnerability; ②Earnings contain significant non-recurring items (18.3%); ③Gross margin jump partially stems from one-time memory price pass-through, with earnings center at risk of decline when prices fall; ④The company holds ~RMB 2.9 billion in cash-like assets while simultaneously carrying ~RMB 1.5 billion in interest-bearing debt (end of 2026Q1), with financial expenses +1825%—of the "cash-like assets," RMB 760 million are other non-current financial assets (external investments) and RMB 270 million are large-denomination certificates of deposit; truly discretionary cash is limited, and the interest-bearing debt is likely used for strategic memory stockpiling—capital efficiency warrants attention.

---

II. Management Assessment

Consistency of Words and Actions: Management's guidance of "building momentum, improving quarter over quarter" has a high fulfillment rate—the 2024 annual report outlook for steady 2025 growth was fulfilled with actual revenue +26.28% and non-GAAP net profit +39.20%; multiple research briefings repeatedly guided "quarter-over-quarter improvement, high growth," and 2025 saw four consecutive quarters of sequential revenue growth, with 2026Q1 revenue +49.35%—fulfilled. However, long-term targets lean aggressive: for the LiDAR chip "mass production in 2026, 10 million units in 2027, targeting global leadership," the SS901 has been mass-produced in flagship models of leading domestic OEM brands (2026Q1, partially fulfilled), but "10 million units/global leadership" is unverified. Assessment: partially fulfilled, with a preference for aggressive long-term expectations.

Shareholder Friendliness: Continuous dividends since listing—FY2024: RMB 2 per 10 shares (payout ratio ~32.9%); FY2025: RMB 3 per 10 shares (payout ratio ~40.86%); January–February 2026: buyback of 1.7448 million shares for approximately RMB 119 million, completed. Dilution: 2024 IPO of 42.11 million shares raising RMB 680 million; three restricted stock incentive plans in 2024/2025/2026 (2025 grant of 927,200 shares @ RMB 33.25); H-share issuance in progress (potential dilution). Assessment: neutral-to-shareholder-friendly (dividends and buybacks are acceptable, but equity incentives and H-share issuance create ongoing dilution).

Risk Signals: No controlling shareholder or actual controller (largest shareholder SigmaStar is MediaTek-affiliated, holding 28.74%; director Chen Xuan-ni comes from MediaTek's finance division); governance structure is stable but lacks an actual controller backstop; 2025 acquisition of Furui Kun (53.31%) created goodwill of RMB 153 million, with additional external investments in Yuanchuan Micro, Beijixin Micro, Tuoyuan Intelligence, and participation in industrial funds—expansion is relatively aggressive; share-based payment expense of RMB 33 million was charged to expenses.

III. Business Segment Breakdown

Core Profit Segment: Gross profit contribution calculated as "revenue share × gross margin" — Smart Security 65.10%×32.38% ≈ 21.1, Smart IoT 22.14%×37.63% ≈ 8.3, Smart Automotive 10.69%×40.84% ≈ 4.4. Smart Security contributes ~62% of gross profit and is the absolute profit engine (despite having the lowest gross margin); Smart IoT and Automotive together contribute ~38% with higher growth and better margin structure, representing future earnings flexibility.

Gross Margin Structure Variance: Smart Automotive (40.84%) vs. Smart Security (32.38%) — an 8.5pp gap driven by different business models — automotive SoCs require AEC-Q100 certification, commanding higher front-loading pricing power and ASP; security faces HiSilicon's return and price wars from Fullhan Micro, Ingenic, etc., forcing price reductions. Smart IoT's gross margin fell 6.7pp in 2025 due to robotics' low-price share-grabbing strategy, confirming the competitive reality of "volume growth at the expense of margins."

---

IV. Financial Engineering and Inter-Period Consistency

Accounting Red Flags

1. Receivables surge far exceeding revenue (Moderate): 2025 year-end receivables of RMB 242M, +110.6% vs. beginning of period, while revenue grew only +26.28% — receivables growth ~4x revenue growth; the annual report's asset change notes are blank on this, with only the Q3 report explaining "increased revenue and higher receivables from extended payment terms" — cash collection ratio declining, possibly related to loosened credit terms for distributors/key customers.
2. Inventory buildup turning operating cash flow negative (Low): Inventory from RMB 930M at 2025 year-end to RMB 1.217B at 2026Q1 end (+30.8%); 2026Q1 operating cash flow of −RMB 16M, explained by management as "proactive strategic stockpiling"; if memory prices decline, inventory write-down risk exists (2025 year-end inventory impairment reserve of RMB 32.44M) — requires tracking of destocking and impairment provisions.

Inter-Period Consistency

1. Gross margin (declined for three consecutive years 2023–2025: 36.5%→35.8%→34.2%, then spiked to 46.05% in 2026Q1) — consistent with management's explanation of "memory cost pass-through + product mix upgrade," but note that the one-time contribution from memory price increases within the 46.05% has not been quantified.
2. Operating cash flow / net profit (OCF/attributable NP: 2.12→1.63→1.45, declining annually, turning negative in 2026Q1) — consistent with the "proactive stockpiling" explanation; methodology consistent across periods.
3. Receivables as % of revenue (~4.9% at 2024 year-end → ~8.1% at 2025 year-end, still elevated at RMB 246M in 2026Q1, ~25% of quarterly revenue) — unexplained by the company; credit pressure persists and requires ongoing monitoring.

---

V. Valuation and Risk-Reward

Current Market Data: Share price RMB 122.31 (close 2026-08-14), total market cap ~RMB 51.58B, float market cap ~RMB 25.6B; PE(TTM) ~108x, PB ~16x (high percentile in ~2.4-year post-IPO window). Earnings not representative: TTM earnings exclude 2026H1 results and already include memory price tailwinds from Q1 onward; PE(TTM) percentile mechanically elevated — provided for context only, not as a headline cheap/expensive signal.

Peer Comparison:

Sigmastar's PB of 16x is significantly higher than peers (excluding Rockchip at 18.3x, others range 5.4–6.1x), yet ROE is only 9.9% — the high valuation implied by PB/ROE is supported primarily by the "edge AI leader" growth narrative rather than asset returns.

Market Implied Expectations: At the current RMB 51.6B market cap, the market implicitly believes sustainable normalized earnings of RMB 1.47–1.72B by 2028–2029 (three-year CAGR ~39%); however, extrapolating from demand math (shipment CAGR 31.1% × normalized net margin 11%–12% × share), 2029 normalized earnings would be only ~RMB 1.08B — a 35%–60% gap. The current price requires "memory tailwinds persisting + growth exceeding industry by 8pp + both share and margin expansion" — a bull-case condition, not a base case.

Three-Layer Value (EPV): Asset value (liquidation floor) ~RMB 7.64/share; zero-growth EPV ~RMB 14.9/share (normalized EPS RMB 1.40, WACC 10%, net cash per share RMB 0.92); ~88% of the current RMB 122.31 price is supported by growth optionality. If the edge AI structural inflection materializes, zero-growth EPV is a floor rather than a conclusion — whether growth deserves payment is determined by the scenario analysis below — but an 88% growth-option share means extremely thin margin for error; if growth or tailwinds are falsified, valuation downside is substantial.

Three Scenarios and Risk-Reward:

Current price of RMB 122.31 falls within the bull-case range (≈bull midpoint of RMB 120), indicating the market has priced in the optimistic path of "extended memory super-cycle + second curve delivery" rather than the base case. Probability-weighted fair value is ~RMB 86, ~29% below the current price.

Proprietary Forward Estimates: FY2026 revenue RMB 5.8–6.4B, attributable NP RMB 1.55–1.9B (H1 already RMB 820–900M + H2 ~RMB 700M–1.0B, with H2 gross margin declining to 36%–42%); FY2027 revenue RMB 6.0–7.2B, attributable NP RMB 900M–1.4B (wide range depending on memory price trajectory). Management guidance is qualitative ("sequential improvement, high growth sustainable" — no numbers); sell-side consensus (pre-announcement) FY2026E EPS RMB 1.50, FY2027E RMB 1.81 — already significantly exceeded by actual H1 results, severely lagging, and cannot serve as a safety cushion.

Conclusion: Overvalued. Quality and price are separate — the company's quality is excellent (global leader + genuine inflection point), but the current price of RMB 122.31 is ~49% above the base-case fair value midpoint of RMB 82 and ~42% above the probability-weighted fair value of ~RMB 86, with a margin of safety of approximately −39% (negative). Base-case exit multiple anchoring rationale: referencing comparable companies' 2026E forward PEs (Amlogic ~41x, Rockchip ~60x), taking the 33–37x lower end, with discounts applied for Sigmastar's memory cycle exposure, 2026→2027 reported earnings decline, and lock-up expiry supply; the current PB of 16x (ROE only 9.9%) versus peer PB of 5.4–6.1x represents a significant divergence that independently does not support the current price.

---

VI. Industry Landscape and Competitive Dynamics

Industry Space: The company's "vision AI SoC / edge AI inference chip" track is transitioning from "introduction" to "scaled deployment." According to Frost & Sullivan (as cited in Sigmastar's HK IPO prospectus), the 2024 global vision AI SoC market was ~246M units (shipment basis), with AI penetration of 39.1% in vision SoCs, projected to reach 954M units by 2029 with penetration rising to 84.8%, corresponding to a shipment CAGR of ~31.1% (note: penetration five-year CAGR ~16.8% — the two metrics should not be conflated). On a broader edge AI chip basis (including SoC/SiP), QYResearch estimates ~USD 65.28B globally in 2024, reaching USD 216.41B by 2031; another source cites RMB 93.22B for 2024 — significant methodology differences requiring verification. China's smart security market overall was ~RMB 591.02B in 2025 (Zhiyan Consulting). Headline figures use Frost & Sullivan shipment data (first-tier institution, cited via company prospectus); QYResearch/Zhiyan Consulting used only for cross-reference.

Quantified Demand Inflection Chain (Structural Inflection): This company captures an "exogenous demand inflection" — the technology shift toward edge AI/edge inference is repricing the "security-dominated existing market" as an "all-scenario edge AI incremental market," and the chain is quantifiable: ①Driver variable trajectory: Vision SoC AI penetration from 39.1% in 2024 to 84.8% in 2029, shipments from 246M to 954M units; ②Unit usage/feature upgrade: AI SoCs with integrated NPU deliver higher per-unit compute and value (SSR670 series at 8T compute, SAC8905 at 32 TOPS, ASP higher than traditional IPC SoCs); robotics/LiDAR represent entirely new categories (LiDAR SPAD SoC from 400K units in 2024 to 12.3M units in 2029, CAGR 98.4%); ③Penetration ramp: In 2026, the company's products fully upgrade to advanced process nodes, with robotics chips, AI glasses, and automotive LiDAR chips entering mass production; ④Company share: Global #1 in security vision AI SoC shipments (41.2%), global #2 in robotics vision AI SoC (23%, 2025H1). Industry-level quantification is complete (CAGR 31.1%); company-level partially quantified (revenue growth of +87%~94% verified, but per-product ASP breakdown not disclosed).

Value Chain and Value Distribution: Upstream is wafer foundry (dependent on a few foundries like TSMC; top five suppliers account for ~80% of procurement), packaging/testing, and key components such as memory (2025–2026 memory price increases are a key variable); upstream concentration is high, and Fabless companies have weak bargaining power over advanced process nodes. Midstream is IC design (IP cores + SoC design + AI toolchain), where value is primarily captured by designers. Downstream includes security camera/robotics/wearables/automotive Tier-1 OEMs. The company sits in the midstream IC design segment, the primary margin retention point (Fabless design gross margins of 30%–40%, higher than downstream OEMs), but has weak bargaining power over upstream wafers/memory and moderate-to-weak bargaining power downstream due to high customer/channel concentration.

Supply-Demand and Competitive Landscape: Demand is driven by both "volume" (penetration ramp, embodied intelligence/robotics scaling, AI glasses, automotive, LiDAR) and "price" (4K+, edge computing, high-compute upgrades raising ASP); supply side consists of Fabless design capacity with numerous players (HiSilicon, Fullhan Micro, Ingenic, Goke Micro, Rockchip, Amlogic, Ambarella, Novatek, Allwinner). During 2025–2026, upstream memory and key component supply is tight, favoring leading players with supply chain reserves; the overall market is in a tight balance with improving sentiment. Competitive landscape: In security vision AI SoC, Sigmastar held 41.2% global shipment share in 2024, ranking #1; after HiSilicon's contraction due to sanctions, share was absorbed by Sigmastar, Fullhan Micro, Ingenic, and Goke Micro (under the old methodology approximately 25%/17%/12%/5% respectively, timed during HiSilicon's retreat — methodology pending verification). Entry barriers: SoC design requires long-term IP accumulation, lengthy customer validation cycles, automotive-grade certification (AEC-Q100), and high capital requirements for advanced process tape-outs. Substitution and competitive threats: HiSilicon returned to security in 2024 with the Hi3516CV610, targeting consumer-grade with high cost-performance "rural-encircling-cities" strategy, normalizing price wars (2025 vision AI SoC average prices down 5%–20%); high-end scenarios face competition from Ambarella (7nm automotive) and Rockchip (AIoT platform + edge co-processors).

Cycle and Regulation: The industry is in an upcycle + early growth stage (2026H1 edge AI explosion + memory shortage amplifying the cycle). Leading indicators: edge AI SoC shipments and AI penetration, upstream memory/component prices, robotics/AI glasses/LiDAR new product ramp pace, company quarterly gross margin and automotive customer wins (30+ customers onboarded). Regulation is neutral-to-slightly-negative: semiconductor localization is a tailwind (security/automotive chip domestic substitution), but US export controls/EDA/advanced process supply restrictions constrain high-end R&D pace; the 15th Five-Year Plan includes low-altitude economy and edge computing as priorities. Memory supply-demand is the largest current variable — TrendForce sees the DDR cycle turning point at end-2027, 36Kr sees 2027H2–2028 as the cyclical peak, directly impacting Sigmastar's "embedded memory" model.

Peer Comparison (Operational Dimension): See valuation comparison table above; operational additions — Amlogic (set-top box/display SoC leader, edge compute chips shipped >20M units in 2025, +160%), Rockchip (AIoT platform + edge co-processor RK182X, ROE 26.2% significantly leading), Ingenic (automotive-grade memory at 61%, direct beneficiary of the memory super-cycle, same tailwind source as Sigmastar), Fullhan Micro (security ISP leader, 2025 revenue/net profit both declining), Ambarella (high-end edge AI vision, 5nm/4nm, 58% gross margin but GAAP loss-making). Sigmastar has the highest focus and shipment market share in vision AI SoC, but its scale and profitability are weaker than Rockchip/Amlogic.

Company Industry Positioning: Sigmastar is the global leader in edge vision AI SoC (2024 shipment market share of 26.7% globally #1, security segment 41.2% #1, robotics vision 23% global #2), with share trending upward (2025 shipments >180M units, cumulative AI-capable shipments >550M units). Moat sources: proprietary full-stack vision IP and 600+ product matrix, leading market share and channel scale as first choice among global security brand customers, robotics/AI glasses/LiDAR/automotive new category positioning + supply chain reserve resilience. Market position methodology note: Company's self-reported figures are broadly consistent with third-party agency data (F&S's 26.7%/41.2% on shipment basis, 23% on 2025H1 shipment basis), with no material methodology conflicts; key weaknesses are declining gross margins (four consecutive annual declines in security), high customer/distributor concentration, weak upstream bargaining power, and smaller scale/profitability versus Rockchip/Amlogic.

---

VII. Comprehensive Conclusion and Tracking

Overall Assessment: Cautiously bearish (confidence 0.52). This is a high-quality global leader positioned at a genuine edge AI inflection point, but ~two-thirds of its 2026 earnings surge is contributed by the cyclical memory price tailwind. The current price of RMB 122.31 already prices in the bull case of "extended memory super-cycle + second curve delivery" (falling within the bull-case fair value range), ~49% above the base-case fair value midpoint of RMB 82, with a margin of safety of approximately −39%. The core risk is not "the company is bad" but "the price is too expensive" — even if memory prices plateau per consensus, 2026→2027 reported earnings will decline from the ~RMB 1.7B peak, and combined with the massive lock-up expiry on 2027-03-29 and high customer concentration, valuation compression pressure exists within 6–12 months.

Strategy: Existing holders advised to reduce positions on strength and realize the valuation premium from cyclical tailwinds; non-holders should avoid and not chase, waiting for two signals before reassessing — ①memory contract prices flattening/declining in 2026Q4–2027 (tailwind retreat confirmation); ②the 2027-03-29 lock-up expiry and sell-down expectation digestion. Target price range RMB 75–90 (base case); below RMB 75 enters the value observation zone; approaching the RMB 52 bear-case midpoint provides sufficient margin of safety.

Key Risks (Contrarian View): If the memory super-cycle extends into 2027–28 (DRAM shortage exceeding expectations, HBM crowding out capacity) and the LiDAR/robotics second curve delivers, 2027 attributable NP could reach RMB 1.6–1.8B (bull case), supporting or even pushing the current price higher; if the 8/27 interim report shows gross margin sustaining >44% and attributes it to product mix upgrade (rather than memory pass-through), the "unsustainable tailwind" thesis would need upward revision. Additionally, in an A-share liquidity-driven market, high valuations for quality leaders may resolve through "sideways digestion" rather than "downward reversion," creating uncertainty in the timing of downside realization.

Key Tracking Milestones for Next 6–12 Months:

• 2026-08-27: Interim report official release — review Q2 gross margin QoQ, shipment volume (units) and ASP breakdown, H2 guidance.
• From 2026Q4: Memory contract price trajectory (TrendForce/DRAMeXchange quarterly data) to determine tailwind retreat timing.
• 2027-03-29: 234M restricted shares unlock; monitor major shareholder/VC sell-down announcements beforehand.
• Ongoing: H-share listing progress (dilution and sentiment impact), LiDAR/robotics/AI glasses new product ramp, receivables and inventory impairment.

---

This report is based on public information as of 2026-08-17, with the current price anchor being the 2026-08-14 closing price of RMB 122.31. Core materials: 2025 annual report, 2026 Q1 report, 2026 interim earnings preannouncement, Frost & Sullivan industry data, TrendForce/DRAMeXchange memory price tracking, and historical investor relations meeting minutes.