| Meta Info | Content |
|---|---|
| Rating | Neutral |
| Target price range (base-case fair value) | RMB 115–150 |
| Current price (close 2026-10-23, price anchor of this report) | RMB 124.79 |
| Margin of safety (base-case fair value floor vs current price) | Approx. -7.8% (no margin of safety) |
| Probability-weighted expectation | Approx. -1% (Bear 30% / Base 50% / Bull 20% weighted fair value approx. RMB 123) |
| Time horizon | 12 months |
| Market cap / total shares | Approx. RMB 65.6 billion / 526 million shares |
VeriSilicon is mainland China's No.1 semiconductor IP licensing player (No.8 globally per IPnest) plus a one-stop chip custom design (SiPaaS) platform, standing right at a structural inflection point in AI ASIC customization demand: as of end-June 2026, orders in hand totaled RMB 12.449 billion (approx. 3.95x 2025 revenue, 83% from AI data processing), and new orders signed from January to mid-August reached RMB 15.142 billion (approx. 254% of full-year 2025). Growth visibility on "volume" is locked in by the order book, but the company has never posted a profitable year in its disclosure history: in 2026H1, while revenue grew +91.4%, net loss attributable to shareholders widened to RMB 612 million, and blended gross margin fell for four consecutive periods to 31.34%. The current price of RMB 124.79 (approx. 67% off the June peak) corresponds to PS(TTM) of approx. 16.2x and implied 2027E PS of approx. 6.3x — the market has already priced in the base-case success path of "orders converting on schedule + adjusted breakeven in 2027." The base-case fair value range of RMB 115–150 essentially overlaps with the current price, leaving no meaningful margin of safety; both tails are wide and roughly balanced (Bear RMB 30–50 / Bull RMB 190–260). Key constraints: zero historical validation for the margin leap (adjusted net margin -22.7% → target 15%+), concentration in the single AI vertical, fund-plate (Big Fund) exit and pending H-share issuance creating share dilution. Conclusion: Neutral. The quality inflection is real, but the price offers no discount. Await confirmation signals from the Q3 report and 2026H2 adjusted EBITDA turning positive before taking a directional stance.
Key evidence:
The order figures themselves are verifiable facts disclosed by the company; however, it must be clear-eyed that the order-to-revenue conversion cycle is 9–12 months for design services and 6–18 months for mass production, and "subject to factors such as customers' product shipment plans and partner foundries' capacity scheduling" (semi-annual report original text) — the conversion bottleneck lies partly with third-party foundries rather than the company. The 83–90% AI concentration means order book momentum is inseparable from AI sector beta; Alchip's -40% revenue drop in 2025 after losing a single major customer is a realized precedent in the same supply chain. Cross-validation of supply chain demand (Broadcom's AI quarterly revenue of approx. USD 17 billion, GUC's +104% revenue growth in the first 8 months of 2026) supports genuine demand, but if AI capex enters a digestion phase in 2027, the absolute order book value does not constitute a revenue guarantee.
Key evidence:
This report's reverse calculation: at Q2's standalone gross margin of approx. 30.6% and an H2 adjusted cash expense run-rate of RMB 810–850 million, adjusted EBITDA turning positive in H2 requires only approx. RMB 2.6–2.8 billion of H2 revenue (+21%–28% YoY vs 2025H2's RMB 2.179 billion; +42%–50% HoQ vs H1) — with order book support, revenue is likely reachable. The risk of missing the timeline stems mainly from gross margin falling below 30% or cost rigidity. For 2027 adjusted net profit to turn positive at a 4–6% net margin requires revenue of approx. RMB 7.0–8.5 billion (this report's estimate, not company disclosure). Be sober about the "incentive alignment": the profit assessment has a zero threshold (≥0 vests 100%), and was inserted as an amendment after the SSE's regulatory working letter in August — a passive patch, not an ex-ante commitment; the adjusted metric is company-defined (adding back share-based payments and acquisition-related non-cash expenses). The first validation point of the delivery path is the Q3 report (note: it is uncertain whether the company will disclose "adjusted EBITDA" quarterly; the 2026Q1 report did not include this metric — if the Q3 report omits it, verification will be deferred to the January 2027 earnings preannouncement/annual report).
Key evidence:
At the current RMB 65.6 billion market cap, consensus 2027E revenue of approx. RMB 10.5 billion implies a 2027E PS of 6.25x — between Alchip's forward (4.5x) and GUC's forward (18.3x), about 1.4x Alchip's: the market has already priced VeriSilicon's volume as a "quasi-Taiwan duo." A reverse DCF (WACC 10%) shows the current price requires 2030 revenue of approx. RMB 15.3–25.6 billion (2025–2030 CAGR of 37%–52%) with steady-state net margin of 15–18% (GUC approx. 12% in 2026H1, Alchip approx. 21% in Q2). On volume, 2027–2028 is supported by the order book, but the 2030 dimension is extrapolation; the leap in net margin from -22.7% to +15–18% has zero validation in the company's disclosure history and is the crux of the entire valuation debate. Evidence on both ends must be presented side by side: positionally, the current price sits only approx. 5% above the 52-week low of RMB 118.75 (down 66.8% from the RMB 375.67 peak) — de-bubbling is quite complete; yet PB at 22.4x remains around the 75th percentile of the past 5 years, so the "not cheap" evidence also holds. The root of the divergence from the sell-side RMB 250 target: the Street's 2026 GAAP breakeven forecast contradicts the H1 reality of -RMB 612 million, and its approx. 10x 2027E PS fails to discount VeriSilicon's disadvantage of a net margin only 1/3–1/2 that of the Taiwan duo.
Key evidence:
Q2 standalone gross margin was approx. 30.6% (this report's calculation: H1 gross profit RMB 584 million minus Q1 gross profit RMB 270 million, divided by Q2 revenue of RMB 1.028 billion) — the decline is ongoing. The SBC accounting needs to correct a common market exaggeration: 2026 SBC is approx. RMB 250–270 million (H1 RMB 133 million annualized; company's 2025 incentive plan full-year figure approx. RMB 254 million). The RMB 400–500 million magnitude will only emerge in 2027, after the new October 2026 grant (5.148 million shares) begins amortizing; the adjusted metric adds back SBC and acquisition-related non-cash expenses in full, so the gap between GAAP and adjusted will widen in 2027 — GAAP breakeven later than adjusted breakeven is a high-probability event. The company's counter-argument (most of the mass production gross profit contributes to net profit; adjusted R&D expense ratio down approx. 20pp YoY in H1) is directionally valid, but H1 net loss still widened 91% YoY — the hedge has not yet shown up at the net profit line.
Key evidence:
The other side of the overhang must be balanced: the Big Fund's first-round plan of 1.70% was only 1.51% executed and completed in April, with the second round only 0.094% — the supply peak from the state-level shareholder may be past (approx. 26.79 million shares remaining, worth approx. RMB 3.3 billion at current price, becoming post-facto-disclosed "invisible selling pressure" after falling below 5%); no concentrated unlock batches in the next 12 months; pledge ratio only approx. 0.2%; while the H-share issuance dilutes approx. 10% (rough calculation at current market cap), it also brings a net cash infusion (current net cash only approx. RMB 758 million against annualized R&D investment of RMB 1.7–1.9 billion — financing is essentially a necessity). At the executive level, only EVP Wang Zhiwei has a symbolic plan to sell 31,000 shares (0.006%).
| Metric | 2023FY | 2024FY | 2025FY | 2026Q1 | 2026H1 |
|---|---|---|---|---|---|
| Revenue (RMB 100mn) | 23.38 | 23.22 | 31.52 | 8.36 | 18.64 |
| Revenue YoY | -12.7% | -0.7% | +35.8% | +114.5% | +91.4% |
| Net profit attributable (RMB 100mn) | -2.96 | -6.01 | -5.28 | -3.41 | -6.12 |
| Non-GAAP (deducted) net profit attributable (RMB 100mn) | -3.18 | -6.43 | -6.14 | -3.35 | -6.22 |
| Adjusted net profit attributable (company basis, RMB 100mn) | Not disclosed | Not disclosed | Not disclosed | — | -4.23 |
| Blended gross margin | 44.75% | 39.86% | 34.19% | 32.29% | 31.34% |
| Net margin attributable | -12.7% | -25.9% | -16.7% | -40.8% | -32.9% |
| R&D investment (RMB 100mn / % of revenue) | 9.54 / 40.8% | 12.47 / 53.7% | 13.49 / 42.8% | 4.85 / 58.0% | 9.41 / 50.5% |
| Operating cash flow (RMB 100mn) | -0.09 | -3.46 | -2.22 | -3.08 | +6.26 |
| Free cash flow (RMB 100mn, OCF−CapEx) | -1.24 | -4.90 | -4.73 | — | +5.45 |
| Cash + trading financial assets (RMB 100mn, period end) | — | — | 29.31 | — | 24.43 (2026-06-30) |
| Interest-bearing debt (RMB 100mn, period end) | — | — | 17.17 | — | 16.84 (2026-06-30) |
| Debt-to-asset ratio (period end) | — | — | 55.7% | — | 67.7% |
| Orders in hand (RMB 100mn, period end) | 20.61 | — | 50.75 | 51.33 | 124.49 |
| Royalty revenue (RMB 100mn) | 1.10 | 1.03 | 1.11 | — | — |
| IP licensing fee revenue (RMB 100mn) | 6.55 | 6.33 | 6.71 | — | — |
Reasons for metric changes (items with YoY ≥ ±20%, with company explanations): ① 2026H1 revenue +91.4% — mass production services revenue of RMB 1.163 billion (+185.29%), design services +50.73%, data processing segment revenue of RMB 732 million (+292.17%, 39.27% of revenue), attributed by the company to AI ASIC order conversion; ② Gross margin -11.98pp — "mainly due to factors such as changes in revenue mix" (low-margin mass production ramp), with the company noting mass production's lower gross margin "but most of the gross profit generated by this business contributes to net profit"; ③ Net loss widened to -RMB 612 million — R&D expense of RMB 880 million (+43.8%, company explanation: increases in R&D staff compensation and share-based payments), SBC of RMB 133 million (vs RMB 1.96 million prior-year period), and RMB 55 million of acquisition-related non-cash expenses related to Pixelworks; ④ Operating cash flow +RMB 626 million (vs -RMB 365 million prior-year period) — cash received from product sales of RMB 3.670 billion (+252%), contract liabilities of RMB 3.157 billion (+150.8% from beginning of year, mass production prepayments); ⑤ Prepayments +135.2% to RMB 1.099 billion — "mass production business expansion, advance payments for wafers to certain mass production suppliers"; ⑥ Goodwill RMB 607 million (+241%), intangible assets RMB 911 million (+62%) — completion of the Pixelworks semiconductor acquisition in January 2026 (net cash paid RMB 884 million; purchase-date fair value assessment not yet complete, provisionally recorded on estimates).
2026H1 is a first-half report in which the "inflection in orders and revenue" and the "counter-inflection in profit and gross margin" appear simultaneously. Revenue hit a record high for a half-year period (RMB 1,864 million, +91.4%), with the volume production services segment alone contributing 62.4% of revenue; data processing (AI ASIC) revenue nearly quadrupled year-over-year — the conversion of the order book into revenue has begun to materialize, and the quality signals are strong: contract liabilities of RMB 3,157 million (+150.8%) show customers paying real cash in advance, and operating cash flow turned positive for the first time (+RMB 626 million). But the profit side faces triple pressure: gross margin declining at an accelerating pace (-11.98pp), losses widening 91% YoY, and R&D expenses up +43.8% — adjusted net loss attributable to shareholders of -RMB 423 million (excluding SBC of RMB 133 million and non-cash acquisition charges of RMB 55 million) still widened from -RMB 320 million in the prior-year period, indicating that core losses are also expanding once one-off items are stripped out. Versus market expectations: sell-side forecasts of GAAP profitability for full-year 2026 (consensus EPS of RMB 0.112) stand in stark contrast to the H1 -RMB 612 million; the -84.4% downward revision of consensus within 90 days after the interim results confirms this. The stock fell from RMB 375.67 at end-June to RMB 124.79 on October 23 (-66.8%), completing the first phase of the "Davis double kill." Next validation points: Q3 standalone quarterly revenue (2025Q3 base of RMB 1,281 million; consensus ~RMB 1.56 billion), whether gross margin can hold above 30%, and whether the order backlog maintains its ~RMB 12 billion scale.
Model in brief: an asset-light "IP licensing + one-stop chip design services" platform (SiPaaS). The IP licensing business (license fees + royalties) is high-margin (~87.8% weighted in 2025) and reusable — a classic asset-light IP business; chip design services (design NRE + volume production turnkey) is a project-based, labor-intensive "quasi-passthrough" business with outsourced foundry management, where the production phase is essentially a scale business in wafer/OSAT procurement. Revenue recurs to a low degree: royalties (quasi-recurring) accounted for only 3.5% of 2025 revenue; IP licensing + royalties combined were 24.8%, growing (+6.3%) well below the design services business — more than 3/4 of revenue is project-based and non-recurring.
Cash content of earnings: the company has been loss-making for years (cumulative attributable net losses of ~RMB 2.04 billion from 2023 to 2026H1), so the OCF/net profit ratio is only a rough reference in a loss context — OCF was -RMB 9 / -346 / -222 million in 2023–2025 (cash absorbed by advance payments for volume production inventory); FCF was negative for three consecutive years (-RMB 124 / -490 / -473 million). In 2026H1, OCF turned positive for the first time at +RMB 626 million and FCF at +RMB 545 million, but this was entirely driven by the +RMB 1,899 million expansion of contract liabilities (advance receipts) — if new orders stall, the working capital tied up in prepayments of RMB 1,099 million plus inventory of RMB 687 million could be given back; sustainability of positive cash flow remains unproven. "Recurring earnings" test: the gap between 2026H1 adjusted net loss of -RMB 423 million and GAAP -RMB 612 million is 31% (SBC and non-cash acquisition charges added back) — what is excluded is not one-off: SBC is a recurring expense incurred every year, so the adjusted basis is aggressive (see C4 for details).
Return on capital: ROIC of roughly -7.8% (2025, our estimate: NOPAT ≈ -RMB 400 million / invested capital ~RMB 5.1 billion) — deeply negative return on invested capital; the moat cannot yet be financially self-evidenced. This partly reflects full expensing of R&D (IP assets are not on the balance sheet), but RMB 2-billion-scale cumulative losses over consecutive years is a hard fact. Benchmark: ARM, which has genuinely made this model work (high royalty share), posts 20%+ net margins, while VeriSilicon's royalty revenue is only RMB 111 million — the "compounding engine" of the IP portfolio has not yet been ignited; the current growth engine is low-margin design services.
CapEx quality: CapEx/D&A of ~0.94 (2025) — heavy on R&D, light on fixed assets. 2026H1 investing cash outflow of -RMB 1.44 billion mainly reflects cash payment for the ArcPoint (Zhudian) acquisition and structured deposits; maintenance vs. growth CapEx is not broken out (intangible asset purchases — third-party IP/EDA — are the main investment), so maintenance investment cannot be determined separately.
Moat and quality red flags: the moat lies in breadth of the IP portfolio (six processor IPs + over 1,700 mixed-signal/interface IPs; NPU adopted by 91 customers in 140+ chip designs, with cumulative shipments of nearly 250 million units), full-node tape-out capability from 4nm to 250nm, and a foundry-neutral position. Red flags: ① high dual concentration of customers and suppliers (top-5 customers 45.51%; top-5 suppliers 67.72%); ② continued reliance on equity financing (IPO RMB 1,862 million + 2025 private placement RMB 1,807 million + Tiansuixinyuan strategic round RMB 570 million; cumulative undistributed profits of -RMB 3,556 million and zero dividends in six years as a listed company); ③ goodwill + acquired intangibles of RMB 1,518 million equal 51.9% of attributable net assets, while ArcPoint's H1 revenue was only RMB 32 million / net loss RMB 62 million — large impairment elasticity.
Track record of words vs. deeds: partially delivered; a preference for steering expectations with order and revenue narratives. ① The 2024 annual report pledged "R&D intensity will decline and return to normal levels" — 2025 R&D ratio of 42.8% (vs. 53.7% in 2024) delivered, but 2026H1 rebounded to 50.5% (absolute amount +47%), reversing direction again; ② the turnaround guidance (adjusted EBITDA positive in 2026H2, adjusted attributable net profit positive in 2027) is the first profitability commitment ever given, made at the 2026 interim report — in progress, pending verification; ③ the share issuance for asset acquisition planned in September 2025 was terminated in December, replaced by joint investors acquiring ArcPoint in cash — the revised plan has been executed, but the reasonableness of the consideration and structure awaits proof from results; ④ in the third vesting period of the 2022 incentive plan, the company-level vesting ratio was 0 (shares forfeited as targets were missed) — enforcement discipline linking incentives to performance genuinely exists, a positive. Order disclosures are frequent and high-profile (11 consecutive quarters of "remaining at a high level" language; new orders announced up to voluntary announcement No. 040 within the year), guidance is concentrated on orders/revenue, and profitability guidance is limited to the adjusted basis — the structure of management's narrative is itself "emphasizing revenue over profit."
Shareholder friendliness: tilted toward dilutive financing; zero returns since listing. No dividends in six years as a listed company (cumulative uncovered losses of RMB 3,556 million); the first buyback in July 2025 was a token 287,000 shares (treasury shares of RMB 24.84 million, earmarked for employee stock ownership); cumulative equity financing from IPO + placement + strategic round exceeds RMB 4.2 billion, and the pending H-share offering will add ~10% further dilution. Verdict: neutral-to-unfavorable — understandable for a growth-stage company, but shareholder returns depend entirely on order-conversion expectations.
Risk signals: the August 2026 incentive plan draft (revenue-only targets, no profit metric, no lock-up after vesting) drew an SHSE regulatory working letter; the October 19 revised draft added profit-based assessment — the governance dispute was resolved through regulatory intervention, a negative record. Actual controller Waymon Tai (Dai Weimin) also serves as chairman of associate company Siengine (persistently loss-making; cumulative equity-method losses recognized of RMB 84.25 million). No material related-party sales.
| Segment (FY2025) | Revenue (RMB 100M) | Share | Gross margin | YoY | Business logic in one sentence |
|---|---|---|---|---|---|
| Semiconductor IP licensing (license fees + royalties) | 7.82 | 24.8% | ~87.8% (weighted estimate) | +6.3% | Develop once, sell many times — an IP reuse business; the margin machine |
| Chip design services (NRE) | 8.77 | 27.8% | 14.22% (+1.35pp) | +20.9% | Project-based labor business; 96 projects in execution, AI computing revenue share 41% |
| Volume production services (turnkey) | 14.90 | 47.3% | 18.14% (-1.15pp) | +74.0% | Outsourced foundry management at scale; the volume driver backed by an order backlog over RMB 10 billion |
Profit-contributing segments: gross profit contribution estimated as "revenue × gross margin" — IP licensing ~RMB 687 million (~64%), volume production ~RMB 270 million (~25%), design services ~RMB 125 million (~11%) — IP licensing, only 1/4 of revenue, contributes nearly 2/3 of gross profit and is the true profit engine; volume production is a revenue amplifier, not a profit engine. In 2026H1 the mix tilted further toward production (62.4% of revenue), which is precisely the mechanical reason for the accelerating decline in blended gross margin. The gross margin structure gap (87.8% vs. 14.2%, a 73.6pp spread) must be highlighted: IP licensing sells reusable intellectual property (near-zero marginal cost, ecosystem stickiness); design services sell labor hours (highly competitive, weak pricing power); volume production is essentially a procurement passthrough (earning a management fee spread) — three entirely different businesses, and the revenue mix change is the entire key to the gross margin change. 114 chips were in volume production shipment (2025); 109 in 2026H1 with another 59 design projects awaiting production — the production pipeline remains full.
Accounting red flags (by severity): ① triple-jump in R&D capitalization rate — 0% (2024) → 2.67% (2025) → 6.46% (2026H1); development expenditures rose from zero to RMB 97 million, while in the annual report's "explanation of significant changes in capitalization" the company checked "not applicable" (i.e., no explanation) — rising capitalization during an expanding loss period suggests expense flattery (medium); ② provisional goodwill — the RMB 429 million goodwill from the ArcPoint acquisition is based on provisional values with the PPA incomplete; risk of retrospective adjustment and impairment (medium); ③ under-provisioned inventory — year-end 2025 inventory of RMB 505 million carried only ~1.3% provision for obsolescence; custom chips iterate quickly, and inventory rose to RMB 687 million by 2026H1 — large impairment elasticity (medium); ④ equity-in-name-debt-in-substance structure — the RMB 570 million investor contributions in Tiansuixinyuan are presented as liabilities (with an obligation to repurchase in cash at principal plus a fixed rate) rather than equity, pushing the debt-to-asset ratio to 67.7% (low); ⑤ chip design revenue recognized over time under the input method (2024 key audit matter), with progress dependent on management estimates (low).
Cross-period consistency: ① gross margin down four consecutive periods (44.75% → 31.34%) — consistent with management's "revenue mix shift" explanation, but volume production's own gross margin also fell 1.15pp, so the mix explanation does not fully cover it; ② 2024 annual report said "R&D intensity will decline" vs. 2026H1 R&D +43.8% and adjusted EBITDA losses widening YoY — inconsistent; ③ the triple-jump in R&D capitalization — unexplained by the company; ④ IP license count disclosure (134 times in 2023 → 216 in 2024) was discontinued from 2025, precisely as IP licensing revenue growth slowed to +6.07% — unexplained; the timing of the metric's discontinuation coinciding with the business slowdown is concerning. Overall judgment: no sign of systematic accounting fraud (revenue is corroborated by contract liabilities and cash flow), but clear traces of "opportunistic adjustments" in expenses and disclosure practices; cross-period quality is medium-to-weak.
Current market data and percentiles: price RMB 124.79 (2026-10-23 close, frozen anchor), market cap RMB 65,629 million, 526 million shares outstanding. PS(TTM) 16.24x (TTM revenue RMB 4,042 million); PE(TTM) negative (consecutive losses; PE and its percentile not applicable and not used as a headline — valuation must be done on PS/forward bases); PB 22.43x (1/3/5-year percentiles of 15%/58%/75% — consecutive losses eroding net assets distort the PB denominator; the 75% percentile overstates how "expensive" it is, reference only). Market cap sits in the lowest 2% of its 1-year distribution (substantial deleveraging after the bubble burst). Liquidity ample (single-day turnover can reach ~5% of market cap); no liquidity risk warning needed.
Peer comparison:
| Company | PS(TTM) | Forward PS | 2026 growth | Net margin | Notes |
|---|---|---|---|---|---|
| VeriSilicon | 16.2x | 2027E 6.25x | H1 +91% | adjusted -22.7% | #8 IP globally + #4 design services |
| Alchip-KY | ~13.0x | 2026E ~4.5x | 9M26 +49% | Q2 ~21% | AWS-affiliated AI ASIC, TSMC-tied |
| GUC | ~21.9x | ~18.3x | Jan–Aug +104% | ~12.6% | TSMC-owned, 3nm ASIC leader |
| ARM | ~43.5x | — | FY26 +22% | 20%+ | Pure IP + royalty ceiling reference |
| CEVA | ~6.8x | — | 2025 +2.5% | GAAP loss | A stalled IP vendor only merits single-digit PS |
(Peer multiples estimated in this report from TWSE/public market caps and revenue as of 2026-10-22/23; Taiwan-listed names computed in NT dollars against themselves, no cross-currency issue)
Market-implied expectations: the current RMB 65.6 billion market cap implies 2027E PS of 6.25x (denominator: consensus revenue ~RMB 10.5 billion). Reverse DCF (WACC 10%, exit PE 25–35x × steady-state net margin 15–18%) solves for: the current price requires 2030 revenue of RMB 15.3–25.6 billion (2025–2030 CAGR 37–52%). In one sentence: the current price requires revenue of RMB 15–26 billion in four years with net margins climbing to Taiwan's twin-leader levels, while reality is TTM revenue of RMB 4 billion, adjusted net margin of -22.7%, and gross margin still declining. Two-way stress test via demand-chain math: the volume side (order book of RMB 12.4 billion on hand + RMB 15.1 billion in new signings, with 9–12 month design / 6–18 month production conversion) solidly supports the 2026–2028 revenue path — implied growth is backed by real orders, not castles in the air; the price side (net margin leapfrog) depends entirely on an unproven jump in operating leverage from -22.7% to +15~18% over the next four years — what is overpriced in the implied expectation is the price, not the volume.
Three-tier value (EPV): asset value ~RMB 5.56/share (attributable net assets RMB 2,925 million; IP portfolio fully expensed, replacement value off balance sheet — a conservative floor). EPV — discounting the current run-rate (2026H1 annualized adjusted attributable net loss ~RMB 850 million, ~-RMB 1.61/share) at zero growth gives a negative value (~-RMB 16/share); the mechanical result is meaningless, but the real implication is: on a zero-growth basis the company has no cash flow value, and ~100% of the current price is a growth option (EPV negative; growth option share exceeds 100% of the price) — the "market prices zero growth" framework does not apply here at all; whether growth deserves payment is answered by the implied expectation math above. Net cash of RMB 758 million (RMB 1.44/share) is already included in the net asset figure.
Three scenarios and odds (12-month horizon):
| Scenario | Probability | Fair range | vs. current price | Core assumptions |
|---|---|---|---|---|
| Bear | 30% | RMB 30–50 | -60% ~ -76% | AI capex digestion in 2027 + order conversion halved + profitability missed; 2027E revenue RMB 5.5–6.5bn × PS 3–4x (benchmark: Alchip's lost-order year and GUC current); floor ≈ net assets ~RMB 5.6/share + IP replacement value |
| Base | 50% | RMB 115–150 | -8% ~ +20% | Orders convert at 55–65%; 2027E revenue ~RMB 10bn (+59%), adjusted net profit turns positive at +RMB 0.3–0.6bn; 2028E net profit RMB 1.1–1.3bn × PE 50–65x cross-checks consistently (RMB 55–84.5bn market cap) |
| Bull | 20% | RMB 190–260 | +52% ~ +108% | Edge AI (Google Coral NPU glasses, etc.) ramp + royalty takeoff; 2027E revenue RMB 12.5bn × PS 8–11x; 2028E net profit RMB 2.2–2.8bn |
Ramp math (explicit path for base/bull): RMB 12.449bn backlog + RMB 15.142bn new signings recognized in 2027 at a 55–65% conversion rate → 2027E revenue ~RMB 10bn (1.9x the RMB 5.327bn incentive-plan revenue threshold, ample margin of error) → adjusted net margin 4–6% → 2027 adjusted net profit RMB 0.4–0.6bn → 2028 revenue ~RMB 14bn, net profit RMB 1.1–1.3bn. Exit multiple anchors: PS set at the low-to-mid 6–8x band between Alchip's forward 4.5x and GUC's TTM 21.9x (VeriSilicon's growth exceeds Alchip's but net margin is only 1/3–1/2 of the two leaders, with higher customer concentration — not granted GUC-tier multiples); PE set at Alchip's current 51x tier (50–65x); the two-method cross-range of RMB 55–84.5bn is consistent with the PS method's RMB 60–80bn. Anchoring discipline note: we do not use VeriSilicon's own current 16.2x TTM PS as the base anchor (circular reasoning); current multiples serve only as reference. The bear scenario is an independent stress construction: AI capex digestion (Broadcom/Taiwan 2027 schedules falling) → order cancellations/deferrals of 25%+ (the company's interim report itself acknowledges "the risk that service contracts may be terminated earlier than expected or payment deferred") → 2027 revenue of RMB 5.5–6.5bn, profitability missed to 2028, priced at PS 3–4x (GUC current and Alchip's lost-order year levels). External comparison coordinates: the most "pessimistic" sell-side action was Capital Securities' September 29 target price cut to RMB 250 (still Buy); no sell ratings exist on the street — the bear case of RMB 30–50 is far below all sell-side targets; the difference lies in our refusal to extend credit for a net-margin leapfrog and our view that street targets do not reflect the 90-day EPS revision trend of -84.4%; the frameworks (12-month PS scenario vs. sell-side 10x 2027 PS) are not directly comparable.
Our own earnings forecast (reconciliation anchor): FY2026 revenue RMB 5.8–6.6bn (+84%~109%), GAAP attributable -RMB 0.9~0.6bn, adjusted -RMB 0.6~0.3bn (adjusted EBITDA ~RMB 0 to +0.2bn) — significantly below sell-side 2026 GAAP breakeven forecasts; FY2027 revenue RMB 9.0–11.0bn (+55%~70%), adjusted attributable +RMB 0.3–0.6bn (net margin 4–6%), GAAP attributable -RMB 0.2 to +0.2bn (SBC rising to RMB 0.4–0.5bn as a drag). Driving assumptions: order conversion 55–65%, volume production gross margin stabilized at 25–30%, blended gross margin 32–35%, adjusted R&D expense ratio down a further ~10pp, royalty contribution still <5%.
Conclusion: fairly priced, no margin of safety. Base-case fair value of RMB 115–150; the current price of RMB 124.79 sits within the range, toward the lower end (~-6% vs. the RMB 132.5 midpoint); probability-weighted fair value ~RMB 123 (~-1%). Assessing quality and price separately: the quality inflection (orders, customers, IP platform reuse) is real and well-evidenced, but the price has been paid in advance — at the current price one buys "a fair price for certain growth + a naked option on margin delivery," with neither an undervaluation cushion nor a June-high-style bubble premium. If 2026H2 adjusted EBITDA confirms breakeven and gross margin stabilizes, the base case can be revised up; if Q3 gross margin breaks below 30% or the order backlog declines sequentially, switch to the bear path.
Market Size: The global semiconductor design IP market reached USD 8.49 billion in 2024 (per IPnest's Design IP Report, including license + royalty), up +20.2% YoY—the highest growth rate ever recorded (prior years were +5–10%), and approximately USD 9.45 billion in 2025 (IPnest/SemiWiki, to be verified). The structural engine is AI interconnect—wired interface IP grew +23.5% in 2024 and processor IP +22.4%. Looking forward: the aggregator MarketsandMarkets projects USD 9.3 billion in 2025 → USD 18.64 billion by 2032 (CAGR 10.2%, cross-referenced estimates). No authoritative total market size exists for chip customization services; first-tier proxies: Marvell's April 2025 investor day projected the data center custom compute TAM at approximately USD 55 billion by 2028; Broadcom management guided in December 2024 to a USD 60–90 billion AI serviceable market by FY2027. Edge AI: IBS projects the edge AI market growing from USD 130.9 billion in 2025 to USD 709.3 billion by 2035. No independent authoritative figure exists for China's IP market (not obtained); proxy indicators: IBS estimates China's semiconductor market at USD 274.4 billion in 2025 with domestic companies holding 38.2% share, rising to 63% by 2030—the addressable space for domestic IP/customization rises with local share.
Quantified Chain of the Demand Inflection Point (the structural inflection this company is riding): Driving variables—hyperscalers replacing general-purpose GPUs with self-developed AI ASICs (Broadcom's AI quarterly revenue of approximately USD 17 billion, Alchip's revenue for the first 8 months of 2026 up +103.8%, and GUC's September single-month +496% are already-realized industry evidence) + leapfrog growth in edge AI per-device IP content (NPU + AI-ISP + GPU multi-IP combinations replacing single cores). Penetration pace—Wayne Dai (April 2026 earnings call) expects ASICs to account for 40% of AI server chip shipments in 2026; per IBS, AI semiconductors will represent 77.7% of the total by 2035. Company share—VeriSilholds the global No. 4 position in full-stack customization (CIC, 2025), No. 8 globally / No. 1 in mainland China in IP revenue (IPnest 2024); the hardest verifiable chain at the company level is new orders: RMB 15.142 billion for January–August 2026 (254% of full-year 2025, 90% AI compute). The industry level (drivers × penetration) is quantified; the company level (IP value per chip × share) lacks third-party public hard data—order disclosures serve as the proxy variable, which is the most honest description of this chain currently available.
Industry Chain and Value Distribution: Along the EDA/IP → chip design → wafer manufacturing → packaging & testing chain, VeriSilis positioned at the design front end (IP + design services), with mass production outsourced via foundry management. Weak bargaining power over upstream players: the EDA Big Three (Synopsys/Cadence/Siemens—the May 2025 China sales restrictions and July reversal demonstrated single-point failure risk), and advanced-node capacity is a seller's market (VeriSil's "foundry-neutral" strategy plus capacity resource pool hedges this, but its bargaining power is weaker than Broadcom-tier large customers). Downstream: strong bargaining power over fragmented chip design companies, weak over hyperscalers and other super-customers. Gross margin primarily accrues at the IP licensing stage (90%-plus), design services are mid-range, and mass-production turnkey is lowest—VeriSil's 2026H1 blended gross margin of 31.34% reflects the mix of these three lines; revenue scale and customer stickiness reside in mass production, while incremental value resides in IP. In 2025, revenue from "non-chip customers" (system vendors/internet/cloud/automotive) was RMB 1.289 billion, or 40% of the total—the downstream customer base has migrated from chip companies toward system/cloud vendors, which is precisely the customer profile of the AI ASIC inflection.
Supply-Demand and Competitive Structure: Demand side is driven by four wheels (cloud ASICs, edge AI, autonomous driving, and system vendors' in-house chip wave); supply side—IP expansion depends on talent rather than capacity, while the customization services bottleneck lies in advanced-node tape-out experience and headcount. Concentration: the top four in design IP in 2024 (Arm, Synopsys, Cadence, Alphawave) held a combined 75% (72% in 2023, per IPnest), with Arm alone >40%—an oligopolistic structure that is still consolidating; VeriSilis the leader of the long tail, with global share of approximately 1.3% based on 2025 IP revenue of RMB 782 million (approx. USD 109 million). No unified market share table exists for customization services; CIC ranks VeriSilglobal No. 4 in full-stack customization (the top three being Taiwan-based players of the Alchip, GUC, and Faraday tier). Entry barriers: silicon validation and software ecosystem accumulation, automotive ISO 26262 qualification, 4nm-class tape-out experience (only a handful of service providers globally have done it), R&D talent density (VeriSilhas 1,883 R&D staff, 87% with master's degrees or above), customer architecture lock-in and switching costs, and UCIe/Chiplet standard positioning. Competitive intensity: on the IP side, VeriSilcompetes in differentiated categories against Arm/CEVA (with direct head-on competition in edge NPUs); on the customization side, the Taiwan Big Three, backed by TSMC capacity, have created an arms race, while mainland peer Britechip (SMIC-affiliated) attacks from both sides with low prices plus supplementary IP (launching PCIe 5.0 IP in September 2026)—price war risk is currently lower than the risk of capacity and talent competition, but Britechip's 2025 mass-production halving and swing to loss demonstrated the fragility of second-tier custom providers. Substitution threats: RISC-V open source is a threat to Arm but a platform opportunity for VeriSil(chairing unit of the China RISC-V Industry Alliance); large customers building in-house design teams pose long-term squeeze.
Cyclicality and Regulation: Semiconductors are in the mid-stage of an upcycle (WSTS: global +22.5% to USD 772 billion in 2025, projected +26.3% to USD 975 billion in 2026) plus the early-to-mid stage of a structural AI ASIC boom. Leading indicators: VeriSil's order backlog, North American hyperscaler capex, Broadcom/Marvell quarterly AI revenue, Arm royalty growth (FQ3 FY26 +27%, a record), and TSMC's HPC revenue share. Regulation cuts both ways: China-side tailwinds—the "AI+" action plan supports AI chips, a national RISC-V policy is in the works (Reuters, March 2025), and the Big Fund is a shareholder; US-side risks—BIS rounds of advanced computing controls, repeated EDA export restriction reversals, and TSMC tightening order acceptance for Chinese-background design companies on advanced nodes. VeriSil's interim report itself acknowledges Entity List contagion clauses (VeriSilHoldings/VeriSilMicroelectronics US cannot sell US-patented-technology products to listed Chinese customers). If US-China tensions escalate to the IP/licensing level, overseas customers (Samsung/Google/Amazon, etc.) businesses also face retaliation risk.
Company Industry Positioning: Absolute domestic leader + global niche challenger. Share metrics in parallel: per the IPnest figures cited by the company, IP licensing revenue ranks No. 1 in mainland China and No. 8 globally (royalty No. 6 globally; among the top 10 IP companies globally, top two in number of IP categories); per CIC, No. 4 globally among full-stack custom solution providers—neither metric is self-proclaimed by VeriSilbut both are third-party commissioned/industry figures; the difference lies in statistical scope (pure IP revenue vs. full-stack including customization), and the trend is consistently upward (2025 revenue +35.8%, 2026H1 +91.4%, new orders +254%). Share trend: rising (order intensity shows AI customization positioning materially improving). Moat sources: breadth of IP portfolio (1,700+ mixed-signal/interface IPs), RISC-V ecosystem positioning + Google Coral NPU partnership, foundry neutrality, and SiPaaS full-flow customer journey lock-in. Weaknesses: not yet profitable, rising mass-production mix dragging gross margins, and 83% of orders concentrated in the single AI data processing track.
Peer Benchmarking (data as of October 2026, each company's own disclosure basis):
| Company | Revenue Scale | Growth | Gross Margin | Profitability | Key Difference vs. VeriSil |
|---|---|---|---|---|---|
| VeriSilMicroelectronics | 2025: RMB 3.152 bn / 2026H1: RMB 1.864 bn | H1 +91% | 31.3% (H1) | Net loss attributable: RMB 612 mn (H1) | — |
| Arm (US) | FY2026: approx. USD 4.92 bn | +22% | 90%-plus (IP vendor) | Profitable (net margin 20%+) | Pure IP + royalty; VeriSilis 1/13 its revenue scale and unprofitable; yet Google chose VeriSilfor Coral NPU |
| Alchip-KY (3661.TW) | 9M26: approx. NT$38.0 bn | 9M +49% | H1 40.3% | Q2 net margin 21.5% | AWS-affiliated 3nm mass-production scale far exceeds VeriSil; 2025 loss of a -40% order demonstrated project-based volatility |
| GUC (3443.TW) | Jan–Aug 2026: NT$36.95 bn | +104% | Not obtained (net margin approx. 12.6%) | H1 net profit NT$3.20 bn | TSMC equity stake, "favorite son" with exclusive capacity access; VeriSil's foundry neutrality trades exclusive capacity for customer diversity |
| Britechip (688691.SH) | 2025: RMB 724 mn | -33.5% | H1 15.7% | 2025 swing to loss: -RMB 110 mn | Direct mainland peer, SMIC-affiliated, no proprietary IP—a control group proving the genuine premium of VeriSil's "IP synergy + AI order structure" (gross margin gap approx. 16pp) |
The generic financial reporting framework for the semiconductor IP/customization industry misses four make-or-break metrics, listed separately here:
Overall Assessment: Neutral (12-month horizon, confidence 0.50). VeriSil's quality inflection (AI ASIC order book, IP platform reuse, customer structure upgrade) is well evidenced, but the current price of RMB 124.79 already prices in the baseline success path (baseline fair value RMB 115–150, probability-weighted fair value approx. RMB 123, no margin of safety). Upside depends on the realization of a "margin leap"—an assumption with zero historical validation—while the downside carries genuine tails from AI capex digestion and order concentration (RMB 30–50). Strategic implication: existing holders may hold and wait for validation with stop-loss on breakdown; non-holders should not chase, but wait for one of two signals to decide direction—profitability validation (Q3 report gross margin stabilizing above 30% + order backlog holding at the RMB 12 bn level → shift to upper baseline/bull range) or falsification signal (gross margin breaking 30% / orders declining sequentially → shift to bear path). Acknowledging judgment boundaries: the current price sits in a high-volatility zone roughly 5% above the 52-week low, and two imminent events—the Q3 report (within days) and H-share pricing—could rapidly unfold either scenario; the non-independence of scenario probabilities to sector beta is reflected in the bear case probability being raised to 30%.
Key Monitoring Points (Next 12 Months):
Risks and Falsification Conditions (Written Summary): ① Order backlog falling below RMB 10 bn at end-September or a marked decline in the mass-production order share—the earliest alarm of order-logic breakdown; ② Blended gross margin breaking 28% in a single quarter, or the 2026 annual report confirming that H2 adjusted EBITDA did not turn positive—the turnaround timetable is missed, switching to bear-case pricing; ③ 2027 new orders turning negative YoY—AI capex digestion confirmed; ④ EDA supply cut exceeding one quarter / company added to the Entity List—a supply-side single point of failure (low probability, high damage); ⑤ H-share issuance at a discount exceeding 30%—dilution beyond expectations and a downward shift of the A-share valuation anchor; ⑥ Pixelworks' annual report recognizing a large goodwill impairment (goodwill of RMB 607 mn vs. Pixelworks H1 revenue of only RMB 32 mn); ⑦ Upside trigger: share price rising above RMB 150 (upper bound of baseline range) while profitability validation has not materialized—a signal of margin-of-safety exhaustion; position-reduction discipline is as important as stop-loss.
Data and Basis Notes: Financial data are from VeriSilMicroelectronics' 2023–2025 annual reports and 2026 Q1/interim reports (cninfo.com.cn); the price anchor is the 2026-10-23 closing price of RMB 124.79 (frozen throughout this report); peer multiples are based on TWSE 2026-10-22 closings and each company's public disclosures; consensus estimates are from investing.com/moomoo aggregations (2026-10-22/23); all figures are company-disclosed originals except those marked "calculated in this report." This report does not constitute investment advice.