Innoscience

All figures are RMB millions under IFRS for years ended 31 December, unless stated. FY2021–FY2023 come from the HKEX prospectus. FY2024 and FY2025 come from the annual reports.

Income statement

The FY2021 growth cell reads 0 because there is no FY2020 base.

Adjusted EBITDA is the operating loss plus depreciation of property, plant and equipment and right-of-use assets, amortisation of intangibles and land-use rights, equity-settled share-based payments, and non-recurring compliance-related costs.

Revenue CAGR. Revenue compounded at about 105% a year from FY2021 to FY2025 and about 43% a year from FY2023 to FY2025.

Commentary. The income statement shows a capital-heavy IDM moving from start-up losses toward break-even.

FY2021–FY2022 losses: Net losses of RMB3.40 billion and RMB2.21 billion were mainly non-operating. Charges of about RMB2.30 billion in FY2021 and RMB0.90 billion in FY2022 reflect fair-value changes on investor instruments with redemption rights. These charges stopped once the instruments were reclassified to equity, which is why reported equity went from −RMB4.1 billion at end-FY2021 to +RMB3.0 billion at end-FY2022.

Operating losses: Operating losses have stayed around RMB0.8–1.2 billion a year. The constant drag is about RMB0.5 billion of annual D&A on fabs built in 2017–2021, together with heavy R&D (RMB662 million in FY2021) and a step-up in SG&A after the IPO. SG&A was RMB549 million in FY2024, including listing, legal and share-based costs.

Gross margin: The inflection came from volume absorption. Gross margin improved from −289% in FY2022 to +7.3% in FY2025 while cost of sales grew only 13.6% in FY2025 against revenue growth of 46%.

Expenses: Management describes period expenses as broadly stable in FY2025, so most of the gross-profit gain fell through to EBITDA.

Income tax: Tax is negligible (RMB0.34 million in FY2024), so pre-tax and net losses are essentially the same.

Half-year progression

1H2026 was the third straight half-year of gross-margin improvement. The net loss fell 28% year on year to RMB309 million, and basic loss per share was RMB0.34. Operating cash flow was +RMB169 million, against −RMB387 million in 1H2025. Total shipments in 1H2026 were 547 million units, taking cumulative shipments past 2.5 billion.

Balance sheet

There is no goodwill, because the company has not made acquisitions.

Commentary.

FY2025 balance sheet: The FY2025 balance sheet was the strongest in the company's history. Two placings raised about HK$2.09 billion (about RMB1.87 billion), and equity rose to RMB4.21 billion.

Cash versus liquid resources: Reported cash fell to RMB879 million while total current assets almost doubled to RMB4.33 billion. A large part of the placing proceeds sits in short-term deposits and treasury-type investments, and inventory was built up (inventories were up 60.8% in 1H2025). Including short-term investments, liquid resources were roughly RMB2.6 billion at year-end, more than total borrowings of RMB2.23 billion.

Gearing: The company reports net gearing of 32.1% for FY2025 (13.4% in FY2024). The current ratio improved to 5.3x from 2.5x.

Debt mix: Borrowings shifted toward long-term facilities. The current portion fell to RMB336 million, which reduces refinancing risk.

Fixed assets: PP&E has fallen every year since FY2021 because depreciation (about RMB500 million a year) outran maintenance-level capex. This will reverse as the capacity expansion funded by the October 2025 placing (HK$482 million earmarked) is spent.

Cash flow

Commentary.

FY2025 operating cash flow: The outflow worsened to −RMB806 million even though EBITDA improved. The cause was working capital: inventory build ahead of data-centre, automotive and appliance ramps, plus receivables growing with revenue.

1H2026 reversal: Operating cash flow turned positive at +RMB169 million, the first positive half-year. This suggests the FY2025 build was timing rather than structural.

Capex: Capex has been low (RMB172–332 million a year in FY2023–FY2025) because the fabs were largely built by FY2021. Management says future expansion depends mainly on adding bottleneck tools such as MOCVD reactors rather than new cleanrooms, so capex per unit of new capacity should fall.

Share purchases: Shares bought by the share-award trustee are held in trust for employees and are not cancellation buybacks.

Key ratios

The current ratio was 2.5x at end-FY2024, 3.21x at 30 June 2025 and 5.3x at end-FY2025. EBIT interest cover was negative: −8.5x in FY2023 and −11.7x in FY2024. Net debt/EBITDA is not meaningful while reported EBITDA is negative. The company does not report a cash conversion cycle. ROIC is deeply negative in every year because NOPAT has been negative throughout. In 1H2026, the annualised return on equity was about −15% (−7.65% for the half-year).

Trend summary. Every profitability ratio has improved for three years in a row, and the improvement sped up in 1H2026. Asset turnover (0.19x in FY2025) is still well below mature power-semiconductor IDMs, which typically run 0.5–0.8x. Further improvement depends on filling capacity rather than adding it.

Management guidance

Management gives no numeric revenue or earnings guidance. It points to:

• further gross-margin gains from a higher-value mix and scale;

• capacity expansion to 20,000 wafers per month, with a longer-term aspiration of 70,000;

• continued growth in AI data-centre, automotive, industrial and energy-storage, appliance and robotics revenue;

• deeper work with ST, onsemi, NVIDIA and Google.

The 1H2026 results statement says the company has built a clear lead in GaN manufacturing scale, cost, customer coverage and supply chain. It says future growth will increasingly come from higher-quality, higher-margin revenue.

Sell-side forecasts

Revenue, Sinolink (August 2025): FY2025: 1357; FY2026: 2270; FY2027: 3749; FY2028: 0

Net income, Sinolink (August 2025): FY2025: -803; FY2026: -255; FY2027: 291; FY2028: 0

Revenue, Soochow Securities International (July 2026): FY2025: 1213; FY2026: 2200; FY2027: 3500; FY2028: 7000

FY2025 is reported actual revenue for Soochow. Sinolink published no FY2028 estimate, so its FY2028 cells read 0. Both brokers put net break-even at around FY2027.

Consensus expects revenue growth of 60–80% in FY2026, with net break-even around FY2027. Actual FY2025 revenue (RMB1,213 million) came in about 11% below Sinolink's August 2025 estimate. Meeting the FY2026 estimates would require second-half revenue of about RMB1.37–1.44 billion, roughly 65–72% above 1H2026.

Bull case

1. AI data-centre power becomes the main growth engine. Shipments into AI and data centres rose 180% in 1H2025 and 183% in 1H2026. Innoscience is the only Chinese company in NVIDIA's 800 VDC ecosystem and has Google design-ins. Commercial deployment of 800 VDC from 2027 could add a high-margin revenue stream large enough to change the company's earnings profile.

2. Operating leverage is visible. Gross margin has improved for three straight half-years to 11.6%. Adjusted EBITDA turned positive in FY2025 and operating cash flow in 1H2026. D&A on the original fabs is expected to run off in 2026–2027, and expansion capex per wafer is falling. Each extra point of utilisation should convert into profit at a high rate.

3. Becoming the merchant GaN wafer supplier for Western IDMs. With ST using its capacity in China and onsemi planning wafer procurement, Innoscience's 8-inch capacity could become a shared industry resource, especially as TSMC leaves GaN foundry work. This would raise utilisation and secure a cost leadership position that is hard to copy.

Bear case

1. IP losses shut out the highest-value Western markets. Infineon has won four rulings (two Munich judgments in 2025 and June 2026, a separate June 2026 utility-model ruling, and the May 2026 ITC determination), and more proceedings are pending. Innoscience's redesigns have worked so far, but each new adverse ruling raises costs and customer qualification risk in exactly the data-centre and automotive markets the bull case depends on.

2. Profitability is still far off. The FY2025 net loss of RMB840.5 million and the 1H2026 loss of RMB309 million compare with a 7–12% gross margin, well below the 38–55% of Western peers. Funding a move to 70,000 wafers per month while still losing money risks further dilution, after two placings in 2025.

3. Geopolitical and concentration risk. Over 90% of revenue is booked in China, and US market access depends on "compliant" arrangements and redesigns. Any US policy move against Chinese power semiconductors in hyperscale data centres would hit the main growth driver.

Key catalysts and monitorables, next 12 months

FY2026 annual results: Expected timing: Late March 2027; What to watch: Whether full-year revenue reaches the RMB2.2–2.3 billion consensus; full-year gross margin above 12%; whether positive operating cash flow holds

NVIDIA 800 VDC deployment: Expected timing: 2027 ramp; design activity through 2026–2027; What to watch: Conversion of partner status into volume orders

onsemi MoU to definitive agreement: Expected timing: 2026–2027; What to watch: Wafer-supply contract terms and volumes

Federal Circuit decision on the EPC ITC appeal: Expected timing: Pending; What to watch: Removal of the exclusion order on legacy products

• Infineon US district case and further Munich proceedings; German appeals — Expected timing: Pending; What to watch: Scope of injunctions and damages

USPTO review of the remaining Infineon '481 claims: Expected timing: Pending; What to watch: Possible invalidation of the remaining asserted claims

Capacity expansion progress: Expected timing: Ongoing; What to watch: Additions toward 20,000 wafers per month; capex intensity

Automotive design wins: Expected timing: Ongoing; What to watch: OBC and LiDAR volume; Tier-1 programmes through the UAES joint lab

Share-award trustee purchases: Expected timing: Ongoing; What to watch: Further on-market buying

1H2027 interim results: Expected timing: August 2027; What to watch: Path to net break-even

Analyst verdict

Innoscience is the most important pure-play GaN power company in the world by volume, capacity and revenue. In 2025–2026 it passed the operating milestones that separate a well-funded start-up from a scaling semiconductor business: positive gross margin, positive adjusted EBITDA and positive half-year operating cash flow. Revenue rose 46% in FY2025 and 51% in 1H2026. The mix is shifting quickly toward AI data-centre, automotive, appliance and industrial uses, which made up about 64% of first-half revenue. Its 8-inch cost position looks increasingly strategic now that TSMC is leaving GaN foundry work and Western IDMs (ST and onsemi) have chosen to partner with it.

The investment case remains contested for three reasons. First, the company is still deeply loss-making: its RMB721 million trailing-12-month net loss is about half its revenue, and gross margin is less than a quarter of Western peers'. Second, the Infineon litigation has gone against Innoscience in Germany and, on legacy designs, at the ITC. Its redesigns and the Suzhou counter-ruling provide only partial protection. Third, the AI data-centre opportunity that drove the 2025 re-rating is real but early. Management itself says the business model, whether selling through PSU vendors or directly to cloud providers, is still being worked out, and 800 VDC volume deployment is a 2027 event.

At HK$32.58 the shares trade about 67% below their September 2025 peak, close to the IPO price and at roughly 18–19x trailing sales. Much of the 2025 enthusiasm has gone out of the price, and every disclosed broker target is more than double the current level. On balance, Innoscience is a high-variance, execution-driven turnaround with real strategic assets. The case for the shares depends on FY2026 revenue landing near RMB2.2 billion, gross margin passing the mid-teens, and IP disputes staying limited to legacy products. If those hold, the operating leverage visible in the half-year data supports a path to break-even around FY2027. If the data-centre ramp slips or Western litigation spreads to current-generation products, losses and dilution are likely to last longer than consensus expects.