International investment banks unanimously favor Innoscience.

I. Innoscience (02577.HK) Trading Rationale: 1. Dongzi Securities: Within the past 90 days, a total of 10 institutions have issued ratings for this stock—6 “Buy” and 4 “Add.” The average target price set by these institutions over the past 90 days is HK$49, representing a 29% premium over the current share price. According to Dongzi Securities’ big-data analysis, the company has received investments from STMicroelectronics and Jiangsu Provincial State-owned Capital Fund. (Date: April 28, 2025) 2. CLSA: Financially, the company has incurred losses over the past several years; however, with growing market demand and expanded production capacity, profitability is expected by 2027. In 2023, the company achieved revenue of RMB 593 million; this is projected to reach RMB 8.01 billion by 2028, representing a compound annual growth rate (CAGR) of 76%. Gross margin stood at −61.1% in 2023 but is forecast to improve to 48% by 2028. Operating expense ratio is also expected to decline gradually as economies of scale take effect. Innoscience holds a leading position in the gallium nitride (GaN) power semiconductor market. Demand for GaN is surging rapidly, driven by growth in electric vehicles (EVs), AI servers, and renewable energy sectors. CLSA assigns the company an “Outperform” rating. (Source: Reuters, cn.tradingview.com/news… Date: April 28, 2025) 3. China Merchants Securities: Benefiting from rising GaN application penetration and market expansion, the company’s revenue grew from RMB 68 million in 2021 to RMB 828 million in 2024, and is projected to reach RMB 3.098 billion by 2027—a CAGR of 55.2% between 2024 and 2027. As business scales up, Innoscience’s gross margin and net profit margin are expected to improve progressively: gross margin is forecast to turn positive in 2025, and net profit margin in 2027. Based on a 30x P/E multiple for 2030 (the peer group’s five-year average), with a weighted average cost of capital (WACC) of 12.0% and an RMB/HKD exchange rate of 1.07, the firm assigns an “Overweight” rating. (Source: Reuters, cn.tradingview.com/news… Date: April 28, 2025) II. Powtech New Materials (300811.SZ) Trading Rationale: 1. Dongzi Securities: Within the past 10 days, four institutions have issued ratings for this stock—two “Buy” and two “Recommended.” The average target price set over the past 10 days is RMB 52, representing a 30% premium over the current share price. Dongzi Securities’ big-data tracking shows that the stock has seen three consecutive days of net inflows from major institutional investors. (Date: April 28, 2025) 2. AVIC Securities: The company’s core business—metal powder core—serves the new-energy sector. By strategically extending its footprint upstream into chip inductors for computing infrastructure, Powtech has established business ties with globally leading GPU chipmakers, further cementing its industry reputation as an “expert in metallic soft magnetic materials and applications.” Its second growth curve has clearly taken shape, positioning it to fully capture the high-growth dividends of the AI era.

A company losing RMB 800 million annually commands a market capitalization of RMB 56 billion: Gallium nitride (GaN), the ‘hidden protagonist’ ignited by AI power supplies.

On May 7, 2026, the U.S. International Trade Commission (ITC) issued a final determination that Innoscience infringed Infineon’s U.S. Patent No. 9,899,481 (claims 4 and 17), and accordingly imposed a Limited Exclusion Order and Cease-and-Desist Order. This means Innoscience’s infringing products are barred from entering the U.S. market. This risk has been markedly underestimated by the market—NVIDIA is one of Innoscience’s largest customers, and NVIDIA’s data centers are heavily concentrated in the United States. The ITC order may compel Innoscience to undertake costly and time-consuming design-around efforts; during this period, orders could be lost. [2] The signal investors should monitor most closely: whether Innoscience files for design-around certification in Q3/Q4 2026—and whether NVIDIA consequently switches suppliers. ⚠️ Risk Two: Fragmentation of Technology Roadmaps—GaN-on-Si vs. GaN-on-SiC vs. GaN-on-GaN Currently, GaN-on-Si (silicon-based GaN) dominates the mainstream. However, in high-voltage applications, a shift toward GaN-on-SiC (silicon-carbide-based GaN) or GaN-on-GaN (bulk GaN substrates) may occur. Such a transition would potentially reconfigure the process-related barriers to entry for integrated device manufacturers (IDMs), rendering existing production capacity vulnerable to impairment. ⚠️ Risk Three: Valuation Bubble in “GaN-Concept” Stocks In the first half of 2026, A-share stocks tagged with the “GaN” concept posted substantial gains. Yet very few possess true IDM-level GaN device shipment capability. For companies engaged solely in epitaxy, substrates, or chip design, GaN-related business typically accounts for only a small portion of their overall operations—meaning their stock price appreciation may far outpace actual business execution. ⚠️ Risk Four: High Customer Concentration Innoscience’s AI/data-center business is scaling rapidly—but if NVIDIA’s rollout of its 800V platform slows (e.g., due to an AI investment bubble), GaN adoption in AI power supplies could be delayed. ⚠️ Risk Five: GaN Is Not a Replacement for Silicon Carbide—They Serve Distinct Applications Do not frame GaN and silicon carbide (SiC) as mutually exclusive alternatives. SiC holds structural advantages in main-drive inverters for new-energy vehicles (1200V+ / high-current), whereas GaN excels in high-frequency, medium-to-low-voltage, and AI-power-supply applications. They are complementary—not substitutive. Conclusion Faced with AI’s soaring electricity bills, GaN is not an “optional upgrade”—it is a “physical necessity.” As AI compute scales from kilowatt-per-GPU to megawatt-per-rack, every 1% improvement in power-system efficiency saves hundreds of millions of kilowatt-hours annually. SiC cannot handle high frequencies; silicon cannot handle high voltages—only GaN can simultaneously satisfy the triple constraints of >800V operation, high frequency, and high efficiency. This is not hype—it is physics. A company losing RMB 800 million per year commanding a RMB 56-billion market cap reflects not a bet on its present performance—but a bet on the future of AI power supplies. References [1] Innoscience 2025 Annual Report / Lao Yao Bar. Innoscience achieved revenue of RMB 1.213 billion in 2025, up 46.3% year-on-year. March 30, 2026. [2] Sohu.com / Intellectual Property Economics. U.S. ITC Issues Cease-and-Desist Order Against Innoscience. May 7, 2026. [4] East Money Forum / Innoscience Interim Report. Innoscience: Key Information from the 2025 Interim Report. September 12, 2025.