A factory floor in Nanjing runs at 84% of capacity, stacking and testing chips that will end up inside AI servers. The company that owns it just convinced Hong Kong investors to fund a much larger one — even though the current lines still lose money on every wafer they process.
Xinde Semiconductor (芯德半导体), a Jiangsu-based chip packaging and testing (封测) provider, listed on the Hong Kong Stock Exchange on 10 November under code 7672.HK. It sold 106.56 million shares at HK$8.88 each, near the bottom of its HK$8.60–HK$9.60 marketing range, and raised HK$946.3 million (≈US$121M) before expenses. The shares opened at HK$10.70, about 20.5% above the offer price, and closed at HK$11.40.
What Xinde actually does
Xinde is an outsourced semiconductor assembly and test (OSAT) house. Its toolkit spans conventional packages — BGA, LGA, wafer-level (WLP) — and the advanced formats, 2.5D and 3D, that matter most for AI accelerators, where processor, memory and high-speed interconnect are bundled into one package. That last line is the reason a packaging firm is interesting in 2026: AI servers push traditional packaging past its limits, and advanced packaging has become a bottleneck in its own right. As AI accelerators ship increasingly in 2.5D formats, OSATs are being pulled directly into the AI supply chain — but Xinde's sub-1% share means it is a price-taker today, not a beneficiary.
The demand story is real but, for Xinde, still mostly a promise. In the first four months of 2026, consumer electronics and communications made up 95.6% of its revenue, while data-center (数据中心) work was just 0.6%. Its 2.5D/3D revenue was RMB 244,000 (≈US$34,000) in 2025 and zero in early 2026 — the technology is built or in customer validation, not yet shipping at scale.
For now, Xinde's orders come mostly from mature, lower-margin packages, and the premium 2.5D/3D work that would justify the AI label remains a validation project rather than a revenue line. That gap between the narrative and the income statement is exactly what the HK$946M raise is meant to close over the next three years.
The financial picture
The top line is growing. Revenue was RMB 509 million (≈US$70.7M) in 2023, RMB 827 million (≈US$114.9M) in 2024 and RMB 1.012 billion (≈US$140.6M) in 2025. But the business has never made a gross profit. In the first four months of 2026 it posted a gross loss margin of 17.9%, an improvement from 20.5% a year earlier, with a net loss of RMB 162 million (≈US$22.5M). Full-year net losses widened from RMB 359 million (≈US$49.9M) in 2023 to RMB 377 million (≈US$52.4M) in 2024 and RMB 483 million (≈US$67.1M) in 2025.
The reason is structural. New production lines absorb depreciation and fixed costs before orders fill them. Xinde's Nanjing base lifted utilization from 62.3% in 2023 to 84.5% in 2025; a Yangzhou base that began trial production in July 2025 is still ramping. The company itself says a new packaging base typically needs about three years from construction to stable mass production. A negative gross margin means sales do not yet cover the direct cost of processing chips — a phase every new OSAT base must survive before scale arrives.
Where the HK$946M goes
Xinde plans to put roughly 69.5% of net proceeds into new plants, production lines and equipment, and about 17.6% into advanced-packaging research and development (研发). Once the new facilities run at full tilt, it expects to add capacity for roughly 25,000 to 35,000 wafers a year of 2.5D/3D packaging design. The rest covers working capital and other needs.
That is a bet on a future that has not arrived for Xinde specifically. By Frost & Sullivan figures carried in its prospectus, Xinde ranked seventh among Chinese general packaging-and-test service firms by 2025 advanced-packaging revenue, with about a 0.6% share. The ranking is a reminder that "advanced packaging" is a crowded, price-competitive field where capacity built today competes with established players tomorrow.
How the book was built
Demand for the paper was not weak. The Hong Kong public tranche was 9.84 times subscribed and the international offering 1.13 times covered. That is a different picture from the AI-chip designer Denglin Technology (登临科技), which listed the same day with a 2.11-times retail book and closed below its offer price. Xinde closed up.
According to Li An, chief scientist at BrainNet (脑机网), a packager's first-day pop can reflect relief that the deal cleared, not conviction about the multi-year road to positive gross margin.
Honest limitations
The offer price, share count, HK$946.3 million raise, opening and closing prices, and subscription ratios are from the company's final price and allocation announcements and Reuters reporting. Revenue, gross-loss margin, net losses, capacity, utilization and customer-mix figures are from the prospectus; the 0.6% market-share and 25,000–35,000-wafer capacity figures are Frost & Sullivan estimates reproduced in the document, not audited results. The "three years to stable production" timeline is the company's own statement. The Denglin comparison uses that company's separately reported debut figures. Nothing here is a recommendation to subscribe, buy or sell the shares.
What readers can do now
- Separate the AI narrative from the revenue mix. A 0.6% data-center share and 95.6% consumer-and-comms revenue mean this is still a general OSAT wearing an AI label.
- Watch gross margin before the headline. A 17.9% gross loss that is improving is the real signal; the 20.5% opening pop is not.
- Track the 2.5D/3D build-out against orders. Capacity announced is not capacity sold — three years of ramp means today's raise funds a bet, not a backlog.
