Everything About AI
NextFin News - Shenzhen Longsys Electronics Co. priced its Hong Kong initial public offering at HK$236 a share, just under the top end of the marketed range, a move that hands new investors an immediate discount of about 44% to the company's Shenzhen listing on a currency-converted basis and lays bare the cautious mood greeting even the hottest names in China's artificial-intelligence supply chain. The memory-module maker, whose profit surged more than 700-fold in the first half of 2026 on a memory-price boom, settled at the low end of the HK$240.60 ceiling after a bookbuild that drew support from a roster of industrial cornerstone investors rather than a rush of financial demand. The pricing, set to be formalized on September 4 ahead of a September 8 trading debut under stock code 9976, is less a celebration of the memory upcycle than a reminder that Hong Kong's revived IPO market still exacts a heavy entry fee from sellers.
The Deal: A Blockbuster Balance Sheet Meets a Discounted Price Tag
Longsys offered 26,077,800 H shares, split between a 2,607,800-share Hong Kong retail tranche and a 23,470,000-share international placement, subject to over-allotment. At the final price of HK$236, the offering raises approximately HK$6.15 billion, slightly below the maximum HK$6.27 billion the company had targeted at HK$240.60. Net proceeds are expected to come in around HK$6.02 billion. The deal is being steered by CITIC Securities and Citigroup as joint sponsors and overall coordinators.
The discount to the A share is the headline within the headline. Longsys closed at 359.63 yuan, or $53.52, in Shenzhen on Thursday. The HK$236 offer price converts to about $30.10, implying the Hong Kong shares trade at roughly 44% below the mainland listing. That gap is not a rounding error; it is the price Hong Kong investors demand for taking liquidity and currency risk on a company whose shares already trade on ChiNext. Even a company printing money in the middle of an AI-driven memory shortage cannot simply transplant its A-share valuation across the border.
Backing for the deal came from the real economy rather than the financial markets. Longsys secured cornerstone commitments totaling about US$151 million, or roughly 19% of the offer shares, from a cast of customers and supply-chain partners including Transsion International, Lenovo Group, TCL Technology's AHGO SPC, Lens Technology, Beijing Injoinic, and Colorful, alongside institutional names such as CITIC Securities Asset Management Hong Kong. The proceeds are earmarked primarily for research and development: about 78.3% will fund chip design and advanced storage-product development, 11.7% is reserved for strategic investments or acquisitions, and the remaining 10% will go to working capital.
"We are an independent-brand semiconductor memory vendor," the company stated in its prospectus, underscoring a business model that buys memory wafers and controller chips from integrated-device manufacturers and controller suppliers, then adds value through controller and firmware design, system-level packaging, testing, and assembly.
Why the Numbers Look So Good: A Memory Cycle on Steroids
To understand why Longsys can command attention in a still-frail IPO market, you have to follow the memory-price curve. The company's financials read like a company that has ridden a rocket. Revenue climbed from 10.125 billion yuan in 2023 to 17.464 billion yuan in 2024 and 22.766 billion yuan in 2025, a 30.4% year-on-year increase. Then the cycle turned violent. In the six months through June 2026, revenue reached 24.088 billion yuan, up 136.3% from a year earlier, while net profit attributable to shareholders jumped to 10.577 billion yuan, a gain of more than 71,500% from the 14.75 million yuan earned in the same period of 2025. Second-quarter profit alone was 6.715 billion yuan, up 3,930.9% year on year.
The driver is not company-specific genius; it is a global supply squeeze with an AI engine. A 2026 industry analysis found that AI-server DRAM prices roughly doubled during the first quarter, with a fourfold increase expected for the full year, as hyperscale cloud providers locked in supply through long-term agreements. Samsung Electronics disclosed that its blended DRAM and NAND average selling price surged 146% in the first quarter of 2026 compared with the 2025 full-year average. New capacity is not expected to come online until 2029 or 2030, meaning the shortage has a long fuse.
Longsys sat on the right side of that move because it had stocked up when prices were low. The company's gross margin on memory products expanded from about 13% in the first half of 2025 to 59.08% in the first half of 2026, a 46-percentage-point swing. But that margin story carries a mirror image: inventory. As of June 30, 2026, inventories stood at 25.777 billion yuan, 60.12% of total assets, up from 21.344 billion yuan at the end of April. Operating cash flow for the first half of 2026 was negative 3.151 billion yuan. The company has, in effect, converted cash into chips and is betting the cycle stays friendly long enough to sell them at a profit.
That is a cyclical bet, not a structural transformation. The memory industry has spent decades oscillating between glut and shortage, and Longsys's own history proves the point: a net loss of 837 million yuan in 2023, followed by profits of 505 million yuan in 2024 and 1.498 billion yuan in 2025. The 2026 explosion is the upswing of that same pendulum, amplified by AI.
The Second-Order Question: Is This a Memory Company or an AI Company?
The market's temptation is to re-rate Longsys as an AI infrastructure play. That is a mistake, and the pricing discipline suggests the bookrunners knew it. Longsys is a memory company with AI exposure, not an AI company with a memory balance sheet. The distinction matters because the two trade on different multiples and are vulnerable to different risks.
An AI company earns recurring, high-margin revenue from software or proprietary silicon whose demand grows with model complexity. Longsys earns its margin by timing the purchase and sale of commodity memory. Its non-consumer business, which includes enterprise and industrial applications, grew from 5.3% of revenue in 2023 to 11.6% in the first four months of 2026, but consumer products still dominate. The company's largest customers include Dell, Lenovo, Mindray, OPPO, Samsung, Transsion, and Xiaomi, a roster that reads like a consumer-electronics directory. Enterprise storage revenue did surge 208.8% in the first half of 2026, to 2.14 billion yuan, but it remains a small slice of the total.
The second-order implication is uncomfortable for bulls. If memory prices stay high, Longsys prints money in the near term, and the HK$236 price looks cheap. But high memory prices are also a tax on the very AI demand that is supposed to sustain the cycle. Every percentage-point increase in DRAM cost flows into server bills, smartphone prices, and automotive electronics budgets. Industry data showed automotive-grade memory prices rose about 180% over a three-month span in 2026, adding 7,000 to 10,000 yuan to the cost of each vehicle. At some point, demand destruction kicks in, and the companies with the most inventory leverage, like Longsys, feel it first.
The Counter-Thesis: This Time, the Cycle Has a Floor
The strongest argument against the cyclical view is that this upcycle is structurally different from those that came before. Bulls point to three facts. First, AI server demand is not a consumer fad; industry analyses estimate a single AI server uses eight to ten times more DRAM than a traditional server, and global AI server shipments are projected to reach 1.5 million units in 2026, up 180% year on year. Second, supply is concentrated in the hands of Samsung, SK Hynix, and Micron, all of whom have prioritized high-bandwidth memory for AI over commodity DRAM, structurally tightening the rest of the market. Third, Longsys is not a passive trader; it has built controller-chip and firmware capabilities, with more than 280 million self-designed controller chips cumulatively produced, that let it differentiate products for automotive and industrial customers, where margins are stickier.
There is weight to this argument. Longsys is, by revenue, the world's ninth-largest memory vendor and the second-largest independent semiconductor memory vendor globally, with a 1.2% share of the global memory-products market in 2025. Its Lexar consumer brand ranks second globally with a 3.3% share, and its FORESEE B2B brand ranks second with a 0.5% share. That scale gives it procurement leverage that smaller rivals lack, and its Zilia brand leads the Latin American independent memory market.
But the counter-thesis still rests on a cyclical foundation dressed in structural clothing. The concentration of HBM production does tighten commodity supply, but it does so only as long as the big three keep making that choice. If HBM margins normalize or AI capital expenditure slows, the capacity will flow back. The falsifying signal is specific and observable: if DRAM contract prices fail to rise in the second half of 2026, or if Longsys's memory-product gross margin falls below 40% in any full fiscal year after 2026, the structural-upgrade narrative is broken and the company reverts to its historical role as a leveraged bet on memory pricing.
The Bigger Picture: Hong Kong's IPO Revival Still Has a Price
Longsys is not listing in isolation. It joins a wave of Chinese technology companies using Hong Kong as an alternative to increasingly difficult U.S. listing conditions, part of a broader rebound in a market that spent much of 2022 and 2023 in a listing drought. Hong Kong's exchange ranked first globally by IPO funds raised in the first quarter of 2026, and industry projections put total 2026 Hong Kong IPO activity as high as US$45 billion. The pipeline includes names from Shein's massive debut to smaller semiconductor listings such as Ingenic Semiconductor's US$410 million offering.
Yet the revival has a price, and Longsys's discount is the bill. Hong Kong investors have been burned before by mainland companies arriving with A-share valuations intact, only to see the H shares trade at deep, persistent discounts. The theoretical H-share discount at the top of Longsys's range was roughly 45% against the A share's late-August close; the final price keeps that gap wide. For issuers, Hong Kong offers speed and regulatory predictability. For investors, it offers a bargain basement, but only if they are willing to wait for the rest of the market to catch up.
The policy angle matters too. Beijing has made semiconductor self-sufficiency a national priority, and Longsys's use of proceeds, with nearly 80% earmarked for chip design and advanced storage development, aligns squarely with that agenda. The company's ability to raise capital in Hong Kong for domestic R&D is a feature, not a bug, of China's technology strategy. But policy support does not insulate a memory trader from the price cycle; it only changes who is willing to fund the inventory build.
What Comes Next: Three Scenarios for the September 8 Debut
The immediate question is how the stock trades when it begins dealing on September 8. Three scenarios frame the risk.
In the base case, the shares open modestly above the HK$236 offer price, supported by the cornerstone lockups and the visible earnings momentum, then trade in line with the memory cycle. Investors get a liquid proxy on memory prices without paying the A-share premium. In the upside case, a fresh memory-price hike or a stronger-than-expected quarterly update sends the stock toward the A-share discount's fair value, and the 44% gap narrows as Hong Kong investors re-rate the company as a genuine AI-supply-chain beneficiary. In the downside case, a broader Hong Kong market selloff, a memory-price pause, or a weak debut from one of the other listings in the same week triggers a break of issue price, and the discount to the A share widens further as investors demand an even larger margin of safety.
For the longer term, the watch list is narrow. Track DRAM and NAND contract-price reports from TrendForce each quarter, Longsys's inventory level and turnover days in its interim and annual reports, and the trajectory of the non-consumer revenue mix. If inventory stays above 50% of assets while margins hold above 40%, the company has earned the right to a higher multiple. If inventory swells and margins compress together, the cycle has turned, and the HK$236 price will look like the top of the market rather than the entry point.
Longsys's Hong Kong listing is a bet that the memory boom has room to run. The company's earnings say the boom is real. The 44% discount says the market is not willing to pay for the sequel.
Explore more exclusive insights at nextfin.ai.
