Quick ReadPlain-language overview · read this first
Iluvatar CoreX（09903.HK） is a domestic general-purpose GPU designer that sells computing power through training cards, inference cards, and AI solutions. The report rating is "Avoid". Revenue in 2025 was RMB 1.034 billion, equivalent to about HK$1.197 billion, with a gross margin of 54.0%, while adjusted net loss narrowed to RMB 438.8 million. The report believes the company has moved past the most dangerous phase of "only R&D, no volume ramp"; the inference ramp is real progress, but the company is still loss-making and has not yet proved it can earn stable profits.
The moat has an institutional flavor: as of the base date, Iluvatar CoreX did not appear in searches of the U.S. Entity List, while peers such as Moore Threads and Biren had already been added, making this status scarce among comparable companies. Together with high project switching costs and inclusion in mainstream domestic computing-power compatibility lists such as 4Paradigm's, all 3 moat lines are still forming. The report cautions that the list advantage is granted by external rules, while ecosystem control remains in the hands of stronger rivals such as NVIDIA and Huawei.
Valuation is the dividing line for the rating: the June 10 closing price was HK$519, with a market value of about HK$132 billion, implying a price-to-sales ratio of about 110 times. The report's estimated ideal buying range is HK$190 to HK$255, placing the current price in the "clearly overvalued" zone. The market is pricing in scarce status, an extremely thin free float, and the computing-power theme; HK$519 has already pulled forward two to three years of delivery and leaves almost no margin of safety.
The biggest risk is lock-up expiry: the cornerstone lock-up expires in July 2026, and the existing shareholder lock-up expires in January 2027. These 2 rounds of unlocks will reshape supply and demand for this recently listed stock. If revenue loses momentum or the list status changes at the same time, the report estimates that a 50% to 65% drawdown would not be extreme. The company deserves serious study, but the current price should be avoided. Only when the share price returns below HK$255 and high growth is validated through another reporting cycle could the conclusion potentially be revised upward. The above is a summary of the report's views and does not constitute investment advice. The stock market involves risk; invest with caution.
LeadA domestic general-purpose GPU designer, Iluvatar CoreX sells compute through training cards, inference cards, and AI solutions, making it a scarce Hong Kong-listed name not yet on the Entity List. In 2025, revenue reached RMB 1.034 billion, gross margin was 54.0%, adjusted loss narrowed to RMB 438.8 million, and the stock traded at roughly 110x price-to-sales. Rating Avoid: inference volume is real progress, but HKD 519 already discounts two to three years of execution, with an ideal buy range of HKD 190-255.
Prices in the article are as of publication; see the valuation band above for the live price.
Metadata
Ticker: 09903.HK
Full company name: Shanghai Iluvatar CoreX Semiconductor Co., Ltd.
Current price and market cap: HKD 519 / HKD 131.99 billion, both as of the 2026-06-10 close. The closing price is cited from the Reuters market page, and market cap from Morningstar's same-day page.
Currency: HKD
Report date: 2026-06-11
Industry classification: Semiconductor
One-line positioning: A domestic general-purpose GPU designer that sells compute through training cards, inference cards, and AI solutions.
This report uses 2026-06-11 as the research base date and Hong Kong shares as the primary quotation framework. Because the company was listed on the Hong Kong Stock Exchange Main Board only on 2026-01-08, its trading history is extremely short. The research focus is therefore on the prospectus, allotment results, first post-listing annual results, share lock-up arrangements, and secondary-market float structure, rather than stretching analysis across a listing history that does not exist. The reporting currency in the financial statements is RMB. Unless original-source wording is specifically noted, RMB/HKD conversions in this report use the 2026-06-10 historical rate of 1 HKD = 0.8641 CNY, equivalent to 1 CNY ≈ 1.1573 HKD.
Research Summary
Iluvatar CoreX has built a business that can genuinely sell, not an empty shell that merely tells a domestic-substitution story. It sells training GPUs, inference GPUs, boards, software stacks, and solutions, deploying domestic compute into the server rooms of government-enterprise, internet, and industry customers. In 2025, revenue rose to RMB 1.034 billion, gross margin reached 54.0%, and adjusted net loss narrowed to RMB 438.8 million. That means it has passed the most dangerous stage of having only R&D and no volume, but it has not yet reached the stage where profits can fund the business by themselves. Put simply, it has proved the product can sell, but has not proved the business can earn steady profits without sustained heavy R&D spending and capital-market tolerance.
What the market is trading now is not just this RMB 1.0 billion of revenue, but three overlapping narratives. First, Iluvatar CoreX is one of the very few directly tradable domestic general-purpose GPU pure plays in Hong Kong. Second, as of the base date, searches for “Iluvatar” and “上海天数智芯” in the U.S. EAR Entity List returned no matches, while the same list clearly returned Moore Threads, Biren, and Shanghai Cambricon entries. This means the capital market is willing to price it as a domestic GPU supply-chain option that has not yet been pinned down by the list. Third, its early post-listing float was extremely thin, while the southbound Stock Connect list adjustment in early June pushed it into a new liquidity narrative. Scarcity, no Entity List entry, and a small free float together mean the share price has behaved more like a high-beta option in the short term than a mature stock already priced by discounted cash flow.
This also explains why the share price has moved so aggressively over the past five months. The IPO price was HKD 144.6, and the stock opened 31.54% higher at HKD 190.2 on its first trading day. By the 2026-06-10 close, it had reached HKD 519, about 2.6 times above the IPO price, with market cap rising to about HKD 132.0 billion. The drivers were mainly scarcity premium in a newly listed stock, the domestic AI compute theme, repaired sentiment toward Hong Kong IPOs, and another acceleration after Stock Connect inclusion expectations and realization in early June. They were not driven by a profit-and-loss statement that had already delivered threefold growth. This looks more like a repricing of a stock with extremely tight float and a sufficiently large story.
The most important bull-bear disagreement can be reduced to one sentence: is Iluvatar CoreX moving from technical validation toward scale delivery, or sliding from real progress into valuation overreach? Bulls see inference volume in 2025, narrowing losses, no Entity List entry, and ecosystem adaptation into mainstream domestic chip lists, implying that its position in domestic AI infrastructure may be more important than reported revenue suggests. Bears see the same company with just over RMB 1.0 billion of 2025 revenue, still losing money and heavily dependent on R&D and supply-chain resources, yet already valued by the secondary market at more than HKD 100 billion. That effectively pre-settles the next two to three years of execution at a perfect score. Neither side is empty talk. The real difference is whether one is willing to pay today's price for upside space that has not yet been disproved.
Putting fundamentals, competitive position, capital-market expectations, and valuation together, my view is this: at the enterprise level, the company has entered a volume inflection point; at the stock level, it has already been pushed into a high-expectation zone. It is a high-risk growth company with real technology and real customers, but still on the eve of profit validation. It is neither a mature cash cow nor a high-quality compounder. If I had to choose one label, it would be “in valuation reset.” The company has indeed moved from the R&D phase into revenue scaling, but the stock has re-rated much faster than the earnings model has matured. Valuation no longer follows current profits and has instead moved to pre-discounting its identity as a scarce domestic general-purpose GPU asset.
Company Development History
Why It Emerged
Iluvatar CoreX was founded in 2015, and the timing matters. That year, China was still using the broad basket of “AI chips” to discuss the future, while local teams capable of building general-purpose GPUs, especially teams able to bring both training and inference into a unified software stack, were rare. Its early logic was to build a complete GPGPU route on the data-center side, rather than a niche accelerator to fill a domestic gap. This path is much harder than building a dedicated ASIC, but once it works, the ceiling is also higher. The product architecture, dual training and inference lines, and “GPGPU + software stack + solutions” combination disclosed later all show that the company aimed from the start at platform capability rather than a single product.
Its starting point also had a rarely discussed but important feature: from the beginning, Iluvatar CoreX was driven by both capital-operations capability and engineering capability, not by a single star architect founder. In the board and equity arrangements disclosed in the prospectus, the Shanghai Shuqi platform series led by Gai Lujiang long served as the core voting-rights hub. After listing, the single largest shareholder group held about 21.25%, while Centurium held about 20.62%. The shareholding structure was highly dispersed. In governance terms, this looks more like a configuration of platform shareholders plus professional managers than a typical founder-controlled company. In May 2025, Shanghai Shuqi's sole shareholder and executive director changed from Diao Shijing to Gai Lujiang, while the relevant platforms continued to exercise voting rights for employee shareholding platforms centrally. The advantage of this structure is efficient financing and listing execution. The downside is that ordinary investors find it difficult to tie the company's destiny simply to one “technical godfather.”
Listing Path and How the Capital Market First Understood It
The company ultimately chose the Hong Kong Main Board instead of continuing to wait for an A-share or STAR Market window. That choice itself carried strong period context. From late 2025 to early 2026, Hong Kong was again absorbing an AI and hard-tech listing wave. Reuters noted that Chinese authorities were accelerating listings of AI and chip companies to strengthen domestic substitution for high-end U.S. technology. Within that window, Iluvatar CoreX issued 25,430,000 shares at HKD 144.6 per share, raising about HKD 3.677 billion, and opened 31.54% higher on its first day. The capital market initially gave it a very simple label: scarce domestic general-purpose GPU target. That label was powerful because investors quickly found that Hong Kong had few direct targets that offered the domestic GPU theme without being upstream materials or equipment companies.
The prospectus and allotment results told the market two additional things. First, the offering size was not large, with 25.43 million shares accounting for only a small portion of total share capital. Second, cornerstone and lock-up arrangements made tradable float even thinner. The allotment results showed that 219,670,165 H shares held by existing shareholders were subject to a one-year lock-up under Chinese company law, expiring around 2027-01-07. H shares subscribed by cornerstone investors were locked until around 2026-07-07. In other words, the company appeared to have 245.10 million H shares, but during the first half-year after listing, the amount that could truly trade freely was far smaller. This planted the fundamental reason the share price was later squeezed upward.
The Three Stages It Has Actually Passed Through
The first stage was the “technology bet” period. During this phase, the company had to solve whether it could build a domestic general-purpose GPU route at all, rather than educate the market. The constraints were hard: design talent was scarce, and tape-out, packaging, memory, boards, and the software chain all had to work, with each step burning capital. The company's repeated emphasis in later filings on both training and inference product lines shows that from the beginning it placed ambition on the harder general-purpose route, rather than choosing the lighter path of doing only edge or only inference.
The second stage was “product landing without financial closure.” This broadly corresponds to the reporting period covered by the prospectus. Revenue started to rise, but net losses expanded at the same time, and R&D expenses were above or close to revenue for years. For a semiconductor start-up, this is the cost of entry, not an anomaly. Only after products are polished enough for customers to deploy in real clusters can later scale opportunities emerge. But capital markets usually do not pay indefinitely for this stage, so Iluvatar CoreX chose to list when revenue had just formed a slope and losses had not yet turned to profit. In essence, it used capital-market funds to cross the final deep-water stretch from validation to volume.
The third stage, beginning in 2025, is “inference volume and valuation reset.” The annual report shows that 2025 revenue and gross profit continued to grow rapidly, adjusted net loss narrowed, and the company for the first time put “growth” and “narrowing loss” on the same answer sheet. The secondary market was more excited than the fundamentals: it interpreted this improvement as a domestic GPU company finally switching from “can it be built” to “can it sell more.” After that, Stock Connect inclusion in early June gave the share price a second acceleration. Fundamentals took one step; valuation ran three steps ahead.
Key Nodes That Still Matter Today
The most important node is the fact that the company is not on the Entity List, not the listing itself. By 2026-06-11, official U.S. EAR Entity List text searches produced no matches for Iluvatar or Shanghai Iluvatar CoreX, while Moore Threads, Biren, and Shanghai Cambricon could be found. This difference does not guarantee the supply chain will always be safe, but it gives Iluvatar CoreX a selling point rarely seen among domestic GPU peers in the capital market: if customers need to find a supplier with domestic GPU capability that has not been directly blocked by the list, Iluvatar CoreX naturally enters the candidate set. A substantial part of the premium the market gives it today comes from this relatively clean identity.
The second key node is the opening of southbound investability. In early June, the Shanghai Stock Exchange updated the Hong Kong Stock Connect eligible securities list and included 09903. Around the same time, AASTOCKS recorded sharp rises in Iluvatar CoreX and comparable AI/chip stocks after the news. For a newly listed stock with only five months of trading history and an already thin float, Stock Connect directly changed the buyer structure. It was far more than a simple liquidity increment. Previously, the game was mainly between international placing investors and local Hong Kong capital. After inclusion, mainland thematic capital could participate more smoothly, and the valuation framework became more prone to converge with high-beta A-share technology stocks.
The third key node is that in June 2026 shareholder-meeting documents, the company simultaneously proposed an H-share incentive scheme, a general mandate for H-share repurchases, and an arrangement stating that there was no distributable profit for 2025. These three points are interesting together. The company knows it does not yet have cash-return capacity, but it is already building frameworks for post-listing equity incentives, secondary-market management, and talent retention. More importantly, the documents explicitly explained why service providers were included in the incentive scheme: chip R&D cycles are long, switching suppliers and partners is costly, and external teams often allocate dedicated resources to projects over long periods. This disclosure indirectly admits that Iluvatar CoreX's business model depends on a deeply bound collaboration network, not simply on selling cards.
Business Model and Industry Cycle
How It Makes Money, and Which Part Looks More Like a Real Business
Iluvatar CoreX's revenue machine can be summarized in three layers. The top layer is general-purpose GPU products, split into training and inference lines. The middle layer is AI computing solutions that package boards, servers, cluster deployment, and tuning. The bottom layer is the software stack and ecosystem adaptation. It may not collect much standalone revenue, but it determines whether customers can actually use the product. The core change in 2025 was that inference-side volume genuinely appeared in revenue, rather than another round of additional signed projects. For a domestic GPU company, this matters more than releasing a new product, because inference is closer to continuous deployment and scalable replication.
The cost structure also explains why stable profitability has not yet emerged. General-purpose GPUs are not an asset-light software business. Memory, wafers, packaging and testing, boards, and server matching all consume substantial materials and prepayments. Supplier-structure disclosures in the prospectus show that procurement concentration is high, and in some stages the largest supplier was mainly a memory-device supplier. This explains why outsiders often use the phrase “the flour costs more than the bread” to challenge the business. In large-model training and inference clusters, memory and surrounding BOM are inherently heavy, and looking only at the material cost of one card can easily produce a pessimistic conclusion. The issue is that what is often sold in the financial statements is packaged delivery of full cards, full machines, software adaptation, and solutions, rather than a lonely bare GPU die. The 54% consolidated gross margin in 2025 shows that, at least at the reporting level, the “flour costs more than bread” claim cannot be treated as proven fact. More strictly, public disclosure does not provide enough unit-card shipment, unit-card ASP, and BOM data for outsiders to verify the claim rigorously using primary data.
Where the Real Moat Is, and Where the Promotional Moat Is
I think Iluvatar CoreX has only three real moats, and each is still forming.
The first is its relatively scarce supply-chain identity. It is not an Entity List company, which gives it a tangible differentiating attribute among domestic GPUs. This moat is a trading qualification temporarily granted by institutional and geopolitical conditions, not a technological moat. It is unstable, but very valuable for now.
The second is project-based switching cost. The company said it plainly in the H-share incentive scheme documents: chip R&D and mass-production cycles are long, external service providers continuously invest dedicated resources, and switching costs are so high as to be almost impracticable. This sentence contains more information than any slogan about a complete ecosystem. It shows that once customers bind a generation of products, drivers, model frameworks, scheduling systems, and operations processes to one company, they will not migrate as easily as replacing an ordinary server.
The third is that domestic ecosystem usability has moved past the sample stage. 4Paradigm's 2026 annual report disclosed that its platform is compatible with mainstream domestic computing platforms including Huawei Ascend, Cambricon, Iluvatar CoreX, Kunlunxin, and Hygon. The value of this type of disclosure is that it proves Iluvatar has entered the adaptation pool of mainstream software and application ecosystems. It is no longer a product that can only run benchmarks in its own PowerPoint. It does not prove Iluvatar CoreX is technically leading.
As for the moats in market promotion, such as “earliest mover,” “full-stack self-development,” and “inevitable beneficiary of domestic substitution,” none can be treated directly as moats. Earliest mover does not automatically mean largest scale. Full-stack self-development does not automatically mean the best ecosystem. Domestic substitution does not mean Iluvatar CoreX will necessarily receive the largest slice of the market. The true determinant remains whether, over the next three years, it can lift revenue from the RMB 1.0 billion level to the RMB 3.0 billion level without sacrificing gross margin or cash-consumption discipline. If it cannot, many of today's “moats” will prove to be promotional words from a favorable-cycle period.
Where the Industry Is in the Cycle
This industry is a technology iteration cycle, a policy cycle, and a capex cycle at the same time. Upstream is a full set of training and inference infrastructure, not a single chip. Downstream includes government-enterprise, financial, telecom, cloud-service, and model-application customers, not a single internet client. Once policy emphasizes domestic substitution, capital markets are willing to give high valuations to AI infrastructure, and customers are willing to deploy early, orders, financing, and valuations for local GPU companies rise together. Reuters' description of the background to Hong Kong AI and chip IPOs in early 2026 already made this environment clear: behind the listing wave is the need to strengthen China's ability to substitute high-end U.S. technology.
But the industry's weak points are equally clear. As long as U.S. export controls continue to evolve, or any key link in the supply chain tightens, product roadmaps, delivery schedules, and customer budgets will have to be rearranged. Iluvatar CoreX's largest current external advantage comes precisely from the fact that it has not yet been named on the Entity List. Equally, its largest external risk is that this state is not permanent. For this company, policy is a variable written directly into the valuation multiple, not a backdrop.
Horizontal Competitor Analysis
What Each Competitor Has Become
If international vendors are included, Iluvatar CoreX faces a highly asymmetric competitive field. NVIDIA defines the global ceiling for training-GPU performance and software ecosystem, and CUDA remains the default standard. AMD plays more like the second international option, offering an alternative when customers do not want to put all chips on NVIDIA. Their moats already lie in software, developers, and long-term large-customer relationships, not in a single chip. NVIDIA's current market cap is about USD 4.89 trillion, with PE around 30.5x. AMD's market cap is about USD 746.5 billion, with PE around 148x. This scale reminds investors that the high valuations given to GPU companies in international markets are usually built on huge revenue and mature ecosystems.
Among domestic comparables, Huawei Ascend is unavoidable in government-enterprise and localization mega-projects, but it is more of a “chip + full machine + framework + channel” system than a directly investable listed pure play. Hygon Information is another route: it is already profitable and exists with a larger revenue base and a more mature data-center platform form. By 2026-06-11, Hygon Information had a market cap of about RMB 660.6 billion, trailing PE of about 227x, and price-to-sales of about 38.7x. This is a company for which the market is willing to pay a high valuation for earnings quality and platform depth. Cambricon is closer to the A-share mapping of a high-beta domestic AI chip leader. On 2026-06-10, its market cap was about RMB 773.36 billion, with 2025 net profit of about RMB 2.059 billion. Its valuation center is clearly higher than traditional semiconductors, but it has at least crossed the profitability threshold.
The companies most similar to Iluvatar CoreX are Biren Technology and Moore Threads. Biren, as a direct Hong Kong comparable, had a closing price of about HKD 55.4 and market cap of about HKD 135.1 billion on 2026-06-10, with price-to-sales above 100x. The market gives it the same “scarce newly listed domestic GPU stock in Hong Kong” valuation framework. Moore Threads has already been added to the Entity List, showing that geopolitical risk in this track has actually occurred and is not an abstraction. Iluvatar CoreX and Biren have similar market caps today, but the largest difference is list status, not the number of product releases. Biren is already on the Entity List, while Iluvatar CoreX had not been included as of the base date. This one difference separates their investment narratives.
A Narrow Table to Locate the Position
The table below puts together the items investors care about most. A-share data in the table are converted at 1 CNY ≈ 1.1573 HKD for horizontal readability only.
| Dimension | Iluvatar CoreX | Biren Technology | Hygon Information | Cambricon |
|---|---|---|---|---|
| Listing venue | Hong Kong | Hong Kong | SSE STAR Market | SSE STAR Market |
| Latest share price | 519 HKD | 55.4 HKD | 335.3 HKD† | 1,405.0 HKD† |
| Latest market cap | 131.99 billion HKD | 135.11 billion HKD | 763.85 billion HKD† | 894.19 billion HKD† |
| Profit status | Still loss-making | Still loss-making | Profitable | Profitable |
| Known valuation metrics | About 110x TTM PS‡ | About 103x PS | About 38.7x PS, about 227x PE | High market cap and high beta; 2025 net profit RMB 2.059 billion |
| Entity List status | Not found | Found | Not used as a core comparison variable here | Related Shanghai entity can be found |
† Converted at the 2026-06-10 exchange rate. ‡ Estimated using 2025 revenue of RMB 1.034 billion after conversion.
The key information in this table is: who is expensive, and on what foundation, not simply “who is more expensive.” Hygon and Cambricon's high valuations are at least partly built on larger revenue bases and the ability to have crossed breakeven. Biren and Iluvatar CoreX's valuations look more like pricing the scarce right to trade a domestic GPU target in Hong Kong. If Iluvatar CoreX can continue to deliver revenue and narrower losses over the next two years, it has a chance to move gradually from “pure theme comparable” toward “operating comparable.” If it cannot, the similar market caps of Iluvatar and Biren today may ultimately prove to be mirrors of newly listed stock sentiment and theme trading.
Iluvatar CoreX's True Ecosystem Niche
Its most accurate position in the industry is this: a challenger in the domestic general-purpose GPU camp. It is neither the leader nor the weakest follower, and it is the one that most resembles an institutional-arbitrage challenger. The market gap it fills is “customers need an option that can make general-purpose GPUs, enter mainstream domestic adaptation lists, and has not yet been named by the U.S. Entity List,” rather than an absolute gap where “China has no training chips.” This means it most directly competes for replacement procurement space among domestic customers under the triple constraints of compliance, supply, and budget, rather than NVIDIA's highest-end global profit pool.
But its weaknesses are also clear. Customers that want the most mature software ecosystem will first look at NVIDIA. Customers that want the strongest domestic system and government-enterprise channels will first look at Huawei Ascend. Those that emphasize earnings quality and platform stability will compare Hygon. Those that simply want high-beta domestic GPU exposure will look at both Biren and Moore Threads. Iluvatar CoreX can still hold a position mainly because it fills the gaps between these paths. That niche is not wide, but in a shortage era it is valuable.
Current Fundamentals and Valuation Analysis
What Actually Happened Over the Past Year
The company's latest annual report sends a calmer signal than market sentiment, but the direction is positive. In 2025, revenue grew, gross profit grew, gross margin remained high, and adjusted net loss continued to narrow. Under the company's framework, adjusted net loss was RMB 438.8 million. Excluded items mainly included fair-value changes of financial liabilities measured at fair value through profit or loss, share-based payments, and listing expenses, while reported losses were higher than this figure. At the same time, the company had no distributable profit in 2025, and the board did not recommend a final dividend. For an AI chip company with thematic heat, this information matters: fundamentals are improving, but the financial reality remains that this is a loss-making company.
This data set also explains why the market is willing to keep going long, and why it can easily be disappointed. It is willing to go long because “high revenue growth + narrowing loss” finally appeared at the same time. It can be disappointed because even so, the company is still some distance from stable profitability. In addition, it has only been listed for five months, and the share price has not yet gone through a full “earnings miss to valuation reversion” test. It is therefore hard to say that the market has fully priced the risks.
What the Current Share Price Is Mainly Trading
In my view, the current share price mainly trades four words: scarcity, list status, float, and theme.
Scarcity means there are too few tradable domestic general-purpose GPU pure plays in Hong Kong. List status means Iluvatar CoreX has not yet appeared on the Entity List while some peers already have. Float means small issuance, high lock-up ratio, and short listing history. Theme means AI infrastructure, domestic compute, southbound capital, and Hong Kong technology re-rating. The company's actual operating progress provides a base for this narrative, but pricing elasticity has clearly amplified these operating improvements. The share-price reaction after Stock Connect inclusion in early June is a typical example.
What bulls truly believe is that inference volume will extend the 2025 improvement into 2026, that the supply-chain window created by the absence of an Entity List entry will be long enough, and that customers will leave room for second and third suppliers during domestic substitution. What bears truly worry about is that the company's current HKD 132.0 billion market cap corresponds to only about HKD 1.197 billion of 2025 revenue, a TTM price-to-sales ratio of about 110x. If 2026 revenue growth falls below market expectations, or if the July and January lock-up expiries turn the float structure from “squeeze” into “clearing,” the valuation swing back will be severe.
How Valuation Should Be Done
For Iluvatar CoreX, PE, DCF, and EV/EBITDA are not the primary methods now. The reason is direct: it is still loss-making, owner earnings are negative, and operating cash flow and net profit have not formed a stable positive cycle that can be extrapolated. At this stage, the most effective method is still forward price-to-sales, but one needs to strip out the free-float distortion and thematic premium before looking at operating delivery one year out.
First, today's valuation should be made clear. Based on 2025 revenue of RMB 1.034 billion, equivalent to about HKD 1.197 billion, and a market cap of about HKD 131.99 billion on 2026-06-10, Iluvatar CoreX currently corresponds to about 110x TTM price-to-sales. This multiple is close to Biren Technology's current roughly 103x price-to-sales and significantly higher than Hygon Information's roughly 38.7x price-to-sales. In other words, the market is valuing it as one of the scarcest domestic GPU options, not as a domestic semiconductor growth stock still proving itself.
The table below gives my three operating scenarios. Here, operations are estimated first, then the multiple, and only then mapped to the share price. This table is a research framework and does not constitute investment advice.
| Dimension | Bearish | Base | Bullish |
|---|---|---|---|
| Revenue/margin assumptions | 2026 revenue RMB 1.30-1.45 billion; inference growth cools; gross margin falls back to 48%-50% | 2026 revenue RMB 1.50-1.65 billion; inference continues to scale; gross margin 50%-53% | 2026 revenue RMB 1.75-1.85 billion; inference and solutions accelerate together; gross margin 53%-55% |
| Cash-flow assumption | Still negative, insufficient internal cash generation | Near breakeven, but still dependent on continued profit improvement | Cash burn narrows significantly, close to a verifiable path to profitability |
| Valuation multiple assumption | 40x-48x 2026E PS | 48x-52x 2026E PS | 54x-56x 2026E PS |
| Key catalysts | Smooth July lock-up expiry; orders do not stall | Two consecutive quarters in 2026 maintain high growth, with losses continuing to narrow | No escalation in overseas restrictions; inference orders exceed expectations; ecosystem partnerships expand |
| Key risks | Lock-up impact, declining inference ASP, competitor price cuts | Growth misses expectations; solution gross margin unstable | List-status change, supply-chain limits, valuation drawdown |
| Return versus current price | About -54% to -39% | About -37% to -25% | About -17% to -9% |
| Permanent capital-loss risk | Trigger: 2026 revenue growth falls below 30% and gross margin drops under 48% | Trigger: inference business keeps growing but price war breaks margins | Trigger: list-status change or delayed deployment by major customers causes multiple collapse |
The financial framework, market multiples, and peer references in the table come from the company's annual report, prospectus, and the latest market quotations for peers.
Mapping these scenarios to price, I arrive at the following ranges: fair value under the conservative operating scenario is roughly HKD 236-316; under the base scenario, HKD 328-390; under the bullish scenario, HKD 430-471. Applying at least a 20% margin of safety, the ideal buy range should fall to HKD 190-255, not today's HKD 519. What the market now implies is not just that “the company will keep growing,” but that “the company will keep growing rapidly, and the list-status advantage, low-float advantage, and thematic premium can survive the next lock-up round.” This assumption set is not impossible, but it is already very full.
Margin of Safety Check
Under the above framework, the current price represents a high premium to the conservative scenario, not a discount. The margin of safety is zero. The most fragile assumption is continued high revenue growth in 2026, especially whether the inference business can expand without meaningfully compressing price and gross margin. There is no need to cut the base-case revenue assumption to an extreme. Even at only 70% of that assumption, with other assumptions unchanged, base-case valuation would fall from about HKD 390 to about HKD 273. Compared with today's HKD 519, this is well beyond “slightly expensive.” It looks more like “if one brick loosens, the whole wall falls.”
Put more plainly: if over the next three years there is almost no earnings growth and the company remains in a “high revenue, low profit” stage, while the market gives it a price-to-sales range closer to mature platforms such as Hygon, a return to about HKD 165 would not be absurd. That corresponds to an annualized return of about -32% over three years. This return does not need a point-by-point comparison with any country's 10-year government bond to show that buying today offers almost no margin of safety. My conclusion is four words: no margin of safety.
Risks, Catalysts, and Horizontal-Vertical Synthesis
Risks That Must Actually Be Watched
The first risk is revenue growth stalling while the market admits it only late. I assign medium probability and high impact under a high valuation. The most valuable thing in Iluvatar CoreX today is the idea that inference volume can continue. If revenue growth in the next two quarters clearly slips below 30%, or if inference growth depends on more aggressive price cuts to gain scale, margins will be eroded first, valuation multiples compressed next, and the share price hit on both fronts. Observable indicators are quarterly revenue growth, gross margin, and whether there is another pattern of “adjusted loss narrowing while reported loss structure worsens.”
The second risk is float structure turning from friend to enemy. I assign high probability and high impact. Around 2026-07-07, cornerstone lock-up expires. Around 2027-01-07, the one-year lock-up on a large amount of H shares held by existing shareholders expires. The allotment results and shareholder-meeting documents both show that this company's early post-listing share-price move was built on extremely thin float. For this type of stock, lock-up expiry changes the pricing mechanism rather than merely increasing supply. Previously, a small amount of stock was chased higher. Later, early holders may seek exit liquidity. If the first July lock-up round coincides with mediocre results, the valuation's squeeze premium can disappear quickly.
The third risk is a change in list status or broader tightening of supply-chain compliance. I assign medium probability and high impact. Iluvatar CoreX's most visible relative advantage now comes precisely from not yet being officially named on the Entity List. The company does not fully control this advantage. Once external rules change, the market's reason for paying a premium is directly cut off. The transmission path is short: supply-chain expectations worsen first, then customer risk appetite falls, and finally the valuation multiple is forced back toward that of an ordinary loss-making semiconductor company.
The fourth risk is that unit economics remain opaque for a long time. I assign medium probability and medium-high impact. Public disclosure is insufficient to verify the relationship between unit-card BOM and selling price, but the market is already betting at a high valuation that inference volume will not materially sacrifice margins. If later disclosures show that inference margins are more fragile than training margins, that solution revenue merely pulls hardware profit forward, or that project-based revenue is not sustainable, today's high multiple will have to be re-rated. Investors should watch not only revenue itself, but also gross margin and changes in the solution revenue mix.
The fifth risk is governance and incentives. I assign medium probability and medium impact. The current governance structure is a joint setup between platform-style shareholders and management, not a typical single-founder strong-control model. The advantage is strong financing and resource integration. The downside is that external shareholders find it harder to determine who ultimately bears responsibility for long-term value. In addition, the company has proposed an H-share incentive scheme. Although no grants were planned as of the latest practicable date, future dilution risk is real. For a high-valuation loss-making company, dilution hurts more than it does for a mature company.
Positive and Negative Catalysts
There are only three most important positive catalysts. First, revenue growth remains high for two consecutive quarters in 2026 while gross margin does not fall. Second, the July cornerstone lock-up expiry is smooth, and the market finds actual selling pressure far below feared levels. Third, the company continues to remain off the Entity List, and more mainstream application vendors publicly disclose adaptation or deployment. Each would strengthen the market's judgment that it is moving through the validation phase rather than being only a sentiment stock.
The negative catalysts are also very clear. First, guidance or order disclosures weaken. Second, after the July lock-up expiry, trading volume rises but price support is insufficient. Third, external list or supply-chain rules change. Fourth, competitors launch a more aggressive price war in inference. Given the current valuation structure, the most dangerous setup is bad news overlapping with the lock-up window, not bad news itself.
Tracking Dashboard
| Metric | Current observation | Normal range | Warning threshold |
|---|---|---|---|
| Quarterly revenue YoY | High growth continuing | >40% | <30% |
| Consolidated gross margin | 54.0% in 2025 | 50%-55% | <48% |
| Adjusted net loss | Narrowed in 2025 | Continues to narrow | Expands for two consecutive periods |
| July lock-up realization intensity | Cornerstone 6-month expiry | Moderate turnover | Volume-driven decline and continued drops |
| January lock-up realization intensity | Existing shareholders' one-year H-share lock-up expiry | Orderly clearing | Excessively concentrated supply |
| List status | Entity List not found | Unchanged | Official new entry appears |
| Ecosystem adaptation disclosure | Already in mainstream domestic compatibility lists | New partnerships | No new public cases for a long time |
| TTM price-to-sales | About 110x | More reasonable below 70x | Stays above 100x for long |
The most important item to watch in this table is whether high growth and a high multiple can both stand, not the share price itself. If any one of revenue growth, gross margin, or list status slips, the current valuation will be hard to sustain.
Horizontal-Vertical Synthesis
Viewed vertically, what Iluvatar CoreX has truly proved is that in an industry that burns extreme capital, depends heavily on supply-chain coordination, and is constantly disturbed by external rules, it can build products, sell them, and push the company to listing. It has done more than tell a domestic GPU story. This requires technical capability, financing capability, and the ability to keep bridging customers, software, hardware, and capital markets. The tailwinds in its past success are obvious: AI and domestic substitution gave it strong wind at its back. But management and shareholder resource integration was also important. Otherwise, it could not have survived from 2015 to 2026 and reached the Hong Kong Main Board at the hottest moment.
Viewed horizontally, its most real advantage versus peers is that it is “capable enough and not yet pinned down by the list,” rather than absolute technological leadership. This advantage is very real today because customer procurement does not look only at theoretical peak performance. It also considers compliance, delivery, adaptation, budget, and project cycle. Iluvatar CoreX sits neatly in the gaps between several stronger rivals: more local than NVIDIA, cleaner than some domestic peers, and more like a genuine GPU company with self-developed chip capability than many pure project-solution vendors. This is not an end-state advantage, but it is enough for the company to keep winning orders over the next two to three years.
The problem is that the stock price now rewards far more than this real advantage. HKD 519 assumes not only that the company can keep growing, but also that it will keep growing rapidly before float loosens, preserve gross margin before competition intensifies, and maintain its list-status advantage before rules change. This pricing treats the outcome as if it were already the process. For a GPU company with just over RMB 1.0 billion in revenue, still losing money, and listed for only five months, the current price is pre-consuming the future rather than simply rewarding past success.
The most important variable over the next year is inference-business growth and actual selling pressure from the July lock-up expiry. Over the next three years, the key variable is whether the company can push its revenue base close to RMB 3.0 billion and make the market start viewing it as a domestic GPU company with a path to profit, rather than an extremely scarce newly listed theme stock. Over the next five years, the key variable is whether it can move beyond the accident of list-status advantage in software ecosystem and customer stickiness, forming a truly sustainable product moat. If these variables are delivered, it can slowly change from a theme stock into a growth stock. If not, it can move from the type of stock that rises fastest into the type that draws down deepest.
Under what conditions would it become a better investment target? The conditions are specific. First, the share price returns to a range with a margin of safety. Second, subsequent 2026 quarters continue proving that inference volume is not a one-off project recognition. Third, after the July and January lock-up rounds, the float structure clearly normalizes while turnover and price support remain relatively stable. Only when fundamental progress and normalized float structure happen together does Iluvatar CoreX deserve a long-term holding discussion. Conversely, if revenue growth stalls, gross margin continues to fall, external list status changes, or the share price breaks sharply once lock-up expiry arrives, this research framework should be overturned rather than forcing a long-term narrative.
Bull and Bear Cases
Bull case:
Real inference volume allowed the company to show high growth and narrowing losses at the same time for the first time, which matters more than simply releasing new products.
As of the base date, no Entity List entry was found, giving it a scarce supply-chain and compliance narrative among domestic GPU peers.
Ecosystem partners such as 4Paradigm have placed Iluvatar in mainstream domestic compute compatibility lists, showing the product has moved beyond the sample stage.
There are very few tradable domestic general-purpose GPU pure plays in Hong Kong, and scarcity itself brings a capital-allocation premium.
Bear case:
Current market cap is about HKD 132.0 billion, corresponding to only about HKD 1.197 billion of 2025 revenue and about 110x TTM price-to-sales, meaning valuation has deeply pre-consumed future execution.
The company is still loss-making and had no distributable profit in 2025. It remains a growth company validating its profit path, not a cash creator.
The 2026-07 and 2027-01 lock-up expiries will reshape supply and demand, with especially large impact on a newly listed stock with extremely thin float.
True ecosystem dominance is not in Iluvatar CoreX's hands. NVIDIA owns the international software ecosystem, Huawei has government-enterprise system capability, and Hygon is the profitable platform benchmark.
Its largest current relative advantage comes from not yet being included on the list, and that is not a permanent moat.
Pre-mortem
The first 50% loss scenario could easily occur within the next seven months. After the July 2026 cornerstone lock-up expiry, the market discovers actual selling pressure is higher than expected. Soon after, the interim or third-quarter report shows revenue growth slipping to 30%-35% and gross margin falling below 48%, making the market realize that inference volume is more about trading price for scale than trading scale for profit. The valuation framework would then compress quickly from more than 70x forward price-to-sales to 40-45x forward price-to-sales, and a share-price fall to HKD 250-300 would not be exaggerated. Relative to the current HKD 519, that is already close to a halving.
The second scenario is slower but more lethal. In the second half of 2026 through 2027, if U.S. export controls continue tightening or the company is placed on a new restriction list, customers will immediately reassess deployment risk, while supply-chain costs and delivery visibility both worsen. At the same time, the one-year lock-up on existing shareholders expires in January 2027, releasing a large amount of supply. At that point, the market no longer gives it a “list-status advantage + low float” premium, and prices it only as a domestic GPU company that has not yet achieved stable profitability. If the multiple compresses to around 35x price-to-sales, the corresponding share price could return to about HKD 160-200, implying a drawdown of more than 60%.
Final Research Conclusion
Iluvatar CoreX is a company worth serious research, but it is not a stock that investors can embrace at any price simply because the narrative is attractive. It has proved it is not a hollow story: products are selling, revenue is growing, inference is scaling, losses are narrowing, and in the domestic general-purpose GPU camp it does occupy a very valuable position, with chip capability, software adaptation, and no list blockage for now. The problem lies at the stock level. The price the market gives it has already pushed the value of that position close to an extreme.
What I truly worry about is investors buying the right industry trend at the wrong point in time, not the company failing to build the product. The current HKD 519 price almost requires the company to complete four difficult things at once over the next year: maintain high revenue growth, keep gross margin from falling, avoid lock-up-driven selling pressure, and keep list status unchanged. If any one of these fails, the drawdown will not be small. Conversely, this company is not permanently unbuyable. If later quarterly reports continue proving inference volume, the float structure clearly normalizes after the July and January lock-up rounds, and the share price returns to a more reasonable margin-of-safety range, the research conclusion can be upgraded.
【Company Profile Scores】
Fundamental quality: Medium
Growth: High
Moat: Medium
Financial resilience: Weak
Management credibility: Medium
Valuation attractiveness: Low
Risk level: High
Suitable investor type: Not suitable for ordinary investors
【Investment Rating】
Rating: Avoid
One-line investment thesis: Inference volume is real progress, but HKD 519 already pre-consumes two to three years of delivery.
【Ideal/Fair Buy Price】HKD 190-255 Basis: corresponds to the fair value under the conservative operating scenario after applying at least a 20% margin of safety.
Acceptable holding price: HKD 328-390
Clearly overvalued price: Above HKD 518
Current price category: Clearly overvalued
Worth waiting for a better price: Yes. The trigger would be a price below HKD 255, with at least one subsequent financial-reporting cycle showing continued high growth and stable gross margin. The opportunity cost of waiting is that if the stock remains strong after lock-up expiries, the share price may continue to stay high. But I think this is the cost that must be paid to avoid permanent capital loss.
Target holding period: If it later enters the buy range, reassess from a 1-3 year perspective. Building a position at the current price is not recommended.
Expected annualized return: Conservative about -54% to -39%; base about -37% to -25%; bullish about -17% to -9%.
Maximum loss risk: If revenue growth stalls, list status changes, and lock-up selling pressure is released in a concentrated way, a 50%-65% drawdown within 12-18 months would not be extreme.
Signals that would trigger reassessment: first, revenue YoY remains above 40% for two consecutive quarters; second, gross margin stays above 50%; third, no high-volume breakdown occurs after the July and January lock-up expiries; fourth, more mainstream application parties publicly disclose deployment or adaptation; fifth, any official change appears in external list status.
【Valuation Range】
current: 519 (as of the 2026-06-10 close)
bear (conservative · ideal buy range): [190, 255]
base (reasonable · acceptable holding range): [328, 390]
bull (optimistic · clearly overvalued line): [472, 518]
Key Data Table
The table below collects the most important figures in the report for follow-up tracking.
| Metric | Value | Note |
|---|---|---|
| IPO price | 144.6 HKD | 2025-12-30 prospectus framework |
| IPO offering size | 25.43 million shares | Small issuance |
| First-day opening price | 190.2 HKD | 2026-01-08 |
| Current closing price | 519 HKD | As of 2026-06-10 |
| Current market cap | 131.99 billion HKD | As of 2026-06-10 |
| 2025 revenue | RMB 1.034 billion | Equivalent to about HKD 1.197 billion |
| 2025 gross margin | 54.0% | Annual-report framework |
| 2025 adjusted net loss | RMB 438.8 million | Annual-report framework |
| Shares issued at end-2025 | 254,317,736 shares | Including 245,101,965 H shares |
| 6-month lock-up observation window | Around 2026-07-07 | Cornerstone lock-up expiry |
| 12-month lock-up observation window | Around 2027-01-07 | Existing shareholders' H-share lock-up expiry |
Research Uncertainties
The company's public disclosures do not provide sufficiently granular unit-card shipments, ASP, and BOM, so outside discussion of “unit-card economics” cannot yet be fully proven or disproven with primary data.
Actual changes in southbound holdings after Stock Connect inclusion require higher-frequency broker-seat and holding data. Public webpages show list changes, but cannot fully reconstruct the capital structure.
“Not on the Entity List” is the status as of the base date, not a future commitment. This type of information is highly time-sensitive.
As a newly listed stock with only five months of trading history, secondary-market price is heavily affected by float and liquidity. Historical volatility samples are far from sufficient to support traditional statistical regression.
For private unlisted competitors such as Moore Threads, MetaX, and Enflame, public financial granularity is limited. The horizontal comparison in this report is more about industry position than strict financial comparability.
Reference Sources
HKEXnews: prospectus, allotment results, 2025 annual report, and shareholder-meeting circular.
U.S. eCFR 15 CFR Part 744 Supplement No. 4 Entity List.
Shanghai Stock Exchange Hong Kong Stock Connect constituent adjustment page and Hong Kong market news.
Reuters news and market pages.
Morningstar, Yahoo Finance, Investing, and other market data pages, used for cross-checking market cap, valuation, and exchange rates.
Domestic compute adaptation disclosures in 4Paradigm's annual report.
Other Companies Mentioned in This Report
06082.HK — Biren Technology, the most direct domestic GPU comparable in Hong Kong, and also the Entity List comparison group.
688041.SHG — Hygon Information, a profitable domestic compute platform company, used to compare earnings quality and valuation center.
688256.SHG — Cambricon, a high-beta A-share AI chip leader, used to compare market-cap tier and maturity.
NVDA.US — The global absolute benchmark for training GPUs and software ecosystem.
AMD.US — The international second supplier, used to compare the market space and valuation of a non-NVIDIA route.
06682.HK — 4Paradigm, both an ecosystem adaptation party and supporting evidence for Iluvatar CoreX's software usability.
00763.HK — ZTE Corporation, an important reference for cornerstone and government-enterprise channel narratives during the prospectus stage.
This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.
Full report
Sign in to read the full report
Sign up free to unlock the full text, the Baillie growth scorecard, and full-text search.
Log in / Sign up free