In-Depth Research on MetaX

From Product Validation to Commercial Scaling

The company's evolution over the past few years looks more like a long-cycle systems-engineering effort than a consumer electronics company dependent on one blockbuster generation. Around 2023, MetaX began to lay out its product lines: the XiSi N series for inference, the XiYun C series for training and general-purpose computing, and the XiCai G series for graphics rendering, with MXMACA as the supporting software stack. Public technical inquiry responses state clearly that the company's core delivery form is GPU boards, and MXMACA is mainly used together with GPUs and is not sold separately to external customers; in other words, the software is currently more a necessary condition for hardware sales than an independent profit center.

By 2024, the company began turning "can build" into "can deliver in volume." Revenue rose to 743 million yuan in 2024, then doubled again to 1.644 billion yuan in 2025. More important is the revenue structure: SSE inquiry responses show that during 2022-2024 and Q1 2025, the XiYun C-series training-inference integrated GPU products contributed 94.31% of main-business revenue across the three years and one quarter. This means the real engine of MetaX's current commercialization is the C series taking the lead, while the N and G series are more about completing the map, expanding scenarios, and maintaining the "full-function GPU" narrative, rather than all three lines flying together. In 2025, IP, technical services, and other revenue was only 13.339 million yuan. Although the gross margin was as high as 98.22%, the scale was very small and could not change the fact that "at this stage, the essence is still making money by selling GPU hardware."

Listing, Extreme Pricing, and Share-Supply Structure

The IPO was the first major test of MetaX's commercial story, and the first time the stock narrative and operating reality were forcibly pulled onto the same table. The listing announcement shows that the company issued at 104.66 yuan, issued 40.10 million shares, and had post-issue total share capital of 400.10 million shares; the IPO price corresponded to a diluted pre-extraordinary 2024 price-to-sales ratio of about 56.35x, with gross proceeds of about 4.197 billion yuan, exceeding the originally planned fundraising size. The proceeds are mainly allocated to new high-performance general-purpose GPUs, next-generation AI inference GPUs, and frontier technology R&D. For a GPU company still in a loss-making phase, the financial meaning of this IPO is direct: it pushes the question of "will it need financing again soon" back by at least several years.

But the secondary market's response was much more aggressive than the primary market's. Reuters reported that MetaX's share price rose about 700% on its first trading day, as the market quickly elevated it into one of the symbols of "China AI chip self-reliance." By the research base date, the stock still closed at 694 yuan, corresponding to a market cap of 277.669 billion yuan, about 6.6 times the IPO market cap. The issue is that this pricing rests on an extremely small free float. The listing announcement shows that only 18.138973 million shares were tradable on the first day, accounting for 4.53% of total share capital. This is a share-supply structure in which a very small amount of incremental capital can move the price meaningfully, and it is hard to understand with traditional market-cap intuition.

Financial Review, Share-Price History, and Lock-Up Windows

The financial trajectory is attractive, but the rhythm is not smooth. The annual-report summary shows that revenue rose from 53 million yuan in 2023 to 743 million yuan in 2024, then to 1.644 billion yuan in 2025; losses narrowed from 1.409 billion yuan in 2024 to 789 million yuan in 2025. On the surface, this is a typical path of "revenue ramping, losses narrowing." But once broken down by quarter, the shape looks different: revenue in the four quarters of 2025 was 320 million yuan, 595 million yuan, 321 million yuan, and 408 million yuan, respectively; net profit attributable to shareholders was -233 million yuan, 47 million yuan, -160 million yuan, and -444 million yuan, respectively. Only Q2 briefly turned positive, while operating cash flow was negative in all four quarters. This means the company's delivery and revenue-recognition cadence still has project-type volatility, and a stable, repeatable quarterly profit model has not yet formed.

A widely circulated market misreading must also be corrected here. When interpreting the 2025 annual report, several media outlets wrote that net cash flow from operating activities was "1.26 billion yuan"; but the annual-report summary's original figure is -1,260,071,793.82 yuan, or -1.260 billion yuan, only improved by 888 million yuan versus 2024's -2.148 billion yuan. The media mixed "improved by 888 million yuan" with "net outflow of 1.260 billion yuan" into "inflow of 1.260 billion yuan." This difference is crucial, because it directly determines whether you are looking at a company that has "already begun self-funding" or a company whose "revenue has surged but still consumes working capital." Based on primary disclosure, the answer is the latter.

The lock-up window further magnifies the valuation issue. The company announced on 2026-06-10 that 389,622 offline placement restricted shares would become tradable on 2026-06-17, accounting for about 0.10% of total share capital; the listing announcement also disclosed another 14.355608 million shares that will enter the 9-month lock-up expiry around 2026-09-17, accounting for about 3.59% of total share capital. For the 12-month tranche around 2026-12-17, the listing announcement can at least confirm that 6.649945 million strategic-placement shares will expire then, accounting for about 1.66% of total share capital. In addition, multiple pre-IPO shareholders' holdings were split into "12-month + 36-month" lock-up segments, so the actual scale of the 12-month expiry will be higher than this number. In other words, before the end of 2026, the free float will at least rise from 4.53% at listing to close to 10%, rather than staying permanently scarce.

The table above only counts the minimum scale that can be clearly verified from primary documents as of the base date, so it is more of a "conservative floor" than a "final total." But even on this floor, share supply is no longer static. For a high-valuation, still-unprofitable recent IPO with an originally tiny free float, lock-up expiry is part of the pricing model itself, not a side issue.

Business Model and Industry Landscape

How This Business Machine Actually Runs

MetaX's revenue core today is clear: it sells training-inference integrated GPU boards, and around those boards sells servers, racks, interconnect, and software adaptation. The SSE inquiry response is blunt: the company's main business is R&D, design, and sales of full-stack GPU products used in AI training and inference, general-purpose computing, and graphics rendering. Its core delivery form is GPU boards, and the MXMACA software stack is mainly used to improve ease of use and is not sold separately. In other words, the company is a typical hard-tech supply-side model: first win orders through hardware, then use software and cluster integration to complete deployment, rather than "first earn money from software subscriptions, then pull hardware along."

This also means its cost structure has two characteristics. First, the truly heavy parts are R&D, EDA, tape-out verification, board and system adaptation, customer onboarding, and ecosystem building; capital expenditure is not the main burden because it is a design company, not a foundry. In 2025, R&D investment was 1.027 billion yuan, equal to 62.49% of revenue and still up 14.04% year on year; in Q1 2026, R&D investment was 253 million yuan, equal to 45.09% of revenue. Second, as shipments scale, working capital will consume cash first. At the end of Q1 2026, accounts receivable were 933 million yuan, prepayments were 1.233 billion yuan, and inventory was 1.422 billion yuan, totaling about 3.588 billion yuan; meanwhile, contract liabilities fell from 116 million yuan at the end of 2025 to 58 million yuan. This suggests MetaX is more like "to scale, first push goods, materials, and supply-chain allocation in," rather than a light-cash model that "collects large prepayments first, then delivers gradually."

From an operating-leverage perspective, once orders expand, this model does have room to improve gross margin and expense ratios. In 2025, gross margin rose to 56.51%, up 3.02 percentage points from 2024; the direct reason losses narrowed was the sharp revenue growth. The problem is that operating leverage has not yet flowed through to cash flow. 2025 operating cash flow was still -1.260 billion yuan, and Q1 2026 was still -550 million yuan. That determines how MetaX should be valued: "margins may ultimately improve" cannot be treated as a fact already realized today.

Moat, Governance, and Truly Effective Advantages

The market repeatedly emphasizes many moats for MetaX, but in my view only four can potentially survive 3-5 years of competition.

The first is the engineering team. The AMD background of founder Chen Weiliang and the core R&D team is a starting condition for how this company got here, not a marketing slogan. A GPU is not a product that can be rushed out through one isolated innovation. Architecture, compiler, drivers, interconnect, thermals, board-level stability, and customer adaptation are all indispensable. MetaX's most valuable asset today is still this team that can build and deliver complex systems. By the end of 2025, the company had 675 R&D employees, 73% of total headcount, and 401 domestic authorized patents and software copyrights, including 350 invention patents.

The second is the route choice of "general-purpose GPU + integrated hardware and software." The company did not lock itself into a single inference ASIC, but has insisted on a general-purpose GPU architecture, which makes it better suited to fast-changing algorithms and multi-scenario migration. In the SSE inquiry response, the company also defined the meaning of GPUs versus ASICs as "more general-purpose and more flexible, able to adapt to rapidly iterating complex cloud algorithms and many non-traditional model structures." This route is difficult to develop and capital intensive, but once commercialization works, customer relationships and deployment experience will be much stronger than those of "single-model customized chips."

The third is that it has crossed the threshold from laboratory to cluster deployment. In 2025, cumulative GPU product sales exceeded 55,000 units, training-inference integrated GPU board sales reached 33,600 units, up 147.31% year on year, and products had been deployed in more than 10 intelligent computing clusters, covering national AI public compute platforms, operator intelligent compute platforms, and commercial intelligent computing centers. These are not monopoly barriers, but they are real thresholds: latecomers without mass production and deployment experience will find them hard to replicate quickly.

The fourth is that as of the base date it had not entered the BIS Entity List. Searches for "MetaX" and "沐曦" in the official eCFR Entity List text returned no matches, while Moore Threads and Biren were formally added to the Entity List by BIS in October 2023. This does not mean MetaX has no geopolitical risk, nor that it will not be added in the future; but at least as of 2026-06-11, it objectively had more room to operate in supply-chain compliance than listed domestic peers.

By contrast, some "moats" amplified by the market are still too early. For example, the three-line product matrix itself cannot yet be considered a real moat, because commercial revenue still mainly comes from the C series. High valuation and a high share price are merely the market result of scarce shares, not a business barrier. On governance, public annual reports show that Chen Weiliang, Shanghai Jiaomai, and Shanghai Xiji are concert parties, with combined holdings of about 20.64%; ownership is not extremely concentrated, and the top ten shareholders include national capital, local capital, VCs, and individual large shareholders. The benefit of this structure is strong financing ability and broad resources; the drawback is that after listing, different investors' exit timing may not align. At least before the base date, public materials did not show major financial fraud, a modified audit opinion, or major governance scandals, but "complex shareholder structure and dense future lock-up expiries" itself will create a liquidity discount beyond any governance discount.

Industry Structure, Cyclicality, and Policy Variables

MetaX operates in a market jointly shaped by AI infrastructure investment, supply-chain localization, and geopolitics, not an ordinary semiconductor sub-sector. IDC data show that China's accelerated server market reached USD 22.1 billion in 2024, up 134% year on year, with GPU servers still accounting for 69%; by the first half of 2025, China's accelerated server market had further reached USD 16.0 billion, doubling year on year, with local accelerator chips at about 35% share. Behind this is the joint push of model-scale expansion, operator and cloud-provider capex, local intelligent computing center construction, and policy direction, not the penetration of a single technology.

Looking at profit-pool distribution, the fattest global layer is still captured by Nvidia. Nvidia's fiscal 2026 revenue was USD 215.938 billion, net income was USD 120.067 billion, and GAAP gross margin was 71.1%; these figures show that the GPU industry's core excess profit lies in who controls architecture iteration, system adaptation, and software-ecosystem pricing power, not merely in whether one can make chips. The reality in China is more complex: Reuters, citing IDC, said domestic GPU/AI chip vendors had already taken about 41% of China's AI accelerator card market in 2025, while Nvidia's share had shrunk to about 55%. In other words, domestic substitution is indeed happening, but it is far from a stage where one company dominates alone.

In cyclicality, MetaX overlays four cycles at once. It has a semiconductor cycle, because supply and demand, foundry capacity, and customer purchasing sentiment all fluctuate. It has a capital-expenditure cycle, because intelligent computing centers and cloud-provider budgets are core demand sources. It has a policy cycle, because domestic substitution, compute infrastructure, and local project reviews directly affect order cadence. It also has a technology-iteration cycle, because model-route changes determine the structure of training and inference compute. It is less like a traditional consumer-electronics inventory-cycle stock. IDC's description of the global server market also illustrates the point: in Q4 2025, global server revenue hit a record high, AI infrastructure construction remained the dominant force, and no clear digestion period was visible.

Policy is both tailwind and constraint for MetaX. The tailwind is that China's top-level AI deployment is clearly strengthening: the State Council's 2025 "Opinions on Deeply Implementing the 'AI+' Action" proposed strengthening intelligent-computing interconnection and supply-demand matching; eight ministries including MIIT released the "'AI + Manufacturing' Special Action Implementation Opinions" in early 2026, making "product breakthroughs" and "ecosystem expansion" explicit tasks; and CAICT estimated that China's core AI industry exceeded 900 billion yuan in 2024 and exceeded 1.2 trillion yuan in 2025. The constraint is also clear: policy has begun to emphasize intensive compute construction and avoiding low-level duplication; at the same time, preference for domestic chips is rising, and Reuters has reported that China gave guidance for new data centers receiving fiscal support to use domestic AI chips. This means MetaX enjoys track expansion, but it will also compete in an environment more tilted toward project-based, policy-guided, and customer-concentrated demand.

Horizontal Peer Analysis

Why I Chose This Comparable Set

For MetaX, a horizontal analysis based on a pure specification table of several products has limited value. Its comparable group should be divided into three layers.

The first layer is the global profit-pool anchor, mainly Nvidia and AMD. They are not same-level rivals that MetaX can replace in the near term, but they define the industry ceiling, software-ecosystem threshold, and international valuation reference. The second layer consists of domestic listed companies that the capital market actually compares it with, namely Cambricon and Hygon Information. Cambricon is closer to a "pure-play Chinese AI chip scarcity asset with a high valuation," while Hygon is more like a "domestic compute foundation that has already achieved scale and profits." The third layer is Huawei Ascend, unlisted but essential on the competitive map, plus Moore Threads and Biren, which have been placed on the U.S. Entity List. The former represents the highest pressure of domestic system-level capability, while the latter two represent players most similar to MetaX but facing heavier supply-chain restrictions.

What Each Company Has Become

Nvidia has become the "profit-pool definer." Its significance is not only that its chips are fast, but that it has tied AI chips, systems, interconnect, and development environment into one whole, which is why its revenue and profits far exceed peers: fiscal 2026 revenue of USD 215.938 billion, net income of USD 120.067 billion, and market cap around USD 4.85 trillion near 2026-06. The reason to use it as a comparison is to remind investors that the companies that really make money in GPUs are those that form a whole system others are reluctant to migrate away from, not simply those that "make a chip." This does not mean MetaX will replicate that path.

AMD has become "number two, but not a side note." It uses EPYC CPUs to hold the data-center platform entrance while using Instinct GPUs to keep chasing the AI training market. Its official homepage even presents the complete AI portfolio of "CPU + GPU + adaptive computing" as a core selling point. Its 2025 revenue was USD 34.639 billion, and its market cap was about USD 760.5 billion in 2026-06. For MetaX, AMD's lesson is that without CPUs, motherboards, system partners, and a long-term software stack, GPU alone makes it hard to build true platform stickiness in the enterprise market. Valuation multiples are secondary. One of MetaX's current largest shortcomings is precisely that it is still far from this breadth of platform-level capability.

Cambricon has become the A-share market's "pure AI chip scarcity asset." In 2025, it had revenue of 6.497 billion yuan and net profit of 2.059 billion yuan, already completely different from a few years earlier; but Reuters data also show that its 2025 operating cash flow was still -498 million yuan, indicating that even after reaching profitability, project cadence and cash-flow quality in this industry can still mismatch. The capital market gives it a valuation based on more imagination and emotion than Hygon. The easiest peer for MetaX today is actually Cambricon, not Nvidia: both are domestic AI chip names, both carry a high narrative, high volatility, and high price-to-sales ratio, except Cambricon has a larger revenue base and a more mature income statement.

Hygon Information is another kind of sample. It had 2025 revenue of 14.377 billion yuan, net profit of 2.545 billion yuan, operating cash flow of 2.097 billion yuan, and a market cap of about 623.3 billion yuan. It makes CPUs and DCUs, following a more mature path of "platform foundation + government and enterprise deployment," with cash flow and profitability clearly better than MetaX. Putting Hygon and MetaX side by side reveals a key difference: Hygon has grown from a "domestic substitution concept" into an "operating asset," while MetaX is still in the first half of the transition from concept to asset.

Huawei Ascend is unlisted, but it is the rival MetaX can least ignore. Reuters, citing IDC, showed that in 2025 China's domestic AI accelerator card shipments were led by Huawei at 812,000 cards, far above most other domestic chip vendors. This difference comes not only from single-chip capability, but also from complete machines, servers, software partners, industry projects, and government-enterprise relationships. MetaX's real position is more like a challenger: it is closer to engineering-led mass production than most startups, but it has not yet formed a system like Huawei's that can bring its own demand.

Peer Data Comparison and the Business Reasons Behind the Differences

The table above is roughly organized by public market cap and latest annual revenue. The purpose is to see how the capital market prices companies at different growth stages, not to create a precise comparable-company model. Data sources include company annual-report summaries, quarterly reports, Reuters, and Yahoo Finance; MetaX's price-to-sales ratio is calculated from the 2026-06-11 closing market cap and 2025 revenue.

The most important conclusion from this table is "what the market is paying for," not "who is expensive or cheap." Hygon's 43x price-to-sales ratio is already not cheap, but behind it are 14.4 billion yuan of revenue, 2.5 billion yuan of profit, and positive operating cash flow; Cambricon's price-to-sales ratio above 100x at least has a larger revenue base and positive profit behind it; while MetaX supports a market cap near 280 billion yuan with 1.644 billion yuan of revenue. In essence, it is using the earliest commercialization results to trade the platform value that domestic high-end GPUs may form over many future years. It may of course grow, but today's price is already paying for much more than the 2025 and 2026 statements themselves.

From an ecological-position perspective, MetaX is a "high-end challenger" in domestic high-performance GPUs. It is neither the leader nor a low-end substitute follower: in engineering capability, product completeness, and capital-market recognition, it is clearly above most unlisted startup teams; but in software ecosystem, customer breadth, system capability, and scaled profitability, it is clearly weaker than Nvidia and also weaker than domestic names closer to platform companies, such as Huawei and Hygon. The benefit of this position is huge upside; the drawback is that it is easy to overvalue. When technological substitution, price war, or policy tightening arrives, leaders usually capture profit, followers usually compete on cost, and challengers often absorb valuation pressure first.

Current Fundamentals, Valuation, and Risks

Latest Four Quarters and Current Bull-Bear Debate

The latest fundamentals continue the combination of "high revenue growth, narrower losses, and still-tight cash." In Q1 2026, revenue was 562 million yuan, up 75.37% year on year; net loss attributable to shareholders was 99 million yuan, versus a loss of 233 million yuan in the same period last year; adjusted net loss after non-recurring items was 103 million yuan. The official reason for revenue growth was a significant increase in GPU product shipments, indicating that demand has at least not stopped. But cash flow did not improve in sync: net cash flow from operating activities was -550 million yuan, slightly worse than -531 million yuan in the same period last year; accounts receivable and prepayments remained high, while contract liabilities fell from 116 million yuan to 58 million yuan. In other words, the most pleasant part of the statements is the income statement, while the most critical issue remains cash conversion.

This company is not in a near-term liquidity crisis, however. At the end of Q1 2026, cash and cash equivalents were 4.111 billion yuan, trading financial assets were 4.435 billion yuan, and debt investments were 736 million yuan, totaling about 9.283 billion yuan; total liabilities were only 460 million yuan, and the statements showed no interest-bearing short-term or long-term borrowings. Based on a rough Q1 operating cash outflow of about 550 million yuan, its highly liquid assets are enough to cover more than 4 years of burn at the same pace. This is where MetaX is most likely to distort judgment: it is a startup with enough money to burn but a share price that may already have priced in many years of future success, not a startup short of cash.

The bulls' strongest evidence today has three parts. First, the revenue ramp is real: revenue grew by more than 30 times from 2023 to 2025 and was still growing in Q1 2026. Second, product scaling is real: C-series sales rose sharply in 2025, and deployments entered operators, public compute platforms, and commercial intelligent computing centers. Third, the balance sheet is extremely clean, and the IPO pushed financing anxiety far back. The bears' strongest evidence also has three parts. First, revenue concentration is high, the C series remains the absolute mainstay, and the software stack does not monetize independently. Second, profitability and cash flow have not yet synchronized: operating cash flow was negative in all four quarters of 2025 and remained so in Q1 2026. Third, the stock is expensive enough that it requires future share gains, margins, and ecosystem breakthroughs that may only be realized years from now. The hardest question here is what price deserves to own this future, not whether the company has a future.

Valuation Analysis

Conclusion first: for MetaX, PE, EV/EBITDA, and FCF Yield are not good tools today. The reason is direct. From 2023 to 2025, operating cash flow was -1.017 billion yuan, -2.148 billion yuan, and -1.260 billion yuan, respectively, while net profit was -871 million yuan, -1.409 billion yuan, and -789 million yuan, respectively. The operating-cash-flow/net-profit ratio has little meaning because both are negative. Although the company is fabless and traditional capex pressure is not high, it must continue investing in R&D to maintain competitiveness. In economic substance, R&D is more like "quasi-capex" than a discretionary expense that can be freely cut. Therefore, using accounting profit to calculate PE already mistakes the heaviest ongoing investment for future freely distributable profit.

A more suitable approach is to use "next 12 months revenue x comparable price-to-sales ratio" for range valuation, then apply discounts for cash flow and lock-up windows. The current share price of 694 yuan corresponds to a total market cap of 277.669 billion yuan; based on 2025 revenue, the static price-to-sales ratio is about 168.9x. Even simply annualizing Q1 2026 revenue leaves the price-to-sales ratio above 123x. This multiple is not only far above Nvidia and AMD at about 22x, and Hygon at about 43x, but even above Cambricon's rough 2025-revenue-based 119x. The market is giving MetaX a premium for being "earlier, smaller, but scarcer." The issue is that a scarcity premium can hold for a period, but usually cannot ignore lock-up expiries, incremental supply, and the speed of profit delivery indefinitely.

I use three scenarios, with only three core variables: 2026 revenue growth, whether gross margin can stay above 50%, and whether the market is willing to continue giving it a price-to-sales ratio far above mature GPU companies.

The table above is a price stress test under the research framework and does not constitute investment advice. Because the company is still loss-making and cash flow has not yet worked, the 12-month returns under all three scenarios are unattractive at the current price. What truly determines whether it is worth buying is whether the market has already taken most of the future growth reward upfront, not whether the company will grow. Combining the assumptions above, my judgment on the 12-month price signal is: 170-220 yuan is the ideal buy zone, 280-380 yuan is a reasonable hold zone, and 500-650 yuan is already within the range covered by optimistic expectations. Above that, the market needs to keep replacing fundamental logic with extreme scarcity logic.

The margin-of-safety review is also direct. The current price is at a significant premium to my conservative scenario's implied value, so the margin of safety is zero. The most fragile assumption is that "the market continues to accept a forward price-to-sales ratio above 40x, even above 60x," because this is the part most dependent on liquidity and narrative sentiment, rather than on the company's own execution. If the company's profits show zero growth over the next three years, there is almost no annualized return to discuss at this price. More precisely, this looks like a "good company but bad price." Conclusion on margin-of-safety sufficiency: none.

Risks, Catalysts, and Tracking Indicators

For MetaX, I see five main risks that could cause permanent capital loss.

The first is demand and capex mismatch, with medium probability and high impact. Its customers are often tied to intelligent computing centers, operator platforms, government, or large-model infrastructure, and such demand is sensitive to policy, budgets, and project cadence. If compute-center construction shifts in 2026-2027 from "race to build" to "utilization first," order delivery may slow before the market narrative does. Observable indicators include single-quarter revenue growth, contract liabilities, and tendering cadence for intelligent computing centers; once revenue growth falls below 40% while valuation still stays near 100x price-to-sales, the stock price usually reprices before the income statement.

The second is cash-flow risk, with high probability and high impact. The company proved in 2025 and Q1 2026 that revenue growth does not automatically bring cash improvement. Receivables, prepayments, and inventory are high, while contract liabilities are falling, showing that the working-capital structure remains unfavorable. Observable indicators include net operating cash flow, accounts-receivable turnover, and quarterly changes in prepayments and inventory; if operating cash flow remains near -500 million yuan for two consecutive quarters while marginal revenue growth slows, the market will begin questioning the quality of growth.

The third is lock-up expiry and liquidity risk, with high probability and medium-to-high impact. The free float at listing was only 4.53%; this week brings the 6-month tranche, September brings the 9-month tranche, and December at least brings the 12-month strategic-placement tranche. For a high-valuation recent IPO, lock-up expiry is a question of whether the market is still willing to support the same extreme multiple with equally thin shares, not whether shareholders want to sell. Observable indicators include lock-up expiry announcements, turnover, concentration on trading-leader boards, and closing-price sensitivity around expiry windows.

The fourth is competition and technology-route risk, with medium probability and high impact. The domestic market has system-type players such as Huawei Ascend on one side, and Cambricon, Hygon, Pingtouge, and several other domestic chip companies still in the field on the other. Overseas, investors must watch Nvidia's compliant chips and AMD's policy changes in China. Reuters reported that domestic share in China's AI accelerator card market rose quickly in 2025, which means future competition will not become lighter; it will shift from "is there a domestic option" to "whose system cost-performance and adaptation efficiency are higher." If a stronger complete-machine integrated solution appears, or if overseas high-performance compliant products re-enter at scale, MetaX's pricing and gross margin may both come under pressure.

The fifth is geopolitical risk, with medium probability and medium-to-high impact. As of the base date, the company did not appear in the official BIS Entity List text, which is a positive difference; but it also means the risk has merely not landed, not that it is zero. BIS can adjust the Entity List, advanced-computing export rules, and related scope at any time. Observable indicators include new BIS/Federal Register rules, eCFR Entity List updates, and the company's supply-chain risk wording in periodic reports. For a GPU company, once geopolitical risk lands, the impact is often on the entire supply-chain path and customer confidence, not just one quarterly report.

Positive catalysts mainly include four items: revenue continuing to beat expectations, Q2 or Q3 operating cash flow improving meaningfully, more large-cluster deployments being disclosed, and domestic substitution accelerating again after U.S. export rules for high-end AI chips to China change. Negative catalysts are the reverse: lock-up windows increase share-supply pressure, quarterly revenue growth slows, operating cash flow continues to worsen, overseas compliant chips expand supply again, or industry procurement slows after the focus shifts from "building centers" to "improving utilization."

Among these indicators, the most important trio is "operating cash flow + receivables/prepayments/inventory + revenue growth," not the share price. For mature companies, revenue growth can automatically flow through to profit; for a company like MetaX, investors must first confirm whether revenue can flow through to cash before they are entitled to discuss profit margin and valuation center.

Key Data Table, Research Uncertainty, and Reference Sources

Data mainly come from the company's 2025 annual-report summary and 2026 Q1 report. 2024 gross margin is back-calculated from "2025 increased by 3.02 percentage points year on year."

There are four main research uncertainties. First, customer-structure disclosure at the public-summary level is still insufficient, making it hard for investors to accurately judge dependence on a single customer or single project. Second, product-line revenue and product-line gross margin disclosures are limited, so the differences in commercial maturity among the C/N/G lines cannot be fully quantified. Third, supply-chain and foundry details are limited, and in a changing geopolitical environment, their impact on future earnings elasticity may be underestimated. Fourth, the listing history is extremely short, so share-price history reflects free float and sentiment more than a true "historical valuation center." These blind spots do not overturn the conclusion, but they mean the confidence interval around the conclusion should be wider.

Reference sources mainly include the company's listing announcement, 2025 annual-report summary, 2026 Q1 report, Shanghai Stock Exchange disclosure files, the company's official website, BIS/eCFR/Federal Register, IDC, Reuters, and official NVIDIA and AMD disclosures. All key facts in the text have been marked with sources in the corresponding paragraphs.

Zen Horizon Cross-Section Summary

MetaX's journey has truly proved two things. First, it has proved that China can indeed take the hardest road of high-performance general-purpose GPUs, relying on a team with complete engineering experience to build products, sell boards, and run clusters. Second, it has proved that domestic compute substitution is not only a policy slogan; at least in 2025, it had begun turning into real revenue and real shipments. Without these two "real things," it could not have reached 1.644 billion yuan of revenue in 2025, nor could it have narrowed losses meaningfully. But to date, the company has not proved the third and most critical thing: whether it can steadily turn revenue into cash, and then into profit, without continuing to consume large amounts of working capital.

Longitudinally, MetaX's success has both an era dividend and team capability. The era dividend is that after 2020, the window for China's high-end compute self-reliance was forcibly opened; the team capability is that it did not stop at "concept financing," but built products, shipments, and cluster deployments. Both remain present today, but neither has reached the point where valuation can ignore price. The real issue is that the market has priced "possible success" as "high-probability success." The current 694 yuan share price looks more like a reward for the possibility that it may become an important domestic high-end GPU player over many future years, rather than a reward for the operating results it has already delivered over the past year.

Horizontally, MetaX's real advantages versus peers are its relatively complete general-purpose GPU route, engineering team, and compliance difference of not yet being on the BIS Entity List as of the base date. Its real weaknesses are that the software stack has not yet formed independent ecosystem pricing power, revenue still depends heavily on C-series hardware, and it remains some distance from being a platform company. Some of these weaknesses are temporary, such as customer count and deployment scale, which can grow over time; others are more structural, such as software ecosystem and system capability in GPUs, which require long accumulation and cannot be filled simply by doubling revenue in one year. The market's easiest mistakes today are reading "not on the Entity List" as "long-term supply-chain safety" and reading "high revenue growth" as "the business model has already worked."

The three most important variables over the next year are hard indicators, not another new-product headline: whether revenue growth can stay above 50%, whether operating cash flow can narrow meaningfully, and whether free-float expansion from lock-up expiries compresses valuation. Over the next three years, the key variable is whether MetaX can expand "more than 10 intelligent computing cluster deployments" into a repeatable large-customer system, and gradually make the software stack and cluster capability a deal threshold rather than a pure supporting tool. Over the next five years, the key variable is even simpler: whether it can live from challenger into platform player, at least forming deployment and migration costs in certain industries and regions that are hard for others to replace. If these variables are realized, today's high valuation will eventually be partly absorbed by growth; if not, share-price reversion will not be absent simply because the company's technology is decent.

Bull and Bear Arguments

Bull arguments:

• Revenue increased from 53 million yuan in 2023 to 1.644 billion yuan in 2025; commercial scaling has already happened, and this is no longer a pure R&D story.

• C-series training-inference integrated GPUs have formed real shipments, with 2025 sales of 33,600 boards, up 147.31% year on year, and deployments in more than 10 intelligent computing clusters.

• Cash, trading financial assets, and debt investments total about 9.283 billion yuan, so there is no near-term liquidity crisis forcing financing.

• As of the base date, the company was not seen in the official BIS Entity List text, preserving greater supply-chain flexibility than domestic peers already listed.

• China's accelerated server and AI infrastructure markets remain in high growth, domestic chip share is rising, and the track is not shrinking.

Bear arguments:

• Operating cash flow was negative in both 2025 and Q1 2026, so growth has not yet turned into self-funding.

• Revenue is highly concentrated in the C series, the software stack does not charge separately, and the business model remains single-engine biased.

• The current share price corresponds to a static price-to-sales ratio of about 169x, significantly higher than Hygon, NVIDIA, and AMD, and higher than most mature comparables.

• The lock-up window begins this week, followed by 9-month and 12-month tranches, so the scarce-share logic will naturally dilute over time.

• Competition at home and abroad is strengthening. Huawei Ascend, Cambricon, Hygon, and policy changes for overseas compliant chips can all compress the company's valuation tolerance.

Pre-mortem

If this investment loses 50% three years from now, I think the most likely first scenario is: from the second half of 2026 to 2027, domestic intelligent computing center construction shifts from "build first" to "utilization first," the company's revenue growth falls from 75%+ to 30%-40%, while operating cash flow remains for several consecutive quarters in the -400 million to -600 million yuan range; at the same time, the 6-month, 9-month, and 12-month lock-up expiries push the minimum free-float ratio close to 10%, the market no longer accepts a 100x price-to-sales ratio, and compresses valuation to 40-50x forward revenue. If 2027 revenue can only reach 3.5-4.0 billion yuan, the corresponding market cap would fall into the 140-200 billion yuan range, and a share price back to 350-500 yuan would not be extreme. In this scenario, the company is not a bad company; the loss comes from paying too much.

The second scenario is harsher: around 2027, overseas compliant high-performance GPUs expand supply again, or domestic system-level players use complete-machine, software, and compute-platform packaged solutions to push prices down. To defend orders, MetaX is forced to sacrifice price and gross margin, gross margin retreats from 56.51% in 2025 to around 40%, working capital continues to be tied up, and the profitability inflection is pushed back as a whole. At that point, the market's label for it would switch from "scarce core domestic GPU asset" to "hard-tech company whose profit model still needs validation," with valuation multiples and sentiment premium falling together. A 50% share-price decline would likewise be a reasonable outcome.

Final Research Conclusion

MetaX is a company worth serious research, but at the 2026-06-11 price, it is not a stock worth owning lightly. The company has crossed many thresholds that many domestic GPU startups cannot cross: a real team, real products, real shipments, real customers, and real revenue. It should not be treated as an empty concept stock. But just as important, it is still far from the stage where both the profit model and cash-flow model run smoothly. The 2025 annual report and 2026 Q1 report tell investors that operating improvement is real, but operating maturity is not.

I think the current share price looks more like "scarce-share pricing" than "cash-flow pricing." If you study the company, MetaX clearly deserves tracking; if you study the stock, the first issues today are price over-discounting and free-float expansion. What would truly change my view are three hard changes, not another market buzzword: operating cash flow continuously narrowing or even turning positive, revenue structure beginning to rely on more than the C series, and valuation falling to a level that leaves future uncertainty to investors rather than entirely to the company itself.

【Company Profile Scores】

• Fundamental quality: Medium

• Growth: High

• Moat: Medium

• Financial soundness: Strong

• Management credibility: Medium

• Valuation attractiveness: Low

• Risk level: High

• Suitable investor type: High-risk speculation

【Investment Rating】

• Rating: Avoid

• One-line investment thesis: The company is improving, but the current share price has already pulled forward years of successful execution expectations.

• Holdable price: 280-380 CNY

• Clearly overvalued price: 500-650 CNY

• Current price classification: Clearly overvalued

• Worth waiting for a better price: Yes; I would wait for the share price to enter below 220 yuan and for at least one quarter of meaningfully improved operating cash flow before reassessing opportunity cost.

• Target holding period: 3-5 years

• Expected annualized return: Bearish -75% to -68%; base -60% to -45%; bullish -28% to -6%

• Maximum loss risk: If growth slows and valuation returns toward 40-50x forward price-to-sales, a drawdown of more than 50% within 3 years is not an extreme scenario.

• Signals that trigger reassessment: operating cash flow better than -200 million yuan for two consecutive quarters; single-quarter revenue YoY growth back above 80%; contract liabilities rebound while receivables/prepayments no longer expand in sync; product lines beyond the C series start forming visible revenue; major changes occur in the BIS or Chinese policy environment.

【Ideal/Fair Buy Price】170-220 CNY Basis: This range corresponds to my margin-of-safety entry point under the company's 12-month conservative scenario valuation, already discounting unverified cash flow, high product concentration, and lock-up expansion risk.

【Valuation Range】

• current: 694.00 (as of the 2026-06-11 close)

• bear (conservative, ideal buy zone): [170, 220]

• base (reasonable, acceptable hold zone): [280, 380]

• bull (optimistic, above the clearly overvalued line): [500, 650]

Other Securities Mentioned in the Report

• NVDA.US - Global AI GPU profit pool and valuation anchor.

• AMD.US - A CPU+GPU combination challenger, also an external reference for China-market policy variables.

• 688256.SHG - A high-valuation A-share AI chip sample, useful for comparing pricing methods for domestic scarcity assets.

• 688041.SHG - A domestic compute-foundation sample that is already profitable and has more mature cash flow.

• Huawei Ascend - Unlisted, with no corresponding ticker; using listed partners and the A-share compute chain as substitute observations is not appropriate. The body only discusses Huawei as an unlisted competitive system.

• 002261.SHE - Talkweb Information is often observed by the market as an Ascend ecosystem proxy, but this report does not treat it as a core comparable and only notes that it reflects sentiment in the domestic compute chain.

• 000977.SHE - Inspur Information is a downstream proxy for China's AI server and accelerated-server demand, and can help observe intelligent computing infrastructure conditions.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.