Moore Threads In-Depth Research

Company Longitudinal Development History

The timing of Moore Threads’ emergence itself carries the imprint of the era. The company was founded on 2020-06-11. Founder Zhang Jianzhong previously worked at China HP and Dell China, and served from 2006 to 2020 as global vice president of NVIDIA and general manager for Greater China. That background meant the company directly targeted general-purpose GPU, one of the hardest tracks in China’s semiconductor industry and also one of the easiest to attach a scarcity valuation premium to, instead of following a path of “first making small edge chips, then slowly testing GPU.” The listing announcement states that the company’s core business is the R&D, design, and sales of GPUs and related products. The annual-report summary lays out the product matrix more completely: from AI compute GPUs, professional graphics GPUs, and desktop graphics GPUs to the software stack based on the MUSA architecture and KUAE AI compute clusters, the company is trying to build a complete full-function GPU platform, rather than a single accelerator card.

Its early stage was plain rather than glamorous. In 2022 and 2023, the company still relied mainly on desktop graphics acceleration and professional graphics acceleration products, and gross margin was once very poor. The inquiry response disclosed that the company’s overall gross margin was -70.45% in 2022 and recovered to 27.84% in 2023, with the biggest drag coming precisely from mid- and low-end desktop graphics acceleration products. The company explained in its response that products in this stage faced strong pricing pressure from international brands, and that certain desktop products had the strategic significance of “public sales and proving technical coverage capability,” but did not represent good unit economics. In other words, Moore Threads initially used graphics cards to plant the technical flag and prove “domestic full-function GPU,” not to make money from graphics cards.

The real first turning point was the Entity List event in October 2023, not the listing. The U.S. Federal Register shows that BIS added Moore Thread Intelligent Technology (Beijing) Co., Ltd. to the Entity List from 2023-10-17; the electronic Code of Federal Regulations page further states that transactions with listed entities are subject to a “presumption of denial.” This means items subject to the EAR face strict licensing constraints for export, re-export, and in-country transfer to Moore Threads. In commercial narrative terms, this was both a blow and a pricing lever: a blow because uncertainty around the supply chain, toolchain, and future iteration was permanently raised; a pricing lever because the market therefore more clearly labeled the company as a “scarce independently controllable GPU asset.”

The second turning point occurred in 2024, and it showed up in the revenue structure rather than in slogans. The Shanghai Stock Exchange inquiry response shows that during the reporting period, the company released one generation of GPU architecture each year, from “Sudi,” “Chunxiao,” “Quyuan,” to “Pinghu” in 2021 to 2024; more importantly, AI compute products formed scaled revenue for the first time in 2024, accounting for 77.63% of revenue, while cluster products alone accounted for 42.42% of total revenue. The company also made its representative products clear in the annual-report summary: on the AI compute side, these include MTT S5000, S4000, and the KUAE AI compute cluster; on the graphics side, S3000, S80, and X300; and on the edge and Agent side, products such as E300. These products do not prove that Moore Threads already has an NVIDIA-like software moat, but they at least prove that the company has moved from “making chips” to “making large systems that can be sold.”

The third turning point was the listing. On 2025-12-05, Moore Threads listed on the STAR Market of the Shanghai Stock Exchange, with an offer price of CNY 114.28, issuing 70 million shares and reaching post-issuance total share capital of 470,028,217 shares; only 29,382,386 shares were tradable on the first day, accounting for 6.25% of total share capital. This set of figures explains why it became a focus for capital: newly listed, pure GPU, low free float, and AI theme. When these four factors stack together, the stock is almost naturally suited to be traded as scarce inventory by the market. Reuters reported that the company rose 425.46% on its first trading day and closed at CNY 600.50. On the listing day, the market first gave it an approximate “sentiment version of comparable valuation,” rather than fundamental pricing in the strict sense.

After the listing, the financials finally began to take over the narrative. The 2025 annual report was the company’s first full annual report after listing, with revenue of CNY 1.506 billion and a net loss attributable to shareholders of CNY 1.001 billion, narrower than the 2024 loss. In Q1 2026, the company generated revenue of CNY 738 million and turned attributable profit positive for the first time in a single quarter, but recurring profit remained negative and operating cash flow showed a very large net outflow. In March 2026, the company also disclosed a CNY 660 million major contract for KUAE AI compute clusters, with payment terms of 30% advance payment within two days, 60% after acceptance, and 10% before year-end. This is a very important announcement because it gave the market its first concrete evidence of a large AI compute cluster order, while also exposing the essence of the model: large order size, long delivery cycle, and revenue recognition not synchronized with cash collection.

Near the research base date, another node that is easy to overlook but important for a newly listed stock appeared. The company’s 6-month lock-up expired on 2026-06-05. The listing announcement shows that shares subject to the 6-month offline placement lock-up totaled only 843,104 shares, accounting for 0.1794% of total share capital. The company’s 2026-05-29 announcement on the listing and circulation of restricted offline placement shares also confirmed that this portion would become tradable from 2026-06-05. The amount itself is not large. The bigger issue lies in the more distant 12-month tranche: within the 14 million strategic placement shares alone, at least 12.6 million shares subject to a 12-month lock-up will enter circulation around 2026-12-05, and many external shareholders listed in the listing announcement also carry lock-up arrangements of “12 months from the listing date.” For a stock whose starting free float was only 29.38 million shares, this is not a formal issue, but a question of whether the future liquidity structure will be rewritten.

Compressing this history into three sentences makes today’s stock easier to understand. First, Moore Threads has proved that it can make full-function GPU and cluster products. Second, it has not yet proved that it can steadily turn this business into high-quality profit and cash. Third, the capital market has already priced it according to a script in which “the probability of success is not low.” Because these three sentences are true at the same time, the company has both industrial appeal and research difficulty.

Business Model, Industry, and Horizontal Competition

The core of Moore Threads’ current commercial machine is selling chips, boards, integrated machines, clusters, and software stack together, rather than selling chips one by one. The inquiry response disclosed that AI compute products are now the revenue axis and had already approached almost all main-business revenue in the first half of 2025; within that, cluster products accounted for nearly 80% of AI compute revenue. The annual-report summary further shows that the company’s products cover scenarios including large-model training and inference, digital twins, cloud PCs, cloud gaming, Agent, and edge intelligence. This means its gross margin can be higher than many people intuitively expect because what it provides is increasingly system-level solutions rather than single cards. For the same reason, the business naturally depends more on delivery, project cadence, and customer budget cycles than “bare chip licensing.”

Its cost structure therefore has a dual character. On one hand, the company is a typical fabless player, and the real hard investment is R&D, with fixed-asset spending not the heaviest burden. The listing announcement disclosed that cumulative R&D investment from 2022 to 2024 reached CNY 3.810 billion, accounting for 626.03% of cumulative revenue over those three years. In 2025, the company’s R&D expense ratio remained as high as 86.68%, and in Q1 2026 it was still about 50%. On the other hand, as revenue migrates from boards to clusters, working capital has become heavier instead. Although the company maintains high gross margin, 2025 operating cash flow did not turn positive in sync, and in Q1 2026, while book profit turned positive, operating cash flow again posted a large net outflow. For a fabless designer, this indicates that what truly consumes cash is the time gap between stocking, delivery, acceptance, and collection, rather than factory depreciation in the traditional sense.

The moat needs to be separated into what is “real” and what only “looks real.” Moore Threads’ real barriers first include technical threshold. The annual-report summary is explicit: the company uses its self-developed MUSA architecture to support AI compute acceleration, graphics rendering, physical simulation, scientific computing, and video codec, while desktop graphics card products also support Windows and DirectX 11/12, which is the difficulty of full-function GPU. Second is product completeness. The company simultaneously works on desktop, professional graphics, training, inference, and edge SoC, rather than only training cards, which can improve customers’ judgment of its “platform capability” instead of “single-point capability.” Third is the policy window and founder-team resources. Zhang Jianzhong’s industry background helped the company enter the view of telecom operators, cloud service providers, and research institutes earlier, while domestic compute policy has given this type of supplier a window to enter project lists.

But the moat most easily misread in market promotion is: “Since it is a domestic full-function GPU, it will naturally replicate the stickiness of the CUDA system.” That has not been established by far. The review inquiry response clearly places NVIDIA and AMD as major international competitors, and cites industry materials noting that after each architecture generation, NVIDIA usually spends one year on ecosystem optimization, driver stability validation, and large-scale pilots. The same response also mentioned that NVIDIA announced on 2025-07-15 that it would resume H20 sales in China and launch a new China-compatible GPU. For Moore Threads, the real shortfall is how far “software ecosystem, developer habits, scale validation, and stable delivery” remain from international standards, rather than whether it has a GPU. In addition, sponsor documents disclosed that the top five customers accounted for more than 80% of revenue in each reporting period, showing high customer concentration and indicating that the company has not yet entered a stage where customers are naturally unwilling to migrate.

On the industry side, the track is strong, but a strong track and the winner are not the same concept. Research from the China Academy of Information and Communications Technology and industry alliances shows that as of June 2025, the total scale of computing-center racks in use nationwide had reached 10.85 million standard racks, national computing hub nodes accounted for 72.6%, and intelligent compute scale reached 788 EFLOPS. The same institute’s 2025 research also noted that the National Data Administration’s 2025 action plan emphasized efficient matching between compute demand and national hub nodes, while the State Council’s August 2025 “AI+” opinion emphasized ultra-large-scale intelligent compute clusters, a nationally integrated compute network, and interconnection of intelligent compute power. In other words, the broad demand direction is not the problem, and policy still clearly leans toward intelligent compute infrastructure.

However, the policy stance has shifted from “build everywhere” to “coordinate construction.” Research materials from 2025-2026 likewise show that policy emphasis increasingly centers on hub-node layout, interconnection, and supply-demand matching rather than disorderly expansion. Reuters’ 2026-06-09 report even mentioned that China was studying a roughly CNY 2 trillion data-center construction blueprint over the next five years and leaning toward a higher proportion of domestic suppliers; the report also noted that the government had previously required new data-center projects using state funds to use only domestic AI chips. For Moore Threads, this is a double-edged sword: one end is a real market entry point brought by domestic substitution, while the other end is that when state funds and telecom-operator budgets become the core source of demand, competition will look more like “who can enter the list, who can deliver stably, and whose ecosystem is easiest to use,” rather than a pure technical-parameter duel.

Placed into the horizontal landscape, Moore Threads’ ecological position looks more like a niche player transitioning from challenger to project-based platform supplier. The best A-share comparisons are Cambricon and MetaX as two different paths, rather than Hygon or NVIDIA: Cambricon has already reached a larger revenue and better profit position, proving that once a domestic AI chip company captures telecom operators, internet companies, and the large-model wave, valuation can become extremely expensive; MetaX shows that A-shares are still willing to give a high premium to scarce pure-play GPU names, even if profitability has not fully stabilized. Hygon’s relevance is to remind investors that a truly mature domestic compute-chip company will have visibly better cash flow and earnings quality than Moore Threads. NVIDIA and AMD remind investors that the strongest weapons of international leaders remain complete software ecosystems and scaled delivery capability, not single-card performance.

Note: MetaX revenue uses annualized 2026Q1 revenue and is only a rough valuation comparison, not equivalent to audited annual data. Market capitalization and share prices in the table are intraday data available on the research base date; Moore Threads’ market capitalization is calculated using post-listing total share capital. Sources and calculation basis are explained at the end of the report.

This table explains many things. First, Moore Threads’ current valuation is much more expensive than Cambricon’s, even though Cambricon had already turned profitable for the full year in 2025 and reached Q1 2026 revenue of CNY 2.885 billion. Second, Hygon Information’s revenue and profit scale are both far above Moore Threads, but its price/sales ratio is only a fraction of Moore Threads’. Third, MetaX shows that “high valuation is not Moore Threads’ problem alone”; A-shares have given very high scarcity premiums to domestic GPU newcomers as a group. In other words, Moore Threads’ position is “within the hottest section of the industry, it still stands in the most expensive layer,” rather than “the industry is cheap, but it is not.”

The international reference is equally important. NVIDIA’s latest price/sales ratio is about 22.6x, and AMD’s about 21.6x. Their latest full-year revenue reached USD 215.9 billion and USD 34.6 billion, respectively. The multiples the market gives them are not low, but they are built on proven global ecosystems, scaled delivery, customer bases, and profitability. Moore Threads’ current valuation does not look like a “discount comparison” against international leaders. Instead, it looks like Chinese scarcity and newly listed stock liquidity covering up the huge quality gap that should exist between mature companies and early-stage companies.

Current Fundamentals, Valuation, and Risks

First look at the past five quarters. The numbers are quite telling. The annual-report summary shows Moore Threads’ revenue in the four quarters of 2025 was approximately CNY 289 million, CNY 413 million, CNY 83 million, and CNY 721 million, respectively, with large fluctuations but an upward trend. By Q1 2026, revenue further reached CNY 738 million. This curve looks more driven by project delivery cadence than continuous scaling of standardized products. What truly supports the market in continuing to give the story credibility is that revenue entered a new level after Q4 2025; what keeps the market from chasing even more aggressively is that seasonality and order cadence remain heavy.

The earnings-quality issue is more important than revenue. In Q1 2026, the company recorded net profit attributable to shareholders of CNY 29.3592 million, seemingly taking an important step. But non-recurring gains in the same quarter reached CNY 83.64 million, including CNY 70.06 million of government subsidies recognized in current profit and loss, and recurring attributable profit still showed a loss of CNY 54.2822 million. Looking further down, net operating cash flow was CNY -1.487 billion, while R&D expenses remained CNY 369 million, about half of single-quarter revenue. For a high-valuation stock, this combination tells investors: the income statement is beginning to tell a story, but the cash-flow statement has not nodded.

The favorable fundamental changes should not be ignored either. The CNY 660 million KUAE AI compute cluster contract disclosed in March 2026 was equal to about 43.8% of the company’s full-year 2025 revenue. The contract terms require the customer to pay a 30% advance within two days of effectiveness, showing that the company has at least entered the stage of being able to win large project orders. If similar-size contracts appear again in the next few quarters and are smoothly accepted and recognized as revenue, Moore Threads’ “project-based revenue” will have a chance to move one step toward “order-pool revenue.” But as of the research base date, this is still a strong signal, not proven continuity.

Putting the key data into one table is more intuitive.

Note: † 2024 gross margin comes from the IPO inquiry response basis; 2025 and 2026Q1 come from periodic reports. ‡ 2025 operating cash flow is summed from quarterly data disclosed in the annual-report summary.

For valuation, Moore Threads cannot tell its story with PE, at least not yet. Once non-recurring gains are stripped out, profitability is still unstable. Once cash flow is examined, owner earnings remain negative. If 2025 attributable net profit is used, the ratio of operating cash flow to net profit is about 2.95, but both are negative, meaning operating cash outflow was nearly three times the accounting loss. Q1 2026 was even more extreme: attributable profit was slightly positive in the quarter, while operating cash flow continued to show a large outflow. Therefore, the most suitable primary valuation method for this company today is forward price/sales, not PE, DCF, or even the apparent price/cash-flow ratio. For Moore Threads, discussing whether revenue can continue and whether gross margin can hold must come before discussing what multiple profit deserves.

The historical valuation change is therefore very dramatic. Based on 2025 revenue, Moore Threads’ price/sales ratio at the offer price was about 35.7x; at the first-day closing price, the ratio was close to 187.5x; intraday on the research base date, the dynamic price/sales ratio was about 192.6x. In other words, after the listing, the fundamentals did indeed improve, but valuation expanded much faster than fundamentals improved. The current price is “requiring fundamentals to continue delivering at high speed,” rather than “waiting for fundamentals to start.”

Among peers, it stands higher. On the current basis, Moore Threads trades at about 192.6x sales, Cambricon about 122.8x, Hygon Information about 43.4x, while international leaders NVIDIA and AMD are both around 22x sales. Different development stages mean this cannot be compared mechanically. But if a company with recurring losses and deeply negative cash flow is valued above domestic comparables that have already achieved scaled profitability, investors must understand that they are buying an “expensive validation period,” not “cheap growth.”

The following valuation scenarios only unpack implied expectations and map price to assumptions. They do not constitute investment advice.

The above share-price ranges are derived from revenue assumptions, P/S assumptions, and current total share capital: the conservative scenario is about CNY 166–194, the base scenario about CNY 316–386, and the optimistic scenario about CNY 702–851. What it conveys is that “if you buy it today at more than CNY 600, you are betting on the optimistic scenario, not the base scenario,” not that “Moore Threads has no value.”

The margin-of-safety review is direct. The current price carries a clear premium to implied value under the conservative scenario, so the margin of safety is zero. In the base scenario, the most fragile assumption is whether “high-level revenue can continue,” not gross margin. If the base revenue assumption of CNY 3.3 billion is discounted by 30%, then at 50x forward sales the corresponding share price is about CNY 246. If over the next three years the company does not truly improve earnings quality and only remains in a state of high valuation and low cash conversion, annualized return will most likely be significantly negative and cannot serve as a substitute for long-term government bond yields. The only conclusion here is: there is no margin of safety.

There are roughly five risks that could cause permanent capital loss. The first is commercialization validation risk, with medium-high probability and high impact. The evidence is that the top five customers accounted for more than 80% of revenue during the reporting period, and the single contract in March 2026 alone equaled 43.8% of full-year 2025 revenue. If the budget, acceptance, or collection process of several core customers runs into problems, revenue, cash flow, and sentiment will be pressured at the same time. The observable signal is whether new large orders continue to appear over two consecutive quarters and whether large orders are recognized as revenue on schedule.

The second is ecosystem and price-war risk, with medium probability and high impact. Internationally, the review inquiry response mentioned that NVIDIA resumed H20 sales in China in mid-2025 and launched a new China-compatible GPU. Domestically, disclosed companies such as Cambricon, Hygon, and MetaX are all expanding revenue and product coverage. For Moore Threads, the real danger is whether cluster solutions will be forced to cut prices to win orders once competitors all enter telecom, financial, and internet project pools, rather than whether there is competition at all. Observable signals are whether gross margin falls below 60% for consecutive quarters and whether the advance-payment ratio in major contracts declines.

The third is cash-flow and working-capital risk, with high probability and high impact. The company had a large net operating cash outflow in 2025, and cash flow remained deeply negative even when book profit turned positive in 2026Q1, showing that order scaling has not automatically turned into cash creation. For a high-valuation newly listed stock, this is the most dangerous type of risk, because it does not necessarily hurt revenue first; it often hurts valuation first. Observable signals are whether net operating cash flow, contract advance-payment ratio, and receivables and inventory turnover improve together.

The fourth is share-supply structure risk, with medium-high probability and medium-high impact. The 6-month tranche unlock involved only 843,100 shares, with limited price impact. But the 12-month strategic placement portion alone involves at least 12.6 million shares, plus a large number of original-shareholder lock-ups of “12 months from the listing date,” meaning the real supply change around 2026-12-05 will be far larger than the small June unlock. For a stock whose initial free float was only 29.38 million shares, the change in share supply itself is enough to alter the valuation center. Observable signals are the scale of subsequent official unlock announcements and turnover and institutional absorption around the unlock.

The fifth is export-control and policy-dependence risk, with medium probability and high impact. The BIS Entity List means the company is always under higher supply-chain uncertainty. At the same time, the company currently benefits from domestic compute substitution and government subsidies, with government subsidies alone reaching CNY 70.06 million in Q1 non-recurring gains. If policy support slows at the margin, or if supply-chain restrictions delay product iteration, the market will reduce both revenue expectations and valuation multiples. Observable signals are the share of government subsidies in profit, the release schedule of next-generation products, and further changes in U.S. export-control policy.

If only the most useful tracking indicators are selected, the following set is enough.

The logic behind these indicators is very direct. Revenue tests whether the high level can continue. The AI compute share tests whether the business is sliding back toward low-margin graphics cards. Recurring net profit and operating cash flow must be viewed together to distinguish a real inflection point from an accounting inflection point. Gross margin monitors price war. Major-contract strength determines whether the project pool is expanding. The valuation premium over Cambricon is the simplest way to observe whether market sentiment has moved into an overheated zone. The main data sources are the company’s quarterly reports, major-contract announcements, comparable companies, and market quotes.

Zen Horizon Cross-Section Summary

Longitudinally, what Moore Threads has truly proved is three more basic and scarce capabilities, rather than already owning an irreplaceable profit machine: first, it has indeed turned full-function GPU, one of China’s hardest chip businesses, into a product matrix; second, it can shift from desktop and professional graphics to AI compute at very high speed, moving the business focus into a larger demand pool; third, at the intersection of policy, industry, and capital markets, it has successfully turned itself into one of the most visible cards among domestic GPU assets. None of these is lightweight. The real problem is that success at the first level does not automatically become reasonable pricing at the third level. Between them lies the second level: business quality.

Horizontally, Moore Threads’ real advantage is the combination of “full-function GPU + domestic supply + system-level delivery,” not any single parameter. The gap it has the best chance to fill is project-based demand from customers that want to reduce reliance on overseas high-end GPUs while not yet finding sufficiently mature domestic substitutes, especially telecom operators, government and enterprise customers, and some industry intelligent compute centers. Its weaknesses are also clear: revenue scale is still small, customer concentration is high, ecosystem stickiness is far from international leaders, and cash-flow quality is currently visibly weaker than mature domestic peers. Cambricon and Hygon have proved that the market is willing to pay high multiples for domestic compute-chip leaders. But they also prove that to deserve those multiples, a company must go further in earnings quality and customer base.

The market’s most likely misjudgment is “how long Moore Threads will need to deserve today’s price,” rather than “whether Moore Threads can make it.” If the question is asked more simply: could this company become an important domestic GPU platform over the next 3–5 years? The answer is yes, it could. But if the question is whether investors at the price above CNY 600 in June 2026 still receive sufficiently thick odds, my answer is no. The current price does not leave investors a buffer of “even if it is slower, they can still earn money.” It looks more like a high-leverage ticket: if the company wins large orders consecutively in 2026-2027, turns recurring profit positive, and sees cash flow catch up, the stock still has room for upward revision; if any link slows, valuation will compress first, and the share price will react earlier than the income statement.

The most important variable over the next year is whether revenue and cash flow can both hold for two consecutive quarters. The most important variable over the next three years is whether customer repurchase and ecosystem validation can turn it from a “project-based supplier” into a “platform supplier.” The most important variable over the next five years is product iteration and supply-chain sustainability under the Entity List backdrop. For investors, this means the research judgment will be strengthened under certain conditions and overturned under others: if the company achieves consecutive recurring profitability in H2 2026, operating cash flow narrows meaningfully, and it proves that institutions can absorb shares around the 12-month unlock, Moore Threads will become a better investment target; if revenue keeps growing while cash flow and recurring profit fail to improve, or if new orders depend heavily on a single customer and policy projects, the current valuation story should be re-examined.

The bull case can be compressed into four points.

• Domestic full-function GPU platforms are extremely scarce in A-shares, and the company has formed a complete product matrix from chips to clusters.

• The revenue structure has completed its shift from graphics acceleration to AI compute, and the commercialization direction has been much clearer since 2024 than in the early period.

• Q1 2026 revenue continued to grow rapidly, and the CNY 660 million major contract proves that large AI compute cluster orders are not only a paper story.

• Domestic compute and intelligent compute center construction still receive policy support, and the domestic substitution window has not closed.

The bear case must be equally specific.

• Nominal profitability in 2026Q1 came from large non-recurring gains, while recurring profit remained negative, so the earnings inflection point has not yet been proved.

• Operating cash flow and the income statement have diverged for a long time, with both full-year 2025 and 2026Q1 showing clear cash-conversion problems.

• Current valuation is significantly above Hygon and also above Cambricon, meaning the market has already bought into a fairly optimistic forward scenario.

• Customer concentration is high and project cadence is heavy, so any change in one large customer or large order will transmit directly into revenue and valuation.

• The 12-month unlock near the end of 2026 will be much more important than the small June unlock, and the share-supply structure may be repriced.

In a pre-mortem, two scripts most easily cause this investment to lose 50% three years from now. The first is competitive deceleration: by 2027, Cambricon, MetaX, and more mature platform solutions win more repurchases among telecom operators and internet customers, Moore Threads cuts KUAE cluster prices by 30%–35% to hold projects, gross margin falls from about 65% to 45%–50%, full-year revenue stays around CNY 3.0 billion, the market compresses its forward price/sales ratio from 60x to 30x, and the corresponding share-price range falls to around CNY 200. The second is delivery deceleration: by 2027, next-generation architecture delivery and the supply chain face two rounds of delays under Entity List constraints, the company still cannot achieve stable recurring profitability for several consecutive quarters, operating cash flow continues to show large net outflows, and the market is no longer willing to pay for “scarcity,” pricing it only by the cash-creation ability of a project-based enterprise. The stock could also return to the CNY 170–220 area.

The final research conclusion is direct. Moore Threads is a company worth tracking continuously because it has already crossed the two thresholds of “having products” and “having revenue.” But it is not a stock worth chasing at any price because it has not yet crossed the two harder thresholds of “having high-quality profit” and “being able to turn profit into cash.” The current price is not cheap. It can even be said that it has already prepaid quite a lot of the smoothest road over the next two to three years. What truly worries me is not that the company cannot make next-generation products, but that the market has pushed valuation to a place with very little tolerance for error while fundamentals are still in the validation period.

If three things appear in the next two quarters, I would clearly raise my assessment: continued new large cluster orders, consecutive recurring profitability, and a significant narrowing of operating cash flow. Conversely, if the company’s revenue keeps growing rapidly but cash flow remains deeply negative and profit continues to rely on subsidies and fair-value changes, I would treat it as a typical case of “good story but bad price,” not a growth investment opportunity. On the research base-date basis, it is more appropriate for me to rate it “Watch,” rather than “Hold” or “Cautious Buy.” The reason is that today’s price already requires it to realize its future quickly enough, not that the company has no future.

【Company Profile Scores】

• Fundamental quality: Medium

• Growth: High

• Moat: Medium

• Financial resilience: Medium

• Management credibility: Medium

• Valuation attractiveness: Low

• Risk level: High

• Suitable investor type: High-risk speculation

【Investment Rating】

• Rating: Watch

• One-sentence investment thesis: Orders and revenue are scaling, but cash flow and recurring profit have not yet validated the valuation.

• Three-tier price signals: Ideal buy price: CNY 166–194

• Acceptable holding price: CNY 316–386

• Clearly overvalued price: CNY 702–851

• Current price classification: Outside the three tiers

• Whether it is worth waiting for a better price: Yes. The core trigger for buying is that the price at least returns to around CNY 200 and the company simultaneously shows consecutive recurring profitability or clear improvement in operating cash flow, rather than simply falling to a round-number level. The opportunity cost of waiting is that if the company rapidly delivers the optimistic scenario in H2 2026, the share price may continue to detach from the fundamental center.

• Target holding period: If a purchase occurs, 1–3 years is more suitable than short-term trading.

• Expected annualized return: Conservative about -69% to -73%; base about -37% to -49%; optimistic about +14% to +38%.

• Maximum loss risk: If competition leads to a price war, or Entity List constraints delay next-generation products, a share price in the CNY 170–220 area would not be extreme, implying a potential drawdown of about 64%–72% from the current price.

• Signals triggering reassessment: Recurring net profit attributable to shareholders turns positive for two consecutive quarters;

• Net operating cash flow clearly narrows to within CNY -200 million for two consecutive quarters;

• Another AI compute cluster contract larger than CNY 500 million is signed during the year;

• Gross margin falls below 55% for two consecutive quarters;

• The officially disclosed 12-month unlock scale is significantly higher than expectations based on the current free float.

【Ideal/Fair Buy Price】166–194 CNY Basis: This corresponds to the conservative scenario valuation range of 2026E P/S 30–35x and 2026 revenue of CNY 2.6 billion, and only then does a discussable margin of safety begin to exist.

【Valuation Range】

• current: 616.87 (intraday as of 2026-06-11)

• bear (conservative · ideal buy range): [166, 194]

• base (reasonable · acceptable holding range): [316, 386]

• bull (optimistic · above the clearly overvalued line): [702, 851]

References

Research uncertainty mainly has four sources.

• First, the customer name in the major contract was exempted from disclosure, so external observers cannot judge its repurchase probability, credit quality, or industry attributes. Therefore, judgments on revenue continuity in 2026-2027 can only be range analysis, not a precise order model.

• Second, for the 12-month tranche and longer restricted-share lock-ups, some share lock-up conditions are related to the dates on which shareholders obtained shares. Near the research base date, what can be confirmed is the 6-month tranche and part of the 12-month tranche; the complete unlock schedule should still rely on subsequent official announcements.

• Third, public information can show changes in the income statement and revenue structure relatively well, but it is difficult to precisely split maintenance capex and expansion capex. Therefore, owner-earnings valuation can only reach the directional judgment that it is “still negative,” not a rigorous DCF.

• Fourth, public disclosures are insufficient on the true intensity of Entity List constraints on future product iteration, the supply chain, and foundry access. Statements in this report on its long-term impact are prudent inferences based on BIS rules and industry common sense.

Core public sources include: the company’s listing announcement, 2025 annual-report summary, 2026 first-quarter report, March 2026 major-contract announcement, May 2026 announcement on circulation of restricted offline placement shares, and Shanghai Stock Exchange review inquiry responses; annual reports and quarterly reports of comparable companies; official financial reports from NVIDIA and AMD; the U.S. Federal Register and BIS Entity List rule pages; intelligent compute and computing infrastructure research published by the China Academy of Information and Communications Technology and industry alliances; and Reuters reports on China’s AI data-center investment and domestic chip usage direction.

Other Securities Mentioned in This Report

• NVDA.US — Global GPU leader, used to compare ecosystem barriers, earnings quality, and the international valuation center.

• AMD.US — International second-tier AI accelerator and GPU reference, used to compare competitive pressure from China-compliant products.

• 688256.SHG — Cambricon is a core A-share AI chip comparable, used to compare revenue scale, earnings inflection point, and valuation.

• 688041.SHG — Hygon Information represents a more mature domestic compute-chip platform, used to compare cash flow and earnings quality.

• 688802.SHG — MetaX is a closer domestic GPU newcomer in A-shares, used to compare the “scarcity premium.”

• 688981.SHG — SMIC is used to illustrate the manufacturing-side reference and market-sentiment reference in the domestic semiconductor chain.

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