Cambricon In-Depth Research

Why did Cambricon emerge? The answer is simple: it grew out of the gap that "China needs its own AI computing architecture," rather than carving share out of a mature market. Since its establishment in March 2016, the company has successively launched terminal intelligent processors including Cambricon 1A, 1H, and 1M, then cloud products including Siyuan 100, 270, 290, and 370, and its intelligent processor IP has been integrated into more than 100 million smartphones and other smart terminal devices. This shows that the first place where it proved itself was IP and terminal-side architecture capability, not the large-model training cards that the market is most excited about today. The company is still led by Chen Tianshi as chairman and general manager, and the market broadly views Cambricon as a technology startup with a strong founder-led character.

The listing path carried a very distinct period flavor. On 2020-07-20, Cambricon listed on the STAR Market at an issue price of 64.39 yuan, closed its first day at 212.4 yuan, up about 230%, and quickly reached a market cap of 84.98 billion yuan. At that time, capital markets understood it very much as a "reserve leader for a Chinese AI chip champion": enormous technological imagination, while commercial execution did not have to be calculated urgently. Reuters also mentioned at the time that the market was enthusiastic about technological self-reliance, while Alibaba as a shareholder strengthened the company’s story in capital markets. The listing starting point was too high. After that, if real operations landed even a little slowly, the share price had to undergo a long reversion.

The table below compresses key milestones into a data timeline. It focuses on how each event changed the market’s anchor for Cambricon, rather than serving as a chronology. Relevant dates, fundraising sizes, and regulatory milestones come from company announcements, official documents, or Reuters reports.

The real turning point occurred from 2020 to 2022. At listing, the market believed in "leading architecture, fixable ecosystem, and inevitable domestic substitution." Two years later, the market was no longer willing to listen to that story alone, because the reality was 2021 revenue of 721 million yuan and net loss attributable to shareholders of 847 million yuan. Looking back through the 2023 annual report, 2022 revenue was 729 million yuan, and 2023 revenue fell again to 709 million yuan. Although losses narrowed in 2023, full-year net loss attributable to shareholders was still 848 million yuan. The most fatal conflict in this stage was that capital markets valued it as a "platform growth stock," while the financial statements looked more like a high-investment, slow-execution chip startup. Being added to the Entity List in December 2022 also formally wrote advanced-process access and parts of the supply chain into the company’s risk framework. The share price fell to 31.27 yuan on 2022-04-27, almost fully compressing the early-listing optimism that had ignored cost.

2023 to 2024 is a period easily romanticized in hindsight, but it was quite difficult at the time. After ChatGPT, any A-share name with AI computing power in it saw valuation rise first. Cambricon benefited especially because it was scarce, pure, and could tell a "domestic substitution" story. But performance did not immediately follow. 2023 revenue still fell 2.70% year over year, and operating cash flow had only just turned positive to 19.7689 million yuan. 2024 revenue recovered to 1.174 billion yuan, but net profit attributable to shareholders still lost 452 million yuan, and operating cash flow dropped again to -1.618 billion yuan. In other words, the previous valuation reset came before profit improvement; what the market was actually buying was "what might happen over the next two years."

From 2025 to early 2026, Cambricon finally translated market imagination into financials. In H1 2025, revenue was 2.881 billion yuan and net profit attributable to shareholders was 1.038 billion yuan. In the first three quarters, revenue was 4.607 billion yuan and net profit attributable to shareholders was 1.605 billion yuan. For the full year, revenue was 6.497 billion yuan and net profit attributable to shareholders was 2.059 billion yuan. Q2, Q3, and Q4 revenue was about 1.769 billion yuan, 1.727 billion yuan, and 1.890 billion yuan respectively, then rose again to 2.885 billion yuan in Q1 2026. That shows revenue ran at a high level for at least four consecutive quarters, rather than reflecting an accidental single-quarter recognition. The market therefore repriced Cambricon from "can it survive" to "can it become a core asset in China’s high-end AI accelerator market." The company also used this window to complete a 3.985 billion yuan private placement and implemented its first cash dividend and capitalization issue in 2026.

This history leaves three long-term effects today. First, Cambricon has proved that it is a company capable of turning technology into real revenue, which matters. Second, it has not yet proved that its revenue structure is healthy enough, because the breakout happened too quickly and both customers and supply chains are highly concentrated. Third, capital markets will repeatedly value it by the logic of "future unification of domestic computing power," while the company’s actual delivery remains mainly the staged ramp of the cloud product line. As long as the distance between these two lines remains, volatility will not be small.

From a longitudinal financial review, Cambricon is a curve repeatedly pulled by R&D intensity and product execution cycles, not a smooth growth curve. Even against the 2025 breakout, R&D expenses reached 1.351 billion yuan, up 11.10% year over year. R&D investment did not brake because of the profit inflection. The company’s 2025 investment in cloud intelligent chips, basic system software, and hardware platforms was 2.888 billion yuan, 2.504 billion yuan, and 630 million yuan respectively, showing that current accounting profit is still built on continued heavy R&D push, not on entering a "harvest period."

Gross margin improvement also needs to be unpacked. In 2025, main-business gross margin was 55.15%, including 55.22% for the cloud product line, 23.89% for the edge product line, and 95.26% for IP licensing and software. It looks good, but because IP and software revenue is too small, this high margin cannot support overall profit. What truly supports profit is the cloud product line maintaining a mid-to-high gross margin at high revenue scale. The issue is whether this margin can continue through customer changes, sanctions constraints, and supply-chain adjustments. That has not yet been fully verified.

Cash flow reminds investors not to treat 2025 profit too casually. In 2025, net profit attributable to shareholders was 2.059 billion yuan, but operating cash flow was -498 million yuan. In the cash-flow statement supplement, the company explicitly shows "decrease in inventories (increase stated with a negative sign)" of -3.303 billion yuan, while receivables also continued to occupy cash. In other words, profit has appeared, but the delivery chain and inventory-preparation chain still consume cash. For a chip design company under the Entity List, this is not a small issue, because it means growth quality has not completed its final proof.

The share price and valuation history can also be summarized in one sentence: first lifted high by identity, then smashed down by delayed execution, then lifted higher by the AI theme and the real profit inflection. It closed its first listing day at 212.4 yuan, bottomed at 31.27 yuan in 2022, reached new highs again under the AI theme in 2023, accelerated further with earnings execution from August 2025, and reached a high of 1,448.01 yuan in May 2026. As of the 2026-06-11 close, the share price was 1,231 yuan, only about 15% below the 52-week high. Capital markets have changed its label three times: from "the first AI chip stock," to "high-investment loss-making company," to "pure Chinese computing-power name." Business quality has indeed improved, but the valuation center has risen even faster than business quality.

Business Model, Industry, and Horizontal Competitors

Cambricon’s commercial machine became very clear after 2025. It makes money by shipping cloud products: not through software subscriptions, not through terminal IP relicensing, and not through slow volume expansion in the edge market. In 2025, cloud product revenue was 6.477 billion yuan, edge product revenue was only 3.39 million yuan, IP licensing and software was 2.29 million yuan, and other business was 14.67 million yuan. By sales channel, direct sales accounted for 98.13%; by region, domestic revenue accounted for an absolute majority. The once-familiar "cloud-edge-device integration" narrative has contracted in the income statement into one main axis: "cloud chips and accelerator cards, training machines, and related systems." Buying Cambricon today is essentially buying its cloud AI accelerator business.

The cost structure is also distinctive. In 2025 integrated circuit product costs, direct materials accounted for 50.40%, other materials 33.91%, packaging and testing 11.80%, and manufacturing expenses 3.89%. This is close to the external manufacturing structure of a typical fabless design company: gross margin depends on architecture, design, software stack, and the ability to organize external manufacturing, packaging, and testing resources, not on owning production lines. The problem is therefore exposed directly: the Entity List affects the stability and cost of these key external links, not the sales end. In its refinancing prospectus, the company states very plainly that "supply-chain stability carries certain risks." The annual report also notes that upstream raw material prices rose overall, and that the company responded through strategic stocking and product iteration. Put those two sentences together with the 178.67% surge in 2025 inventories, and it becomes clear why the market still watches the supply chain so closely after profit turned positive.

I think Cambricon has only three real moats. The first is self-developed intelligent processor architecture and instruction set capability. The annual report discloses that its next-generation intelligent processor microarchitecture and instruction set are under R&D, that existing technologies already cover cloud, edge, and terminal product lines, and that it has obtained 693 patents, including 239 overseas patents. The second is its basic system software platform. The company is not only making cards; it is also continuously iterating training software platforms and inference software platforms. The 2025 annual report even disclosed Day 0 support for DeepSeek-V3.2, showing that its software work is not only about "having something," but also about ecosystem access speed. The third is customer-level engineering capability. Except for the third-ranked customer, four of the top five customers in 2025 were new customers, meaning the company is rapidly sending products into new procurement systems, not simply enlarging old orders.

But "moats in market promotion" must be separated from "real moats." Cambricon has not formed a global standard software ecosystem like NVIDIA’s CUDA. It has not formed broader server processor coverage in CPU and DCU like Hygon. It does not have Huawei’s integrated capabilities across whole machines, networking, servers, and industry solutions. The company’s current revenue structure is too single, which itself shows that the edge and IP lines have not formed stable profit moats. Put more plainly, Cambricon has proved that it can build cloud chips that can be sold, but it has not proved that it has secured the ecosystem entrance.

In management and governance, Cambricon remains a typical founder-led, R&D-first company. The main line of capital allocation is R&D and refinancing, not dividends: in 2023 it completed the 2022 issuance to specific investors, raising 1.672 billion yuan; in 2025 it completed another 3.985 billion yuan private placement, at an issue price as high as 1,195.02 yuan per share. What deserves real credit is that after turning profitable in 2025, the company implemented a cash dividend of 1.50 yuan per share and a capitalization issue in 2026, rather than continuing to use "high R&D investment" as a shield against shareholder returns. On the audit side, the 2025 annual report received a standard unqualified opinion from Pan-China Certified Public Accountants. My view is that the governance discount is currently not large. The core controversy is whether operating quality can pass through a high-volatility cycle, not governance.

Putting the industry back into the broader context, Cambricon sits in a track where both supply and demand are shaped by policy and geopolitics. In 2023, six Chinese departments issued the "Action Plan for High-Quality Development of Computing Power Infrastructure," proposing that by 2025 computing-power scale exceed 300 EFLOPS and intelligent computing power account for 35%. That means local AI chip demand is jointly driven by infrastructure construction, government and enterprise digitalization, and large-model deployment, rather than by a natural market curve. On the other side, in 2025 the United States added new China licensing constraints on NVIDIA H20 and AMD MI308, directly changing the availability of high-end AI chips in China. Demand is lifted by policy, and supply is compressed by geopolitics. That is the real environment of Cambricon’s industry.

Horizontally, Cambricon is one of the most watched Chinese challengers in the industry by capital markets, not the industry leader. The most comparable listed company is actually Hygon Information, not NVIDIA or AMD. Hygon’s products are CPU plus DCU, covering servers and workstations. Cambricon is more concentrated in AI accelerator chips and accelerator cards. Hygon’s 2025 revenue was 14.376 billion yuan, net profit attributable to shareholders was 2.542 billion yuan, Q1 2026 revenue was 4.034 billion yuan, and net profit attributable to shareholders was 687 million yuan, still growing quickly. Its market cap was about 623.3 billion yuan, P/S about 38.71x, and P/E about 227x. By contrast, Cambricon’s current market cap is about 773.4 billion yuan, trailing P/S about 93.5x, and trailing P/E about 284.7x. Its revenue scale and customer structure are weaker, yet it receives higher pricing. The market is paying a premium for the idea that "if China is to produce a true local GPU leader, this looks most like the candidate," not for "operations already mature."

The table below puts the most important domestic listed comparables together. It intentionally does not put NVIDIA and AMD into the same table, because their business boundaries are too broad and side-by-side numbers would make people think they are simply GPU card vendors. After the Cambricon and Hygon table, looking at overseas and domestic private GPU startups in text is closer to the real competitive landscape. Market caps and multiples in the table use the closest publicly available market data around the reference date.

Bring overseas players in, and the industry picture becomes clearer. NVIDIA’s PE on 2026-06-11 was about 30.5x, and AMD’s about 148.3x. The U.S. 2025 new China licensing requirements hit H20 and MI308 respectively. For Chinese customers, this means "available overseas cards are less stable," but for Cambricon it does not automatically mean "all orders come to me." The real beneficiaries are often those that can provide whole machines, software, services, and large-customer delivery as an integrated platform. In 2025 China AI accelerator server market data cited by Reuters from IDC, Huawei led the domestic camp in shipments, while Cambricon and Hygon each had about 116,000 cards. This shows Cambricon has reached the table, but it is not the only major player at the table.

Looking toward the edge of the industry, competition will only become more crowded. Moore Threads and MetaX have successively listed on the STAR Market, and Enflame is also advancing its listing process. Once capital markets again assign high valuations to Chinese GPUs, startups will simultaneously accelerate financing, sampling, and customer grabbing. Cambricon’s advantage is that it has already moved through the profit inflection first. Its weakness is that other competitors do not need to wait for it to stabilize before attacking. For investors, this means Cambricon looks more like a front-row challenger in a high-growth track than an established dominant leader.

Current Fundamentals and Long-Short Debate

Looking only at the most recent four quarters, Cambricon’s operating curve is indeed almost unrealistically attractive. Q2 2025 revenue was about 1.769 billion yuan and net profit attributable to shareholders about 683 million yuan; Q3 revenue was 1.727 billion yuan and net profit attributable to shareholders 567 million yuan; Q4 revenue was 1.890 billion yuan and net profit attributable to shareholders 455 million yuan; Q1 2026 then jumped to revenue of 2.885 billion yuan and net profit attributable to shareholders of 1.013 billion yuan, with revenue rising about 52.6% sequentially. This string of numbers says two things: first, 2025 was not a one-off recognition; second, Q1 2026 was still accelerating, so at least in the short term orders did not visibly turn down.

The table below organizes revenue and net profit for the last four quarters based on the company’s statutory disclosure basis. Q2 and Q4 2025 are derived by subtraction from the semiannual and full-year reports, with the purpose being to see trend rather than pursue accounting-line granularity.

The market reaction after financial disclosure also shows that the market is now trading the core question of "whether growth can continue enough to digest the current valuation," rather than "whether Cambricon can grow." After the company disclosed its 2025 annual report, Reuters recorded a 2.4% early-session rise, with the market treating "first full-year profit" as a reason for another valuation step-up. At the same time, in August 2025 the company proactively clarified rumors about large substrate orders, revenue forecasts, new products, and potential customers, reminding investors not to treat informal news as evidence of continued earnings. In other words, market expectations for it are high, high enough that the company itself has to keep cooling trading sentiment.

The current share price really reflects four stacked narratives. The first layer is delivered performance growth: 2025 and Q1 2026 revenue and profit both significantly exceeded the previous few years. The second is domestic substitution: China licensing constraints on H20 and MI308 make local customers more willing to reserve budgets for domestic solutions. The third is scarcity pricing: very few A-share companies can carry the imagination of "pure AI chips." The fourth is a slightly more distant technological aspiration: the market is willing to pay in advance for a next-generation cloud chip that has not yet been formally disclosed as mass-produced. The first two layers have financial-report support; the latter three are mainly driven by expectations.

The strongest bull evidence is that Cambricon is no longer a paper domestic-substitution story. In 2025, except for the third-ranked customer, the rest of the top five were new customers, showing that the company is truly expanding its customer base rather than only digging deeper into old customers. The cloud product line still had a 55.22% gross margin in 2025, showing that revenue was not simply piled up through low-price shipments. Training and inference software platforms both showed clear progress in the 2025 annual report, which even mentioned Day 0 support for DeepSeek-V3.2; that ecosystem speed matters for domestic cards. Reuters-cited IDC data also proved that domestic AI accelerator share is moving from "almost entirely overseas-dependent" toward "meaningful local share." Bulls therefore say Cambricon once proved it "could build it," now proves it "can sell it," and will next prove it "can earn sustainably."

The strongest bear evidence also comes from the annual report itself. Customer concentration is still extremely high: the top five customers accounted for 88.66% of sales, and a change in a single large customer is still enough to rewrite the income statement. Suppliers are also highly concentrated: the top five suppliers accounted for 75.23% of procurement, and the largest supplier accounted for 55.34%. Inventory rose to 4.944 billion yuan in 2025, up 178.67% year over year, with raw materials increasing significantly. The company still uses high-alert language on the Entity List and supply-chain stability risks. Public disclosure remains limited on the mass-production path, yield, foundry, and advanced packaging partners for the latest generation cloud chip. Bears therefore say Cambricon now looks more like a single-product company in an order explosion than a platform company that has already validated its supply chain and business model.

My understanding is that both sides have caught the real issues, but their time frames differ. Bulls look more at 3–5 years and believe that as long as China has a local opportunity in high-end AI chips, Cambricon will not be absent. Bears look more at 12 months and believe that as long as customer concentration, supply-chain transparency, and cash-flow quality do not improve together, the current valuation will struggle to absorb any weaker-than-expected quarterly report. For a stock with a market cap above 770 billion yuan and trailing P/S above 90x, the latter matters more for pricing.

Valuation Analysis

Cambricon is a classic case where the valuation method must be chosen correctly before discussing numbers. It only achieved its first full-year profit in 2025. Operating cash flow was still negative in 2025. Although Q1 2026 continued to grow rapidly, single-customer exposure, single-product exposure, sanctions constraints, and inventory build-up have not been fully digested. In this situation, judging cheapness directly with 2025 static PE will almost certainly produce a misleading conclusion. I think the most appropriate main method for this company now is forward P/S, cross-checked with medium- to long-term normalized PE. DCF can be done, but sensitivity is too high for it to serve as the core anchor.

Start with cash-flow penetration. In 2025, net profit attributable to shareholders was 2.059 billion yuan, but net operating cash flow was -498 million yuan, and the operating cash flow/net profit ratio was about -0.24x. The annual report already explains this divergence: inventory increase corresponded to a cash outflow of about 3.303 billion yuan in 2025, while receivables also continued to tie up cash. On capex, 2025 "cash paid for the purchase and construction of fixed assets, intangible assets, and other long-term assets" was 569 million yuan. Depreciation, amortization of intangible assets, and amortization of long-term deferred expenses in the same year totaled about 285 million yuan. Combined with the clear rise in prepayments for fixed assets and long-term assets, I am more inclined to treat about 40% of 2025 capex as maintenance and the rest as expansion. On this basis, 2025 owner earnings were still negative, so owner-earnings PE at the current price does not hold, and FCF yield is also negative. Therefore, the valuation below first looks at revenue and cash-conversion quality rather than anchoring on 2025 accounting net profit.

How should historical valuation be understood? After listing, Cambricon was long priced as a "scarce technology story asset," not as a "profitable semiconductor company." The current price of 1,231 yuan corresponds to Reuters’ trailing P/E of about 284.6x and trailing P/S of about 93.5x. The 52-week range is 349.44 yuan to 1,448.01 yuan. In other words, even after the June pullback, the current price remains in the absolute high range of the past year. The permanent change in valuation center came because it moved from "long-term loss-making" to "first profit" and the market became willing to discount a more distant future into today, not because Cambricon has already become an NVIDIA-scale platform.

Put it next to comparables, and the expensiveness becomes more obvious. Hygon Information’s current P/S is about 38.7x and P/E about 220x, while its 2025 revenue and profit were both significantly higher than Cambricon’s. NVIDIA’s current PE is about 30.5x, and AMD’s about 148.3x. The premium given to Cambricon exists because A-share capital treats it as the purest concept mapping for Chinese AI accelerators, not because its current operating quality is best. This premium has some basis, but in magnitude it has already discounted two or three layers of long-term success together.

The valuation scenarios below are a research framework based on the company’s 2025 annual report, Q1 2026, peer multiples, and cash-flow constraints, and do not constitute investment advice. There are three core assumptions: first, 2026 revenue continues to grow, but whether the Q1 pace can beat the full year carries great uncertainty; second, net margin cannot simply annualize Q1 2025 because customer structure, product structure, and supply constraints may all change; third, multiples need to fall clearly from today, unless Cambricon can, within 12–18 months, deliver customer diversification, cash flow, supply-chain transparency, and next-generation chip mass production together.

The most important conclusion from this table is that even in the bull scenario, the odds at the current price are still unattractive, rather than the range itself. To make a 1,231 yuan share price stand, the market roughly needs to believe that Cambricon can rapidly push revenue above the 17.0 billion yuan level in 2026–2027 while multiples fall only mildly. For a company whose top five customers still account for nearly 90%, whose 2025 operating cash flow was still negative, and whose supply-chain information disclosure is limited, that premise is too heavy.

Where will expectation gaps appear? The four most important indicators are not absolute quarterly revenue: whether the top five customer share continues to decline, whether operating cash flow can turn positive for consecutive quarters, whether inventory as a share of revenue can fall, and whether the next-generation cloud chip enters statutory disclosure as "formal mass production and customer delivery" rather than "market rumors." If the next financial report still shows high revenue growth but cash flow is again swallowed by inventory, the market will again question earnings quality. If the customer base continues expanding and cash flow turns positive, part of the valuation bubble can be caught by fundamentals.

The margin-of-safety review needs a hard answer. The current price is clearly at a large premium to the conservative scenario of 350–430 yuan, so the margin of safety is zero. The most fragile assumption across the three scenarios is that "high growth will bring sustainable cash conversion," not "whether revenue can continue growing." If the cash-flow delivery and customer expansion assumptions in the base scenario are cut by 30%, the base valuation is closer to 500–650 yuan. A very simple stress test gives the same message: if earnings over the next three years stay around 3.31 yuan per share after the 2025 capitalization adjustment, and the market is still willing to give it a very generous 150x PE three years later, the corresponding share price would be only about 497 yuan, implying a three-year annualized return of about -26% versus the current price. At this level, Cambricon is a good story at a bad price. My margin-of-safety conclusion: none.

Risks, Catalysts, and Tracking Indicators

The risks that could truly cause permanent capital loss for Cambricon are several specific chains, not vague statements like "the semiconductor industry is volatile."

The first is the pullback risk after highly concentrated orders. I assign it a medium-high probability and high impact. In 2025, the top five customers accounted for 88.66% of sales, and the first and second largest customers together exceeded 3.1 billion yuan. Except for the third-ranked customer, the rest of the top five were new customers in the period. This means the 2025–2026 breakout was likely built on the rapid introduction of a few large orders. Once any of these customers slows procurement after becoming more cautious on capex, delaying validation, or replacing products, Cambricon’s revenue, gross margin, and market narrative will pull back together. The most important indicator to watch is whether "the top five customer share can fall below 80% over the next 12 months," not slogans about whether there are new customers.

The second is supply-chain risk under the Entity List. I assign it medium probability and high impact. U.S. official documents show that Cambricon Technologies Corporation Limited and multiple related entities were added to the Entity List in December 2022, with a license policy of presumption of denial. The company’s refinancing prospectus also clearly warns that some supplier products and services are scarce and proprietary, and that the company’s supply-chain stability carries certain risks. Because Cambricon has almost no overseas revenue, sanctions have limited direct impact on the revenue side. The real issue is that if advanced processes, HBM-related links, substrates, advanced packaging, or key EDA/equipment substitution see new variables, next-generation chip iteration and delivery rhythm may be slowed. The hardest part of this risk is that the company has not publicly named current foundry and packaging partners in the annual report, making through-checking difficult for external investors.

The third is profit quality and inventory risk. I assign it medium probability and high impact. In 2025, net profit attributable to shareholders was 2.059 billion yuan, but net operating cash flow was -498 million yuan. Year-end inventory was 4.944 billion yuan, up 178.67% year over year, which the company explained as an increase in raw materials. For a company, strategic stocking and demand ramp-up may both push inventory higher on the surface. The difference lies in whether inventory can turn into cash through later deliveries. If revenue growth starts slowing over the next two quarters while the inventory/revenue ratio remains high, the market will likely reinterpret "stocking" as "demand pulled forward and then backlogged." The transmission path would be fast: first compress the quality premium, then compress the growth premium.

The fourth is competition and ecosystem risk. I assign it medium-high probability and medium-high impact. Reuters-cited IDC data show that in the 2025 China AI accelerator server market, domestic vendors’ combined share had reached about 41%, with Huawei leading the domestic camp and Cambricon and Hygon each at about 116,000 cards. Meanwhile, Moore Threads, MetaX, Enflame, and other competitors are all using IPOs or financing to advance a new expansion round. For Cambricon, the real danger is that the industry has demand, but demand is captured by integrated players with stronger ecosystems, rather than that the industry has no demand. When customers buy high-end AI computing power, they rarely buy only chip parameters. They buy server systems, interconnect, software stacks, model adaptation, after-sales service, and government/enterprise delivery capability. Cambricon’s software platform is improving, but it is still far from "ecosystem lock-in."

The fifth is pure valuation compression risk. I assign it high probability and high impact. The current share price corresponds to trailing P/S of about 93.5x and trailing P/E of about 284.7x. Multiples like this do not require performance to worsen; "growth not continuing to exceed expectations" alone is enough to trigger compression. Style rotation, rate changes, and capital moving from theme stocks to semiconductor leaders with higher earnings certainty can all bring Cambricon’s valuation down first. For holders, the biggest illusion is often treating high volatility as a side effect of high growth. But when valuation is at this level, volatility itself can evolve into permanent capital loss.

Positive catalysts are concentrated. If the next two quarters continue to show high revenue growth, positive operating cash flow, and falling customer concentration, the market will accept that "2025 was not a one-time volume release." If the company formally provides statutory disclosure of next-generation cloud chip mass production, sample validation passing, or delivery rhythm to leading customers, valuation will be pushed higher again. If U.S. restrictions on AI chips to China continue tightening while domestic policy continues to favor local computing power, the substitution narrative will continue to strengthen. Conversely, any rumor about products, substrates, foundry, or potential customers that is later again clarified by the company as false will hit expectations first and the share price next.

The table below is the dashboard I think is most worth tracking continuously. The "normal range" and "warning threshold" in the table are research discipline, not company guidance. Current observations come from disclosed financial reports and market data.

Among these indicators, only three are truly most important. First, customer concentration determines whether revenue is "truly platformized." Second, operating cash flow and inventory determine whether profit is "truly cashed." Third, the statutory disclosure rhythm of next-generation products determines whether the story can keep rolling forward. The other indicators are only supporting evidence for these three. Investors who want to track Cambricon long term do not need to stare at intraday charts every day. Tracking these three things is enough.

Cross-Section and Longitudinal Summary

Longitudinally, the ability Cambricon has truly proved is "I can survive in an extremely difficult track and push products into customers’ hands at the right time," not "I can tell a big story." The company first proved architecture capability, then proved cloud products could form large revenue, and only in 2025 did it first prove it could turn revenue into profit. Its past success includes both period tailwinds, namely China’s AI infrastructure and domestic substitution trend, and execution from management and the R&D team. Otherwise it could not have delivered four consecutive quarters of numbers like 2025–2026. But its success also carries strong cyclical and policy attributes, because demand-side prosperity and supply-side compression pushed it to the front at the same time. Those success factors remain today, but the weakest link has not disappeared: customers, supply chain, and cash flow are all not yet at the state of a mature platform company.

Horizontally, Cambricon’s real advantage versus competitors is that within the visible A-share universe it has completed the leap "from loss to profit," and earlier than many private GPU startups it has received the threefold recognition of public capital markets, scaled customers, and statutory disclosure. Its weaknesses are also clear: compared with Hygon, its revenue and customer structure are narrower; compared with Huawei, its whole-machine and ecosystem capabilities are weaker; compared with NVIDIA and AMD, its software stack and global developer ecosystem lag by a very wide margin. Cambricon’s weakness is that the niche it occupies in the industry is inherently more dependent on large-customer orders and policy windows, not simply that it is "temporarily not that large." As long as the industry continues growing rapidly, this position looks good. Once the industry moves from "install equipment first" to "calculate output efficiency," this position may suddenly narrow.

The current valuation is clearly pre-consuming the future, not rewarding past success. The place where the market is most likely to misjudge is whether, even if Cambricon continues to succeed, the speed and quality of that success are enough to deserve today’s price, not whether Cambricon will fail. I think the biggest misjudgment comes from equating "first profit" directly with "business model stabilized." For most companies, first profit means an inflection point. For Cambricon, first profit only means the first step from 0 to 1 has been completed. From 1 to 10, it still depends on whether customers can diversify, cash flow can turn positive, new products can be mass-produced, and the supply chain can pass through constraints.

The key variables over one year, three years, and five years are not the same. One year is about customer structure and cash flow, because they are the bottom line for whether valuation avoids collapse. Three years is about whether the next-generation cloud chip and software stack can push Cambricon from a "single hit-product supplier" toward a "stable platform supplier." Five years is about two harder questions: whether China’s local high-end AI chip market can form a sustained and profitable domestic profit pool, and whether Cambricon can occupy a higher and more stable position in that pool than it does today. As long as the answers to these two questions remain uncertain, it is hard to say today’s share price has a margin of safety.

The bull case can be compressed into four sentences. First, 2025–Q1 2026 revenue and profit have proved that Cambricon has truly entered the large-customer procurement cycle and is no longer at the technology-sample stage. Second, four of the top five customers in 2025 were new customers, showing that the company is not only relying on renewals from one old customer and has signs of expansion. Third, the continued progress of training and inference software platforms, especially rapid adaptation to mainstream large models and DeepSeek-V3.2, shows that Cambricon is filling the most important soft-ecosystem shortcoming. Fourth, tighter export licensing of overseas high-end AI chips to China makes the domestic substitution window more real than it was two years ago.

The bear case can also be compressed into four sentences. First, almost all 2025 revenue came from the cloud product line, while edge and IP licensing are basically irrelevant to the income statement, meaning the business is unbalanced. Second, customer concentration and supplier concentration are both extremely high, and volatility on either end is enough to affect full-year performance. Third, first profit in 2025 did not turn into positive operating cash flow, so growth quality has not been fully proved. Fourth, current trailing P/S above 90x has already discounted the next two rounds of success in advance.

In a pre-mortem for a 50% loss three years later, I think there are two most likely scripts. The first happens in 2027: the large-customer procurement of 2025–2026 is proved to have been significantly pulled forward, the top two new customers enter a digestion period, and Cambricon’s full-year revenue growth falls from triple digits to below 20%. Inventory fails to come down at the same time, impairments and price concessions compress gross margin together, and main-business gross margin falls from 55% to 42%–45%. The market switches its pricing from "pure Chinese AI name" to "high-volatility chip growth stock," P/S compresses from around 90x to 25–30x, and the share price falls to the 350–450 yuan range. The second happens in 2027–2028: domestic large-customer procurement increasingly tilts toward whole-machine and integrated solutions, Huawei and Hygon take more of the profit pool on the ecosystem and server side, and although Cambricon’s next-generation cloud chip can be mass-produced, it does not form strong enough software stickiness. The company is forced to concede on price and support costs, net margin falls from the 2025–2026 high level back to the low teens, and the valuation center moves toward Hygon rather than continuing to enjoy a higher premium.

My final judgment is clear. Cambricon is a Chinese AI chip company that has completed the first step from "R&D proof" to "commercial proof." It deserves long-term tracking and even deserves to be treated as an observation center for the competitive landscape in China’s local high-end AI accelerator market. But at the 2026-06-11 price, I am not willing to write "worth tracking" as "worth owning." The essence of the current share price is that it is prepaying very high tuition for a new stage that has not yet been fully verified. The most respectable thing about the company is that it finally turned the income statement positive. The most cautionary thing is that cash flow, customer concentration, and supply-chain transparency have not kept up with valuation. If two of these three fail to keep improving, today’s holders will struggle to receive a decent risk-reward ratio.

【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: Avoid

• One-sentence investment thesis: The earnings inflection has appeared, but the current price has nearly bought the next two rounds of execution in advance.

• Three-tier price signals: Ideal buy price: 280–340 CNY

• Acceptable hold price: 510–870 CNY

• Clearly overvalued price: 1,080–1,260 CNY

• Current price classification: Clearly overvalued

• Worth waiting for a better price: Yes. The trigger is not only falling to the buy point, but also seeing the top five customer share fall below 80% and operating cash flow turn positive for two consecutive quarters.

• Target holding period: If the stock enters the buy zone in the future and meets operating conditions, prioritize 3–5 years; new positions are not recommended at present.

• Expected annualized return: Based on the current price, about -35%/year in the conservative case, about -22%/year in the base case, and about -7%/year in the bull case.

• Maximum loss risk: If large customers cut orders, inventory digestion falls short, and valuation compression overlaps, a 60%–70% drawdown within 12–24 months is not impossible.

• Signals that trigger reassessment: Operating cash flow positive for two consecutive quarters, with OCF/net profit returning above 0.8x;

• Top five customer revenue share falling below 80%;

• Main-business gross margin stable above 50% for two consecutive quarters and inventory/revenue ratio clearly falling;

• Next-generation cloud chip appearing in statutory disclosure with clear mass-production and leading-customer delivery language;

• If new sanctions block key supply-chain links, all scenarios need to be revised down immediately.

【Ideal/Fair Buy Price】280–340 CNY Rationale: This corresponds to the conservative intrinsic value of 350–430 yuan with at least a 20% margin of safety applied. Only in this price band would investors have enough room to cover the triple uncertainties of customer concentration, supply chain, and cash flow.

【Valuation Range】

• current: 1231 (as of the 2026-06-11 close)

• bear (conservative · ideal buy zone): [280, 340]

• base (reasonable · acceptable hold zone): [510, 870]

• bull (optimistic · above the clearly overvalued line): [1080, 1260]

Research uncertainties:

• The company has not disclosed the real names of major customers in the annual report, so external research cannot precisely map order sustainability to specific purchasing entities.

• The company has not publicly disclosed current wafer foundry and advanced packaging partners, so true supply-chain flexibility under the Entity List can only be partially observed.

• The market closely watches next-generation cloud chip progress, but as of the 2025 annual report, first-hand wording remains principle-based and lacks hard indicators on mass production, yield, and customer validation.

• Chinese GPU startups are mostly unlisted entities, so the financial and product information available for horizontal comparison is far less than for listed companies.

• Whether the large profit increase in 2025 and acceleration in Q1 2026 contain concentrated-recognition elements requires at least two more quarters of observation to improve confidence.

Reference sources: Company 2025 annual report, 2025 semiannual report, 2025 first-quarter report, 2025 third-quarter report, 2026 first-quarter report, 2025 annual equity distribution implementation announcement, announcement on first profit and removal of stock abbreviation mark U, and 2025 and 2022 announcements related to issuance to specific investors; Entity List documents from the U.S. Federal Register and eCFR; export-control related filings submitted by NVIDIA and AMD to the SEC; the "Action Plan for High-Quality Development of Computing Power Infrastructure" by six ministries including MIIT; public market data and news releases from Reuters, TradingView, Google Finance, FT/Reuters, and others.

Other Securities Mentioned in the Report

• 688041.SHG — Hygon Information, the most important A-share Chinese computing-power comparable, with broader business and a current valuation significantly below Cambricon’s.

• NVDA.US — NVIDIA, the center of the global AI accelerator profit pool and a direct reference point for China export controls and domestic substitution logic.

• AMD.US — AMD, whose MI308 is also subject to China export licensing constraints and serves as an important benchmark for second-tier overseas GPU competitors.

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