Hygon Information In-Depth Research

Qualitative profile label: in valuation reshaping. The basis is direct: it looks more like a company that has completed first-stage validation while the market is already pricing it as if the second stage has succeeded. It is not a cash cow, because R&D intensity remains high and profit has not been fully released. It is also not pure high-quality compound growth, because the second curve lacks transparency and profit realization lags revenue expansion. Nor is it a turnaround, because the core business has not collapsed.

2. The Company's Longitudinal Development History

The starting point was a designed industrial gap, not a start from zero

Hygon Information was founded in 2014. It emerged against the backdrop of a long-standing domestic gap in high-end CPUs and a national push to address weaknesses in server and key-industry basic software and hardware, rather than as a traditional garage start-up. The prospectus lays out the policy background clearly: from the 13th Five-Year Plan to information-industry development guidance, high-end CPUs, chip and complete-system coordination, and AI-related technologies were repeatedly named. From the beginning, Hygon was filling a gap that had already been confirmed by both policy and industry, but where market-based supply was scarce. It was not searching for a new demand.

What truly determined Hygon's early path was its relationship with AMD. According to the prospectus, as of its signing date, Chengdu Hygon Integrated Circuit Design was 70% held by Hygon Information and 30% by AMD; Chengdu Hygon Microelectronics was 49% held by Hygon Information and 51% by AMD. The former handled high-end processor R&D design and sales; the latter handled core R&D design for high-end processors. In other words, Hygon's starting point was entry into the high-end CPU world through AMD licensing and a joint-venture structure, rather than completely self-developed technology from scratch. This starting point gave it the fastest route to productization, and also planted the core controversy of the following decade: after a licensed start, how far can independent evolution go?

The first stage was climbing with a ladder

After 2016, Hygon took a very pragmatic route: first build high-end x86-compatible processors, then put those chips into domestic complete machines, servers, and industry systems, allowing users to complete substitution with as little software-stack change as possible, instead of adapting from zero to a new world. Compared with routes such as Loongson, Phytium, and Kunpeng, Hygon's value lies in transition cost, not in the originality of the instruction set itself. Its CPU story has never been the purest one, but for many government and enterprise customers it may be the route with less migration friction. The prospectus also did not avoid external benchmarks: in general-purpose processors, Intel and AMD CPU products still hold an absolute advantage globally. Hygon chose to build China's localized alternative at the edge of that dominant ecosystem.

This route worked largely because of Sugon. Sugon is both the largest shareholder and one of the strongest partners in complete systems and industry delivery. By the 2025 annual report, Sugon still held 27.96% of Hygon and remained its largest shareholder. Hygon designed chips; Sugon sold servers, storage, security, and data-center products. The two roles naturally fit together: one provided the "heart," the other the "body." This industrial fit explains why Hygon formed a commercial loop faster than many domestic peers that only make chip IP or single products.

The second stage was being sanctioned before listing

In June 2019, the U.S. Department of Commerce added Higon, Chengdu Hygon Integrated Circuit Design, Chengdu Hygon Microelectronics, Sugon, and related entities to the Entity List. For Hygon, this was the most important break point in its commercial history, not later geopolitical noise. The prospectus says directly that after being added to the Entity List, AMD no longer provided relevant technical services, and the company could only realize subsequent product and technology iteration by itself. Hygon was forced to prove its real capability at the moment external technical support was cut off, not during the IPO roadshow.

The importance of this moment is easily obscured by later high-growth financial statements. In hindsight, the Entity List was both bad news and a filter. The bad news was long-term constraints on advanced foundry capacity and access to some technologies. The filter was this: if the company could still keep launching products that customers could buy, install, and adapt after that point, it proved at least two things. First, the original license was not the whole of its capability. Second, customers could not continue buying only because of a domestic-substitution slogan. A sizable part of Hygon's valuation today is paid for the ability to survive after 2019, not for its 2016 starting point.

The third stage turned an R&D race into a capital race through listing

Hygon listed on the STAR Market of the Shanghai Stock Exchange on 2022-08-12 at an issue price of 36 yuan. Reuters at the time described it as a "poster child" for Beijing's technology self-reliance and noted that the listing was a bet on China's vision of "dual autonomy" in high-end processors. Listing meant more than fundraising. It gave Hygon a capital-market ticket that could support high R&D, heavy equity incentives, and ecosystem investment over the long term. High-end CPUs and GPGPUs are not one- or two-year businesses. Only those that can endure the R&D cycle are qualified to discuss the business model.

From then on, Hygon's narrative began to shift. In the early listing period, the market mainly viewed it as a domestic x86 CPU name. After 2024, as domestic AI infrastructure investment accelerated, overseas high-end GPU supply to China became restricted, and the domestic AI chip narrative heated up, Hygon began to be re-read as a dual-engine CPU plus AI accelerator asset. This is a change in the valuation coordinate system, not a simple addition of themes: a pure domestic CPU company can be valued as Xinchuang hardware, while a company that can tell an AI accelerator story is pulled into a pricing framework with higher volatility and more imagination.

The fourth stage is rapid profit growth, while the market cares more about the unsplit revenue

2025 was the clearest acceleration in Hygon's commercial history. The annual report shows full-year revenue of 14.377 billion yuan, up 56.92% year on year; net profit attributable to shareholders of 2.545 billion yuan, up 31.79%; and net operating cash flow of 2.097 billion yuan, up 114.61%. Management explicitly attributed growth to the explosion in AI computing power, digital infrastructure upgrades, product-performance advantages, service capability, and ecosystem synergy. The problem is that the annual report still lists the main product only as "high-end processors" and does not disclose revenue or gross margin for CPU and DCU separately. External investors can see the result, but not the structure.

Q1 2026 enlarged that contradiction again. Revenue grew 68.06% year on year, net profit attributable to shareholders grew 35.82%, and the company also disclosed that net profit attributable to shareholders excluding the impact of share-based payment grew 74.86%. This means reported profit growth is already fast, but operating elasticity is stronger once equity-incentive expense is excluded. Conversely, it also shows that share-based payment is no longer a negligible small number. Hygon now looks like a high-speed boosted machine, but under the hood the real pressure is borne by R&D and incentives.

The failed restructuring was the climax of the longitudinal narrative and also a reminder

In June 2025, Hygon proposed a share-exchange absorption merger with Sugon. The market saw an almost complete domestic computing-power chain: chips, servers, data centers, software, industry customers, and ecosystem partners, nearly all placed into one listed platform. The proposal summary stated clearly that Sugon, the target, was Hygon's largest shareholder; after completion, Hygon would still have no actual controller. On December 9, the transaction was terminated. The reasons given by both sides were that the plan required a long verification period, involved many parties, faced major changes in the market environment, and did not yet have mature implementation conditions. They also emphasized that termination would not affect subsequent cooperation and that they would continue to promote optimized collaboration across chips, hardware, and software as well as AI full-stack product R&D.

In hindsight, this episode did not change Hygon's business nature, but it exposed the easiest mistake capital markets make when valuing it: mistaking "strong synergy" for "no boundary." The synergy between Hygon and Sugon is real, and the scale of related-party transactions is large enough. But the stronger the synergy, the more investors need to watch governance, related-party pricing, and the boundary of independence. The failed restructuring did not remove this issue. It only changed "legal integration" into "continued commercial binding."

3. Longitudinal Financial Review

Hygon's financial appearance over the past two years has been strong, but the real issue is growth quality. In 2025, revenue increased 56.92% year on year, but operating cost increased 82.40%, and gross margin fell to 57.78%, down 5.92 percentage points from the previous year. In other words, scale accelerated, but margins did not expand in parallel. The company's own explanation was specific: price increases in upstream raw materials and storage auxiliary materials, tight foundry and packaging/testing capacity, higher processing fees, and business-scale expansion jointly pushed up cost. For a design company, outsourced manufacturing is normally an asset-light advantage. When supply is tight, that advantage becomes a margin-transmission channel.

The expense side tells the same story. In 2025, selling expenses increased 260.69% year on year; the company attributed this to rapid expansion of the marketing team and higher spending on market promotion and ecosystem building. R&D expenses increased 42.45%; the company also disclosed a 28% year-on-year increase in R&D headcount and a decline in the R&D capitalization rate from 17% in the previous year to 9%. Taken together, these changes mean Hygon is front-loading and expensing more costs, pressing future investment more directly onto the current income statement. In the short term, this weighs on margins. In the medium term, it improves accounting quality, because profit is not being heavily beautified through capitalization.

Cash flow says more than profit about Hygon's current stage. In 2025, net operating cash flow was 2.097 billion yuan, versus net profit attributable to shareholders of 2.545 billion yuan, implying OCF/net profit of about 0.82x. In 2024, net operating cash flow was 977 million yuan; reversing from the year-on-year growth rate disclosed in the 2025 annual report, 2024 net profit attributable to shareholders was about 1.931 billion yuan, so OCF/net profit was only about 0.51x. Cash conversion is improving but is not yet stable. In Q1 2026, the indicator weakened sharply again: net operating cash flow was only 68 million yuan, down 97.32% year on year. The company explained that material stocking payments increased and that the same period last year had received more contract prepayments. The explanation is plausible, but it also reminds investors that when Hygon's revenue expands rapidly, working capital can consume cash aggressively. Profit alone is not enough.

From the balance sheet, Hygon looks more like a design company that has put cash, inventory, prepayments, and R&D staff into the balance sheet than a heavy-asset factory. At the end of Q1 2026, fixed assets were only 858 million yuan, but inventory reached 7.333 billion yuan, prepayments reached 3.440 billion yuan, and accounts receivable were 3.833 billion yuan. Compared with the end of 2025, prepayments increased about 19.2%, inventory increased about 14.5%, and accounts receivable declined about 5.0%. What does this look like? A company preparing ahead of time for larger shipments while also trying to collect receivables formed earlier. It is the typical working-capital tug of a high-growth hard-tech company, not an immediate liquidity alarm.

The table below puts the most important financial indicators together. It does not mechanically fill every accounting line item; it keeps only the lines truly related to business quality: growth, margins, cash flow, and R&D intensity.

† 2024 net profit attributable to shareholders is reversed from the year-on-year growth rate disclosed in the 2025 annual report. ‡ 2024 gross margin is reversed from the disclosure that "2025 gross margin decreased by 5.92 percentage points from the previous year." Data sources: Hygon Information 2025 annual report and 2026 first-quarter report; some figures are calculated from disclosed data.

The three things worth remembering from this table are more granular than the near-doubling revenue speed. First, margins are moving down, which means growth contains higher costs and heavier investment. Second, cash-flow coverage of profit is volatile; Hygon is not an easy business where profit naturally converts quickly into cash. Third, the company is pressing the accelerator simultaneously on high R&D, high equity incentives, and high stocking, which shows aggressive judgment on future market opportunities. This financial shape is most vulnerable when revenue slows, because profit, cash flow, and valuation can all be squeezed at the same time, rather than merely losing a little short-term profit.

4. Share-Price and Valuation History

Since listing, capital markets have applied three labels to Hygon. The earliest was "domestic x86 CPU," a pricing method tilted toward industrial security and Xinchuang logic. Later, as the AI computing-power narrative heated up, it gained the additional label of "domestic AI chip platform." When the proposed Sugon absorption merger appeared in June 2025, the market briefly traded it as a "domestic computing-power ecosystem integration platform." Once the label changes, the valuation center changes with it.

This re-rating has been intense over the past year. Public market pages show Hygon's 52-week low at 132.01 yuan and high at 358.80 yuan. In the single month from 2026-05-11 to 2026-06-11, the share price moved from near 349 yuan at the high, down to 259.70 yuan, and back to 288.86 yuan. More interestingly, the violent swings from late May to early June came from overall sentiment in the AI hardware sector, residual heat from the M&A narrative, and rapid shifts in risk appetite toward high-valuation assets, not from fundamental changes at the company within a few days.

Putting the current valuation back into the financial statements makes the picture clearer. Based on the 2026-06-11 closing price of 288.86 yuan and total share capital of 2.32434 billion shares, Hygon's market value is about 671.4 billion yuan. Based on 2025 revenue of 14.377 billion yuan, net profit attributable to shareholders of 2.545 billion yuan, and equity attributable to shareholders of 22.493 billion yuan, the current static P/E is about 263.8x, P/S about 46.7x, and P/B about 29.8x. This is clearly pricing it on the side of "scarce national-team asset plus AI optionality," not as an ordinary domestic semiconductor design company.

Why did the valuation center move up? Because the market began to interpret Hygon's revenue in two layers, not because the business model suddenly changed. The first layer is the realized CPU business: x86 compatibility, Xinchuang penetration, complete-system channels, and industry customers. The second is the not-yet-fully-realized DCU business: as long as demand for domestic AI accelerators is growing, the market is willing to capitalize part of the future imagination in advance even without sufficient segment revenue disclosure. The problem is that this valuation uplift is reversible. If DCU realization falls short of expectations, or if high revenue growth comes more from CPU and channel volume than AI-chip penetration, the valuation may move back to a coordinate system closer to a "high-growth CPU design company."

5. Business Model and Moat

Hygon's real business machine has at least three layers, far more than "invent a chip and sell it." The first layer is processor design and iteration. The second is adaptation and delivery with servers, complete systems, industry systems, and ecosystem partners. The third is using related channels and industrial-chain coordination to place chips into end customers' budgets. This is why Hygon's annual report explains growth not only through product performance, but also specifically through "deepening industry collaboration with vendors and ecosystem partners and expanding application fields." It earns revenue from chip shipments plus ecosystem penetration, not merely from IP licensing fees.

The biggest issue in revenue structure is what is not disclosed, not what is disclosed

In the 2025 annual report, the product section of the main business still combines all main products under "high-end processors," with revenue of 14.362 billion yuan and gross margin of 57.78%, and does not split CPU and DCU. In the segment-information section, the annual report again says financial information for reportable segments is shown in the main-business analysis, effectively still a consolidated view. For outside investors, this matters a lot: without a CPU and DCU split, it is impossible to judge accurately which line is driving growth and which line is diluting or lifting margins. Hygon's greatest information advantage sits inside the company; the greatest information discount also comes from there.

From known numbers, 2025 revenue grew sharply, but gross margin declined and R&D investment stayed intense. This looks more like a mixed pattern of "mature CPU business scaling plus new business continuing to spend heavily" than single-hit-product growth. The Q1 2026 report reinforces this impression: the company attributes growth to the explosion in AI computing power, digital infrastructure upgrades, product-performance improvement, and an open ecosystem strategy, but still does not provide segment revenue or gross margin. For Hygon, revenue growth itself has already been validated; the valuation question is how much future incremental revenue is an extension of CPU and how much is a new DCU profit pool.

The company has operating leverage, but not easy leverage

Hygon has the typical operating leverage of a design company: R&D teams, software ecosystem, validation platforms, and sales systems are all front-loaded fixed costs, so in theory margins can expand as revenue scales. The issue is that Hygon is still reinvesting much of the leverage that should be enjoyed later back into the front end. Selling expenses surged in 2025, and the R&D expense ratio remained near 30%, showing the company is still fighting for position and does not yet want to harvest profit. This is very different from mature CPU leaders and also different from many single-point AI chip companies: Hygon is choosing high investment, not being forced into it.

When revenue declines, Hygon's hardest costs to compress are R&D and ecosystem spending, not factories and depreciation. The fabless model means it does not carry massive depreciation like an IDM, but high-end CPUs and AI accelerators are by nature a long race of continuous spending on R&D and software adaptation. In Q1 2026, R&D investment was 1.211 billion yuan, accounting for 30.03% of revenue. Net profit attributable to shareholders after excluding share-based payment was 923 million yuan, about 236 million yuan higher than the statutory figure. In other words, this company already has profit, but to maintain its competitive position it must accept that R&D and incentives will continue to eat part of that profit.

Three moats are real; one and a half in the market pitch are only partly real

The first real moat is the x86-compatible niche. The backbone of the global server CPU ecosystem has long been in the hands of Intel and AMD, and the prospectus acknowledges this. Hygon is not a global player, but its value in China is that it gives server customers needing domestic substitution an option with relatively lower migration friction. Compared with fully proprietary ISA routes, this compatibility is a practical procurement reason, not a romantic technology story.

The second real moat is channel and complete-system synergy. Behind Hygon stand Sugon and a set of industry customers, complete-system integration capabilities, and data-center solution entrances. It is not sitting alone with chips waiting for customers to knock. The 2025 annual report shows the company sold 8.140 billion yuan of goods to "Company A and other companies under its control." In April 2026, the company expected total recurring related-party transactions over one shareholder-meeting cycle to reach 13.949 billion yuan, of which expected sales of goods and services to "Company A and other companies under its control" alone reached 11.015 billion yuan. This binding certainly brings a governance discount, but it also shows that Hygon sells chips inside a system, not in isolation.

The third real moat is organizational capability to keep iterating products under Entity List constraints. After being placed on the Entity List in 2019, AMD no longer provided technical services, and the company stated clearly in the prospectus that subsequent product and technology iteration was achieved by itself. This capability does not equal global leadership, but it at least shows Hygon is not a company that loses its ability to operate once the technology license stops. For a Chinese high-end processor company, that is enough to form a survival barrier over a 3-5 year horizon.

The "one and a half" moats in the market pitch are DCU and the AI ecosystem. The half that is real is that domestic customers do have real demand for domestic AI chip substitution. Reuters cited IDC data saying that in 2025, Chinese local GPU and AI chip vendors had captured nearly 41% of China's AI accelerator server market, while NVIDIA's leading position was shrinking. The other half is not yet real: Hygon has not publicly disclosed enough data to prove it has achieved a sufficiently high position inside that 41%, nor has it provided public numbers proving that the migration cost of its CUDA-like ecosystem has fallen enough to confront the industry leader directly. This remains imagination, not financial-statement proof.

Governance is not a fatal flaw, but it is not a small issue to ignore

The most distinctive feature of Hygon's governance structure is that synergy and independence coexist. The transaction proposal and termination announcement both emphasized that the company has no actual controller, which gives it a certain neutrality in legal structure. But the 2024 annual report also shows that director Li Jun concurrently serves as director and president of Sugon and holds positions in several Sugon-system subsidiaries. Board-level cross-appointments, the largest-shareholder position, and large related-party transactions make it hard to value Hygon as a fully independent pure chip-design company.

The company has not been found to have major financial fraud, major penalties, or audit abnormalities, and the annual report also discloses no serious violations over the past three years. That is a positive. But a governance discount does not have to come from illegality; it can also come from unclear boundaries. Hygon's issue is whether synergy will make pricing power, customer relationships, and profit attribution too ambiguous, not whether synergy exists. For an already very expensive stock, the market can amplify that ambiguity.

6. Industry and Cycle Analysis

Hygon stands at the intersection of two cycles, not in a single industry. The CPU line is closer to a long-term domestic-substitution and enterprise IT refresh cycle. The DCU line is closer to an AI infrastructure capex cycle. The former moves more slowly but has stickier customers; the latter has greater elasticity but faster technology iteration, and its pricing system and ecosystem pattern can be rewritten in one or two years. Hygon is valuable because it sits on both lines. Hygon is risky because it is exposed to both cycles at once.

The profit pool for CPUs is still defined by the international x86 ecosystem

The prospectus says plainly that in general-purpose processors, Intel and AMD CPU products occupy an absolute advantage in the global market. Hygon is making "x86 compatibility plus domestic supply chain plus local ecosystem adaptation" into a usable solution in the Chinese market, not challenging that global structure. The most direct gap it fills is the need in key industries, under domestic-substitution pressure, for a server CPU option with lower migration cost and better ecosystem compatibility. It is not absolute top-tier performance leadership.

The profit pool in this market line mainly sits in server procurement, industry IT upgrades, government and party systems, and localization in key industries. Industry profit is jointly determined by chips, complete systems, adaptation, operations and maintenance, and procurement qualification, not by a single chip parameter. This is exactly why a complete-system and system partner such as Sugon is so important to Hygon. Discussing CPU benchmarks alone cannot explain Hygon's commercial position; the key is whether it can be put into a complete delivery solution.

The DCU cycle looks more like a capex wave

The decisive variable in the AI accelerator market is who actually receives training and inference budgets, not whether a company announces a new product. Reuters cited IDC data in April 2026 saying that in 2025 Chinese local GPU and AI chip vendors had captured nearly 41% of China's AI accelerator server market. This shows that against the background of restricted high-end GPU supply to China, domestic-substitution demand has moved from slogan to orders. But that 41% is a cake cut jointly by many vendors, not Hygon's share alone. It includes Huawei Ascend, Cambricon, Moore Threads, Muxi, MetaX, and many other players. The industry is booming, and competition is genuinely harsh.

For Hygon, the biggest challenge for the DCU line is why customers should choose it rather than someone else, not whether demand exists. In AI chips, the software stack and migration cost are often more important than theoretical peak hardware performance. NVIDIA is formidable not only because its cards are powerful, but because CUDA makes it hard for developers and customers to leave. If Hygon wants DCU to become a real second profit curve, it must cross not only the chip-performance threshold, but also the slower and harder threshold of toolchains, framework adaptation, stability, and customer usage inertia.

Policy and geopolitics are operating variables for Hygon, not background noise

Hygon has been inside the policy cycle since birth. The policy path summarized in the prospectus shows that since 2016, high-end CPUs, core chips, complete-system coordination, and AI have all been included in national industrial directions. Later, Xinchuang, digital infrastructure, and localized procurement logic were added, so part of Hygon's demand side is naturally supported by policy. It does not live entirely off policy, but policy does determine whether its market ceiling can continue rising.

On the other side, geopolitical constraints are equally specific. In 2019, Hygon and related entities were added to the U.S. Entity List. In May 2025, the U.S. Department of Commerce announced the rescission of the AI Diffusion Rule issued on January 15, while also strengthening chip-related export controls. On the surface, rules are changing. In substance, one thing has not changed: export restrictions related to advanced computing and semiconductors remain a long-term variable that continues to tighten and adjust. Hygon's uncertainty is where the next round of impact lands: foundry, EDA, IP, packaging and testing, interconnect, or customer procurement pace. The question is not whether it will be affected.

Therefore, Hygon belongs simultaneously to the policy cycle, technology-iteration cycle, and capex cycle. In the upcycle, the most beneficial variables are the pace of domestic substitution and the release of AI infrastructure budgets. In the downcycle, the most fragile variables are failure of the high-valuation narrative, weaker-than-expected validation of the DCU ecosystem, and any sudden supply-chain restriction.

7. Horizontal Peer Analysis

The hardest part of analyzing Hygon is that there is no truly clean peer set. Looking only at CPUs misses the optionality of DCU; looking only at AI chips misses the already mature and profitable CPU base. The better method is to admit that it spans two profit pools and then examine its position in both coordinate systems, rather than forcing a basket of "identical" companies.

First clarify the competitive landscape

On the CPU side, Intel and AMD set the ecosystem benchmark. Intel represents the legacy mainstream x86 data-center ecosystem. AMD is both the important shadow behind Hygon's technology origin and the most realistic advanced route in current global server CPU competition. Among domestic listed peers, Loongson is the CPU name most often placed next to Hygon, but the two routes differ substantially: Loongson emphasizes proprietary instruction sets and autonomy, while Hygon emphasizes x86 compatibility and lower migration friction.

On the AI accelerator side, the only global reference is NVIDIA. It is not only the hardware leader, but also the monopolistic definer of software and developer ecosystems. Domestically, the most important names to track are Huawei Ascend, Cambricon, Moore Threads, Muxi, MetaX, and other local players. But Hygon's public DCU financial split is insufficient for a fully precise same-basis comparison. Among the same-day public market data obtained, I would rather use NVIDIA as the global capability ceiling, Loongson as the domestic CPU reference, and Cambricon as a narrative reference, rather than force a false "same-category valuation table."

The data table answers only half the question

Data sources: Hygon calculated from the 2026-06-11 closing price and 2025 annual report; Loongson uses the same-period Google Finance page; NVIDIA uses the same-period Yahoo Finance page.

This table shows two things. First, Hygon is no longer an ordinary CPU company in the domestic capital market: its market value is about 12.6x Loongson's. Second, compared with NVIDIA, the world's strongest AI chip company, Hygon is already more expensive on both sales and earnings multiples. At this level of expensiveness, the market is clearly buying a much larger future, not the 2.5 billion yuan profit already earned in 2025.

What each company has become explains why customers and capital price it differently

AMD is the most explanatory global reference for Hygon, because Hygon's starting point was AMD licensing and it later had to prove it could follow an R&D route that had diverged from AMD, not because Hygon is competing with AMD for global share. The prospectus states clearly that since the technology license agreement was signed, the two sides have conducted R&D independently; AMD's processor core microarchitecture and SoC architecture have been updated, and product and technology paths have diverged. Hygon's most valuable feature today is that it is no longer purely dependent on AMD, not that it is like AMD.

Intel is more of an ecosystem reference for Hygon. It remains important because much of enterprise server procurement compatibility, software stack, and usage inertia was originally defined in the Intel/AMD world, not because Chinese customers are now using Hygon to replace Intel's global status at scale. Hygon's ability to enter depends on that world still operating. Put differently, Hygon is a domestic substitute living within a mature ecosystem, not a builder of an ecosystem from scratch.

Loongson is another way of living. Loongson's story is about more pure autonomy and control, but purity has a cost: higher migration cost, slower ecosystem switching, and more need for policy pull. Hygon does not have Loongson's degree of narrative purity, but may be more easily accepted by government and enterprise customers in commercial deployment. Capital markets therefore price the two very differently: Loongson looks more like a long-term autonomy option, while Hygon looks more like a profitable domestic-substitution platform.

NVIDIA represents the hardest line for Hygon to cross today: whether the software and developer ecosystem can form a flywheel, not whether chips can be made. If Hygon ultimately succeeds in DCU, its likely method will be to sell a "good enough and more available" solution in specific Chinese scenarios through policy, supply-chain constraints, industry customer demand, and local adaptation. It will not be a copy of global CUDA. The budget it directly competes for is in constrained scenarios, not the entirety of NVIDIA's global profit pool.

Hygon's ecosystem position

Hygon's real position in the industry is "a challenger with a certainty base," neither a leader nor a simple follower. In the CPU line, it challenges the shortage of domestic high-end server CPUs. In the DCU line, it challenges domestic substitution for AI accelerators under Chinese market constraints. The gap it fills is not that no one in the world can do it, but that under China's constraints there are too few solutions that can be supplied at scale, enter systems, and be accepted by key industries. This is the fundamental reason the market pays a high price for it.

8. Current Fundamentals and Bull-Bear Debate

Over the most recent four quarters, the biggest change at Hygon is that revenue acceleration has outpaced profit acceleration, not that it moved from low growth to high growth. Full-year 2025 revenue grew 56.92%, and net profit attributable to shareholders grew 31.79%; Q1 2026 revenue grew 68.06%, and net profit attributable to shareholders grew 35.82%. Excluding share-based payment, adjusted Q1 net profit attributable to shareholders grew 74.86%. This shows two things happening at once: demand is strong, and expenses and incentives are heavy. The company is expanding the market while also putting more R&D and incentive costs into the income statement.

The most important Q1 issue to unpack is cash flow and working capital, not revenue. Net operating cash flow fell 97.32% year on year. The company attributed this to higher stocking payments and the impact of larger prepayments in the same period last year. The balance sheet supports this explanation: inventory and prepayments rose together, while accounts receivable declined slightly. For a high-growth design company such as Hygon, this is not an immediate red light, but it tells the market that if revenue growth slows after 2026, cash will come under pressure before revenue does.

The market currently trades Hygon on the overlap of three narratives. The first is performance growth, because the 2025 and 2026Q1 numbers are solid. The second is domestic AI substitution, especially when overseas high-end GPU supply is controlled and local AI chip vendors' share is rising. The third is the afterglow of ecosystem integration: although the Sugon absorption merger was terminated, the "chip, complete system, software, AI full stack" story has not fully dissipated. The problem is that among these three narratives, only the first has been sufficiently realized in financials. The latter two still contain a large amount of expectation.

The bulls' strongest evidence now comes in three categories. The first is performance: revenue is growing rapidly and still accelerating. The second is structure: the company explicitly attributes growth to both general computing and AI computing markets, while maintaining high R&D investment. The third is ecosystem: although related-party transactions and channel synergy raise governance concerns, they also show that Hygon is already inside the delivery chain for large projects and industry customers. It is not a laboratory company with decent technology but no sales.

The bears' strongest evidence also comes in three categories. The first is transparency: the company has not disclosed a CPU/DCU split, so the market cannot verify the realization of the AI narrative. The second is valuation: current P/E, P/S, and P/B are all extremely high and are already hard to explain with 2025 profit. The third is quality: declining gross margin, significant share-based payment, volatile cash flow, and concentration in related customers. Any single item may not be enough to overturn the investment logic, but together on an ultra-high-valuation stock, they amplify drawdown risk.

9. Valuation Analysis

Choose the right valuation method before discussing numbers

Hygon cannot be valued with only one method. Using only P/E would undervalue the optionality of DCU that has not fully entered the income statement. Using only P/S would mix the mature and profitable CPU business with an AI business still under validation. A more realistic approach is to treat it as a high-growth semiconductor design platform, cross-check forward P/E and forward P/S, and then apply a discount for cash-flow quality.

Historical valuation and current position

As of 2026-06-11, Hygon's share price was in a rebound position after sharp volatility over the previous month. It still had visible distance from the 52-week high of 358.80 yuan, but had already more than doubled from the 52-week low of 132.01 yuan. Based on the 2025 annual report, the current static P/E is about 263.8x, P/S about 46.7x, and P/B about 29.8x. Using NVIDIA, the world's strongest AI chip ecosystem company, as a reference, NVIDIA's current trailing P/E is 31.88x and P/S 20.09x; Hygon is more expensive. This comparison is meant to show that the market is already granting Hygon a very expensive scarcity premium, not an average domestic-substitution valuation. It is not meant to argue that Hygon has no value.

How to compare peer valuations without misleading yourself

Domestic investors often place Hygon alongside Cambricon and Loongson. But these three companies correspond to different profit pools: Loongson is more about the proprietary-ISA CPU logic, Cambricon is more about pure AI chips and high-elasticity expectations, and Hygon is a hybrid of CPU cash flow plus AI optionality. Mechanical comparison using the same multiple easily produces the wrong conclusion. A more explanatory set of references is this: Loongson gives a lower-bound reference for the autonomous CPU route, NVIDIA gives an upper-bound reference for the AI accelerator ecosystem, and Hygon captures a dual premium in between. From that angle, Hygon's market value is about 12.6x Loongson's, while its valuation relative to NVIDIA is not cheap.

Cash-flow look-through

In the verifiable past two years, Hygon's operating cash flow/net profit ratio improved from about 0.51x to 0.82x, but it is still not stable. In Q1 2026, the ratio fell again to about 0.10x. The core difference is not heavy-asset capex, because the company's fixed assets are only several hundred million yuan and the fabless model means factories do not support the asset base. The real drags on cash conversion are stocking, prepayments, accounts receivable, high-intensity R&D, and share-based payment. Public disclosures do not separately list maintenance capex and expansion capex. I do not think it is appropriate to mechanically value Hygon by multiplying net profit by a high multiple, nor do I think all investing cash flow should be treated as maintenance expenditure. The more reasonable approach is to recognize a cash-conversion discount in the profit-based valuation.

Absolute valuation scenarios

The following research scenarios are based on public information and do not constitute investment advice. My method is to set three levels of 2026E operating assumptions based on 2025 and 2026Q1 growth, margins, and cash-flow quality; then cross-check forward P/E and forward P/S to derive an implied value per share range. Because the CPU/DCU revenue split is missing, the multiples are not as aggressive as the current static valuation and include discounts for cash-flow quality and information opacity.

Scenario basis: the current static valuation is extremely high, showing that the market has already pulled some optimistic expectations forward. Upside comes from DCU realization and margin stability; downside comes from valuation compression, not necessarily from earnings decline alone. Put differently, Hygon's upside mainly depends on proving it is more than CPU, while downside may come from the market reclassifying it back as CPU.

Expectations-gap analysis

The market's current embedded expectations include at least three points: first, 2026 revenue will still maintain high growth; second, AI-related business will continue increasing its share of revenue; third, gross-margin decline will stabilize at some stage. These are also the easiest places for expectations gaps to appear. If subsequent public disclosures still do not split CPU/DCU and total revenue growth begins to slow, the market will find it increasingly difficult to keep assigning such high multiples on an "AI platform" basis. Conversely, if the company starts using clearer customer cases, ecosystem adaptation, and product-shipment structure to prove DCU's presence, Hygon may gain a second round of valuation support.

Margin-of-safety review

Under the scenarios above, the current price of 288.86 yuan sits at a clear premium to the bearish implied value of 200-220 yuan, leaving no margin of safety. The most fragile assumptions across the three scenarios are DCU penetration speed and margin stability, not CPU demand. If both AI realization speed and margin assumptions in the base scenario are cut by 30%, base value quickly contracts to around 200-225 yuan. Put more simply, at the current static P/E of about 264x, the apparent earnings yield is only about 0.4%. Even if profit is flat for the next three years and valuation does not expand, that entry return has almost no cushion. Hygon may well be a good company, but for new capital today it looks more like a bad price. My margin-of-safety conclusion: none.

10. Risk Analysis

The first risk that could truly cause permanent capital loss is supply-chain and export-control risk. I assign a medium-high probability and high impact. Hygon was added to the Entity List in 2019, and the prospectus clearly stated that AMD no longer provided technical services afterward and that the company advanced subsequent iteration by itself. In 2025, the U.S. BIS continued to adjust AI diffusion rules and chip export controls, which also shows the rule environment is unstable. The indicators to watch are product-iteration pace, material prepayments, inventory changes, and any disclosure about restrictions on access to foundry, packaging/testing, IP, or EDA, rather than a single news headline. If this area goes wrong, the transmission path will be fast: first product schedule and customer orders, then working capital, then gross margin and valuation.

The second high-impact risk is weaker-than-expected validation of the DCU line. I assign a medium probability and high impact. Hygon's current valuation clearly contains AI optionality, but the company has not publicly disclosed a CPU/DCU split, so outsiders cannot confirm how much revenue and profit the AI business has already contributed. At the same time, although the domestic AI accelerator server market is expanding and the combined share of local vendors is rising, this means more competitors, not automatic victory for Hygon. The indicators to watch are whether later financial reports or investor communications include more evidence on AI computing products, ecosystem adaptation, customer landing, and margin changes. If this information does not appear for a long time, the market will start doubting the basis of the high valuation.

The third risk is the governance discount from related-party transactions and customer concentration. I assign a medium probability and high impact. In 2025, Hygon sold 8.140 billion yuan of goods to "Company A and other companies under its control," and in 2026 expected total recurring related-party transactions to reach 13.949 billion yuan. At the same time, there are cross-appointments between the board and Sugon. This structure can become a channel advantage in an upcycle and a valuation discount in a downturn or during a sensitive regulatory period. The indicators to watch are the share of related-party sales, whether expected limits continue to be increased, and whether the company discloses clearer pricing mechanisms and boundaries. Once the market begins questioning independence, stocks of this kind usually suffer valuation compression first, not a slow decline.

The fourth risk is high-valuation compression. I assign a high probability and high impact. Hygon's current static P/E, P/S, and P/B are all extremely high. This means that as soon as growth moves from "very fast" to "still fast but not stunning," valuation may adjust before earnings do. The sharp pullback in global high-valuation semiconductor names in early June already demonstrated this fragility. The most direct observation indicators are whether revenue growth falls below 30%, whether gross margin can hold around 55%, and whether the market remains willing to pay high multiples for the domestic AI hardware narrative. Hygon's biggest risk is sometimes simply that the market decides it does not need to pay this much, rather than that performance deteriorates.

The fifth risk is that profit quality is obscured by share-based payment and working capital. Probability is medium, and impact is medium-high. In Q1 2026, net profit attributable to shareholders excluding share-based payment was 923 million yuan, about 34% higher than the statutory 687 million yuan. Operating cash flow also fluctuated heavily because of stocking and contract-prepayment timing. Looking only at net profit can lead to excessive optimism on operating quality; looking only at cash flow can understate the nature of front-loaded investment during rapid expansion. The indicators to watch are OCF/net profit, the gap between profit excluding share-based payment and statutory profit, and whether inventory and prepayments keep growing materially faster than revenue over a long period.

11. Catalysts and Tracking Indicators

Positive catalysts first come from financial results continuing to beat expectations. If revenue in the remaining quarters of 2026 still grows around 50% while gross margin stops falling, the market will be more willing to believe high growth is not just volume grabbing but also product-structure improvement. The second positive catalyst is improved transparency around the AI business. Even without full segment disclosure, clearer proof of revenue contribution and customer landing for AI computing products would give valuation a steadier support point. The third catalyst may still come from "soft integration" with Sugon and other ecosystem partners: using joint solutions, joint customer projects, and industry delivery cases to realize ecosystem synergy, rather than attempting another major restructuring.

Negative catalysts are also clear. First, revenue growth falls too quickly while the company still refuses to split CPU and DCU. Second, gross margin continues downward, indicating competition or cost pressure above market expectations. Third, operating cash flow remains weaker than profit while inventory and prepayments keep building rapidly. Fourth, export-control rules tighten again enough to affect product pace. Fifth, related-party transaction estimates continue expanding, reinforcing market concerns over customer concentration and governance discount.

The table below is the tracking dashboard I think most worth watching. It keeps only hard indicators that can change the bull-bear judgment and does not seek completeness.

Data sources: Hygon Information 2025 annual report and 2026 first-quarter report; some figures are calculated from disclosed data.

The logic behind these indicators is simple. Revenue growth and gross margin must be read together to distinguish growth by volume from growth by quality. OCF/profit, inventory, and prepayments must be read together to judge whether growth is overdrawing cash. Share-based payment impact determines whether statutory profit represents true operating profitability. Related-party sales concentration determines whether Hygon's synergy is strengthening the moat or accumulating a governance discount. Finally, the share price must be read together with the valuation range, or even a good company can be bought into a bad investment.

12. Horizontal and Longitudinal Synthesis

The abilities Hygon has truly proved on its path to today are threefold. First, it has proved that a CPU route that began with external licensing can be turned into a scaled high-end processor business in the Chinese market. Second, it has proved that after being placed on the Entity List in 2019, the organization could continue iterating rather than shutting down after external support was interrupted. Third, it has proved that a chip-design company does not have to sell chips alone. Binding with complete systems, servers, industry projects, and ecosystem partners can instead create a deeper commercial moat.

But Hygon's past success is a composite result, not something contributed solely by technical advantage: it includes the dividend of the times, policy demand, management execution, the historical opportunity created by the AMD licensing start, and the channel and delivery capability brought by the Sugon system. If it is understood only as a "domestic CPU technology hero" narrative, the company is seen too narrowly. If it is understood only as a policy-supported name, its real productization and commercialization capability is underestimated. Hygon's true scarcity lies in putting together the direction policy wants, the path customers are willing to buy, and the story capital markets like to hear.

These success factors still exist today, but their strength distribution has changed. The certainty of the CPU line remains, and it is still Hygon's steadiest value base. The synergy with Sugon and industry channels remains, and the scale of related-party transactions is evidence. Policy and domestic-substitution logic remain as well. The biggest change is that the market is no longer satisfied with this certainty and is demanding that Hygon turn DCU into a true second pole of revenue and profit. The problem is that Hygon has not yet completed that proof with public segment data. The market's willingness to keep paying a high price does not mean the evidence is complete.

Horizontally, Hygon's real advantages over domestic CPU peers are lower migration friction, stronger system synergy, and earlier profit realization. Relative to global x86 leaders, it does not have a global ecosystem and works only under specific Chinese constraints. Relative to NVIDIA, it certainly does not have software-stack dominance and can only compete for budgets through domestic substitution and availability. Some of its weaknesses are temporary, such as insufficient disclosure of AI revenue split. Others are more structural, such as the governance discount caused by related-party transactions and channel binding, and the permanent difficulty of replicating a global developer flywheel like NVIDIA's in DCU.

The current valuation rewards the future more than the past. The market is pricing a string of optimistic premises: a stable CPU base, a high probability of DCU success, continued ecosystem integration, and sustained scarcity in domestic AI hardware. It is not pricing the 2.5 billion yuan profit Hygon earned in 2025. I think the market's easiest misjudgment is overestimating the realization speed of the still-undisclosed DCU business, rather than underestimating Hygon's CPU. Hygon can be an improving company, but its share price is not necessarily an improving entry point.

The most important variable over the next 1 year is whether the growth structure can be verified. The most important variable over the next 3 years is whether DCU can move from "products, R&D, and narrative" to "sustained revenue, visible gross margin, and customer inertia." The most important variable over the next 5 years is whether it can run a stable iteration path for both x86 CPUs and AI accelerators under continuing export controls. When would Hygon become a better investment? When two conditions appear together: the price returns clearly to a margin-of-safety range, and the company provides more public evidence on the AI business structure. Conversely, if gross margin keeps falling, cash conversion remains weak, and AI revenue transparency fails to improve, the investment judgment should be rewritten even if revenue is still growing.

10.1 Bull and Bear Cases

Bull case:

• The company has already turned domestic high-end CPUs into a real business with more than 10 billion yuan of revenue, not just a policy concept.

• After being placed on the Entity List in 2019, it still advanced product iteration, showing organizational capability stronger than the stereotype of a licensed start.

• Deep synergy with complete-system and system partners makes it easier for Hygon to enter large projects and scaled procurement in key industries.

• Q1 2026 revenue continued to grow rapidly, showing demand momentum did not quickly fade after 2025.

Bear case:

• The annual report does not disclose CPU and DCU revenue or gross-margin splits, so the market cannot verify the realization of the AI narrative.

• Current valuation is extremely high, with static P/E and P/S already above many more mature global semiconductor leaders.

• Gross margin declined by 5.92 percentage points in 2025, showing volume growth did not automatically release profit leverage.

• Related-party transactions and customer concentration are high; synergy and governance discount are two sides of the same coin.

10.2 Pre-mortem

If this investment loses 50% three years from now, I think the most likely first scenario is this: in 2027, domestic AI accelerator competition enters a harder stage of delivery and ecosystem execution. Huawei Ascend, Cambricon, and other local vendors take more training and inference budgets from key customers. Hygon's DCU line fails to form a sufficiently clear revenue contribution. To retain projects, the company has to concede margins, gross margin falls from 57.78% in 2025 to below 50%, revenue growth slips below 20%, and the market stops valuing it as a "CPU plus AI platform" and instead values it as a "high-growth CPU design company." If P/S compresses from above 40x to about 20x, a halving of the share price would not be extreme. The real-world basis for this scenario is that the AI server market already has many domestic competitors, while Hygon has not yet disclosed independent financial realization for DCU.

The second scenario is this: from 2026 to 2028, U.S. rules for advanced computing and the chip supply chain continue to adjust. Hygon does not shut down, but product pace slows, prepayments and inventory keep rising, and operating cash flow remains weaker than profit for a long period. At the same time, the procurement pace of related channel customers slows, and the market begins questioning both supply-chain sustainability and revenue quality. Valuation falls from an extremely high range back to a more ordinary domestic semiconductor level. The frightening part of this scenario is that it only requires the market to reclassify Hygon from a "scarce platform" into a "restricted but decent domestic chip company." It does not require the company to lose money.

10.3 Final Research Conclusion

Hygon Information is a domestic high-end processor company that has already crossed the two thresholds of survival and profitability. The CPU business has proved that it can commercialize, scale, and continue advancing under external restrictions. This is Hygon's hardest base. What truly determines whether the share price can continue rising is whether the DCU business, the part for which the market is most willing to pay a premium, can be verified more clearly in public disclosures, not whether CPUs will continue to sell. On fundamentals, Hygon certainly deserves serious research. On entry point, today's problem is how much future has already been pulled into the price, not whether the company is good enough.

What I worry about most is not single-quarter volatility but three things happening together: AI business realization slower than expected, gross margin continuing to fall, and cash flow staying weaker than profit. Any one of these alone may not overturn Hygon's long-term logic. But if they overlap during a high-valuation phase, the share price will re-rate faster than the financial statements. Under what conditions would I change my view? If the company begins to prove DCU revenue and customer landing more transparently, while stabilizing cash conversion as it maintains high growth, I would be willing to raise valuation tolerance. If gross margin and cash flow keep weakening while the market still assigns ultra-high multiples, I would more firmly classify it as a good company at a bad price.

【Company Profile Scores】

• Fundamental quality: Medium

• Growth: High

• Moat: Medium

• Financial stability: Strong

• Management credibility: Medium

• Valuation attractiveness: Low

• Risk level: High

• Suitable investor type: Not suitable for ordinary investors

【Investment Rating】

• Rating: Hold

• One-sentence investment thesis: CPU profit provides the floor and DCU provides upside optionality, but the current valuation has already priced in substantial optimism.

• Three price signals: Ideal buy price: a range corresponding to at least a 20% margin of safety below the bearish implied value

• Holdable price: within +/-15% of the base implied value

• Clearly overvalued price: at least 10% above the bullish implied value

• Current price category: Holdable

• Whether it is worth waiting for a better price: Yes. For new capital, the more reasonable action is to wait for the price to return below 176 yuan and see cash flow and gross margin stabilize at the same time before discussing a buy. The opportunity cost of waiting is that if the company discloses stronger AI realization evidence earlier than expected, the share price may not return to a deep discount.

• Target holding period: 1-3 years

• Expected annualized return: bearish -31% to -24%; base -12% to +0%; bullish +13% to +26%

• Maximum loss risk: 50%-60%; triggering conditions refer to the two pre-mortem scenarios above. The core is weaker-than-expected DCU realization combined with valuation compression, or supply-chain rule changes slowing iteration and shipments.

• Signals triggering reassessment: Revenue growth falls below 20% for two consecutive quarters

• Gross margin is below 52% for two consecutive quarters

• OCF/net profit attributable to shareholders is below 0.5x for two consecutive reporting periods

• Related-party sales concentration continues to rise significantly, or related-party transaction estimates are raised substantially again

• Export-control changes affect product iteration or supply-chain arrangements

【Ideal/Fair Buy Price】160-176 CNY Basis: the result of applying at least a 20% margin of safety to the bearish implied value of 200-220 yuan. Only back in this range does Hygon move from a company worth researching to a price worth risking.

【Valuation Range】

• current: 288.86 (as of the 2026-06-11 close)

• bear (conservative, ideal buy zone): [160, 176]

• base (reasonable, acceptable holding zone): [220, 335]

• bull (optimistic, above clear overvaluation line): [360, 405]

13. Key Data Table

Data sources: Hygon Information 2025 annual report and 2026 first-quarter report; market capitalization calculated from the 2026-06-11 closing price.

14. Research Uncertainties

• The company does not disclose revenue, gross margin, or order splits for CPU and DCU. This is the largest blind spot in this report. Many judgments about the second growth curve can only be inferred indirectly, not directly verified.

• Related-party transaction announcements exempt disclosure for several related parties. Public materials only show anonymous labels such as "Company A" and "Company D," so each counterparty cannot be fully looked through.

• Export controls, foundry and packaging/testing paths, and availability of advanced process nodes are highly sensitive information, and public disclosure is naturally insufficient. Research on supply-chain sustainability can therefore only reach directional conclusions.

• The primary materials obtained are sufficient to verify 2025 and 2026Q1, but segment quality and capex details over a longer history are limited. Therefore, the 5-10 year cash-flow look-through must be treated conservatively.

• In horizontal valuation, domestic AI accelerator companies have insufficient business overlap with Hygon, and same-day available metrics are not fully consistent. This report therefore emphasizes the comparison framework rather than pretending to provide precise decimal-point accuracy.

15. Reference Sources

• Hygon Information 2025 annual report.

• Hygon Information 2026 first-quarter report.

• Hygon Information initial public offering prospectus for listing on the STAR Market.

• Hygon Information announcement on expected recurring related-party transactions.

• Hygon Information proposal summary for share-exchange absorption merger with Sugon and subsequent progress and termination announcements.

• Public materials from the U.S. Department of Commerce Bureau of Industry and Security and the Federal Register on the Entity List, AI diffusion rules, and chip export controls.

• Reuters reports on China's AI accelerator server market share, Hygon's listing, and global chip valuation volatility.

• Public market pages such as Google Finance, Yahoo Finance, and Investing, used for prices, ranges, and horizontal valuation references around the research base date.

16. Other Securities Mentioned in the Report

• 603019.SHG - Hygon Information's largest shareholder and key complete-system ecosystem partner; related-party transactions and the failed absorption merger both centered on it

• NVDA.US - The global benchmark for AI accelerators and software ecosystems, used to measure the true competitive ceiling for Hygon's DCU

• AMD.US - An important reference for Hygon's CPU technology origin and the source of the later debate over independent evolution

• INTC.US - The long-term definer of the global x86 server ecosystem, used to understand the commercial meaning of Hygon's compatible route

• 688047.SHG - A domestic listed CPU peer, representing another domestic route that emphasizes proprietary ISA more strongly

• 688256.SHG - One of the most typical A-share AI chip narrative names, used to understand the AI scarcity premium Hygon receives

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