Moore Threads: H1 2026 Revenue Grew 147.4% to 1.736 Billion CNY While Operating Cash Flow Fell to Negative 2.169 Billion, and 357.85 CNY Sits 175% Above the 130 CNY Ideal-Buy Ceiling

The stock went through an equally important transition. It listed on 2025-12-05 at CNY 114.28, opened at CNY 650, up 468.78%, and closed its first session at CNY 600.50. By the fifth trading day it had traded as high as CNY 941.08. That was scarcity pricing for the first major pure-play domestic full-function GPU listing, not a valuation grounded in contemporary earnings. The current CNY 357.85 is 62.0% below that post-listing high but still 213% above the IPO price. Sitting at a post-listing low says almost nothing by itself about value.

The September collapse started with a specific mechanical catalyst. On 2026-09-07, 25,774,510 IPO institutional-placement shares became tradable, 5.48% of total shares. Before that date, only about 30.23 million shares were freely tradable. The unlock lifted free float to approximately 56.00 million shares, an 85% one-day increase. The stock hit its 20% STAR Market limit and closed around CNY 415.48–415.49, taking market capitalization below CNY 200 billion. The newly tradable institutions had subscribed at CNY 114.28, so even the limit-down price represented approximately 3.64 times their cost, or a 264% gain. Concentrated selling was economically rational.

The fall did not end with that mechanical event. From the September 7 close to September 11, Moore Threads lost another 13.9%, ending at CNY 357.85 while the broader AI-compute and STAR technology complex was also weakening. From the prior house report’s CNY 616.87 reference to September 11, the decline is now about 42.0%, materially larger than the roughly one-third decline visible earlier in the refresh window. Fundamentals improved over that interval. The market price nevertheless fell. The evidence points to a combination of unlock-driven supply and a genuine compression of the scarcity/AI-compute valuation premium, rather than deteriorating H1 operations.

The next supply event is much larger. Approximately 185.91 million shares are scheduled to emerge from 12-month IPO lock-ups on 2026-12-07, roughly 39.55% of the total share count. If fully tradeable, free float would rise from about 56.0 million to 241.9 million shares, or from 11.9% to 51.5% of total shares. That is a 4.32-fold free-float base versus today. This is the single most predictable capital-markets variable over the next quarter.

A planned H-share listing adds another layer. Shareholders approved the Hong Kong Main Board plan on 2026-08-28. Before any greenshoe, new H-shares may be no more than 10% of post-issue share capital, implying a maximum initial issue of about 52.23 million new shares against today’s 470.03 million A shares. The plan also permits a greenshoe of up to 15% of that H-share issue; full exercise would take new H shares to roughly 60.06 million and dilute pre-existing shareholders by about 11.33%. The authorization lasts 24 months from shareholder approval. The latest official issuer record found for this research does not announce a completed CSRC overseas-listing filing or HKEX approval; press reports that the company had “confidentially filed” appeared in early September, but that detail remains unconfirmed by a company announcement at the base date.

The valuation argument has changed substantially. At CNY 616.87, 470.03 million shares implied a market capitalization near CNY 290 billion; divided by 2025 sales of CNY 1.506 billion (15.06 亿元), the previous report’s approximately 192.6 times P/S was mathematically correct. At CNY 357.85, the same stale 2025 denominator produces 111.7 times P/S. More appropriately, H2 2025 revenue was approximately CNY 804 million (8.04 亿元), so trailing-twelve-month revenue through June 2026 is about CNY 2.541 billion (25.41 亿元), producing 66.2 times TTM P/S. Annualizing H1 produces CNY 3.473 billion (34.73 亿元) and 48.4 times annualized-sales P/S.

The old relative-valuation argument that Moore Threads is plainly more expensive than Cambricon no longer survives on a current TTM-sales basis. Cambricon is valued around CNY 653 billion against roughly CNY 9.61 billion of TTM revenue, about 68 times TTM P/S, almost the same multiple as Moore Threads’ 66 times. The important distinction has moved from the multiple to the business underneath it: Cambricon is already profitable, with H1 2026 revenue of CNY 5.996 billion (59.96 亿元) and attributable net profit of CNY 2.311 billion (23.11 亿元), while Moore Threads still has negative owner cash flow and materially smaller scale.

The valuation reservation itself remains. A 66 times trailing-sales multiple for a company with negative operating free cash flow, uncertain advanced-node supply access, concentrated/lumpy cluster contracts and another 39.55% of total shares approaching unlock is still demanding. What has changed is that the reservation is no longer “nothing in the fundamentals has caught up.” A meaningful portion has caught up. The question is now whether revenue can compound fast enough, convert inventory into cash and turn the MUSA/KUAE stack into recurring customer adoption before the valuation and supply structure normalize.

The qualitative portrait is a company in transition: technologically a full-function GPU developer, commercially an AI-cluster supplier, financially approaching operating breakeven in accounting terms while consuming substantial cash, and in capital markets moving from extreme post-IPO scarcity pricing toward a larger-float, potentially A+H valuation regime.

Company vertical history, financial review and price narrative

From Nvidia experience to a domestic full-function GPU stack

Moore Threads was established in Beijing in 2020. Founder, chairman and general manager Zhang Jianzhong spent years at Nvidia and served in senior China leadership roles. That background helps explain the decision to pursue a general-purpose/full-function GPU instead of a narrow AI accelerator: the company sought a unified architecture and software environment spanning graphics and accelerated computing, broadly echoing the architectural breadth that made GPUs useful beyond graphics. The IPO prospectus describes MUSA as the foundation for products across AI computing, graphics, video and scientific workloads.

The strategic choice was unusually capital intensive. Building a full-function GPU means funding silicon design, drivers, compilers, developer tools and application compatibility at the same time. Moore Threads entered a field in which Nvidia’s CUDA ecosystem represented the global usability benchmark, while domestic peers including Cambricon, MetaX, Biren and Enflame generally entered from more specialized AI-accelerator or data-center directions. That breadth created optionality, but it also meant that R&D spending had to arrive years before scale revenue.

The first phase, roughly 2020–2022, went to architecture and product validation. Revenue was only about CNY 46 million (0.46 亿元) in 2022, while the net loss was approximately CNY 1.84 billion (18.4 亿元). The company was financing an engineering organization rather than harvesting an installed base.

The second phase, 2023–2024, brought both commercialization and the company’s defining external constraint. Revenue rose to roughly CNY 124 million (1.24 亿元) in 2023 and CNY 438 million (4.38 亿元) in 2024, but losses remained around CNY 1.67 billion (16.7 亿元) and CNY 1.49 billion (14.9 亿元), respectively. In October 2023 the U.S. Commerce Department added Moore Threads-related entities to the Entity List under the advanced-computing control package. BIS explicitly stated that Footnote 4 treatment requires foundries producing covered chips for those listed entities to obtain a BIS license before sending them such chips when the foreign-direct-product rule applies.

The third phase began in 2025, when commercialization shifted from cards and products toward larger cloud-compute systems. Full-year revenue reached CNY 1.506 billion (15.06 亿元), up roughly 243%, while the company remained loss-making. That revenue base, plus the strategic importance attached to domestic GPU supply, allowed Moore Threads to enter the STAR Market despite still being unprofitable.

The company chose STAR Market listing standard 2.1.2(2): expected market capitalization of at least CNY 1.5 billion (15 亿元), latest-year revenue of at least CNY 200 million (2 亿元), and three-year cumulative R&D expenditure equal to at least 15% of cumulative revenue. Its 2024 revenue was CNY 438.46 million (4.38 亿元), while 2022–2024 cumulative R&D was CNY 3.81 billion (38.10 亿元), approximately 626% of cumulative revenue. This was a revenue-plus-R&D listing standard, not an earnings standard.

At the CNY 114.28 IPO price, post-issue market capitalization was about CNY 53.72 billion (537.15 亿元), already more than 120 times 2024 sales. The offer comprised 70 million new shares and raised gross proceeds of about CNY 8.00 billion (79.996 亿元); post-IPO share count became 470,028,217. Only 29,382,386 shares, 6.25% of the company, were freely tradable on listing day.

That small float explains the extraordinary first week. On December 5, the stock opened at CNY 650 and closed at CNY 600.50. Within five trading days it reached CNY 941.08, implying a market capitalization around CNY 442 billion even though the company had generated only CNY 1.506 billion (15.06 亿元) of 2025 revenue. Price discovery was dominated by scarcity and domestic-GPU thematic demand.

The financial turn

† H1 2025 gross profit and margin are reconstructed from the disclosed H1 2026 gross-profit growth rate and current gross profit.

Source: company interim reporting and prospectus/annual disclosures.

The table exposes the central tension. Operating leverage is visible in the income statement: CNY 1.03 billion (10.34 亿元) of incremental year-on-year revenue produced roughly CNY 504 million (5.04 亿元) more gross profit and a CNY 166 million (1.66 亿元) improvement in deducted net earnings. Yet cash conversion deteriorated because production and procurement ran ahead of revenue recognition.

Sales expense increased 121% to CNY 158.19 million (1.58 亿元); administration was nearly flat at CNY 150.94 million (1.51 亿元); R&D increased 38.16% to CNY 769.15 million (7.69 亿元). That cost structure shows real operating leverage: administration is no longer scaling with revenue and R&D is rising far more slowly than sales. It also shows why normalized profitability remains some distance away. R&D alone absorbed 44.3% of H1 revenue.

The unlock map and the true float

The IPO created a highly unusual distinction between total capitalization and tradeable capitalization. The company had 470.03 million shares outstanding from listing, but only a tiny fraction could actually set the market price.

§ The IPO documents impose several different locks: 12 months from listing, 24 months for the sponsor co-investment and 36 months either from listing or from the date particular pre-IPO shares were acquired. Public market calendars show multiple late-2027/early-2028 dates rather than one single release. The approximately 82.682 million residual above is an inference from current locked shares less the dominant December 2026 and December 2028 buckets; future issuer announcements will determine actual tradable quantities on each date.

The December 2026 release matters more than September. September increased the free float by 85%; December can increase it another 332% from the post-September level. At the September 11 price, today’s 56 million-share free float has a market value of only about CNY 20.04 billion (200.40 亿元), while total capitalization is CNY 168.20 billion (1,682 亿元). That gap is why relatively modest marginal flows can still produce violent price moves.

The original shareholders are not one homogeneous seller group. The listing announcement identifies major 12-month holders including several institutional venture investors, while Zhang Jianzhong, his aligned entities and certain employee platforms carry longer commitments. The sponsor-related CITIC Securities Investment holding of 1.4 million shares has a 24-month lock; strategic placements other than that sponsor co-investment generally carry 12 months.

Price attribution

The H1 report was published on August 10 after a July 17 voluntary results preview, so I do not assign a clean one-day “earnings reaction” to August 10. More usefully, the improved earnings did not prevent a subsequent valuation contraction. By September 4 the stock was CNY 519.35; after the unlock and further sector weakness it reached CNY 357.85.

The September decline began as a supply event and evolved into a broader re-rating. Calling the entire fall “sentiment” misses the contractual unlock. Calling the entire fall temporary also goes too far: the additional 13.9% decline after the limit-down, amid a weaker domestic AI-compute complex, indicates that investors were also resetting the multiple appropriate to a larger-float asset.

The unprofitable-company suffix

Moore Threads listed while unprofitable and entered the STAR Market’s 科创成长层, or “STAR Growth Layer.” Its current market name carries the “-U” designation for an issuer that was unprofitable at listing. The suffix still appears in market quotations as of the research date. Under the SSE’s 2025 special-identification rules, “U” is removed when such an issuer first achieves the prescribed profitability condition or exits the Growth Layer. Cambricon’s removal of “U” after its profitable 2025 annual report provides a recent practical example.

H1 headline profitability does not by itself remove the suffix. Moore Threads’ H1 attributable result was still a small loss and its deducted result remained negative. More importantly, the listing regime does not impose a simple “must be profitable by a fixed year after IPO” covenant. The company’s immediate structural issue is not an imminent profit-deadline delisting event; it is whether a business that remains unprofitable on a deducted basis can justify its capital requirements and valuation as the special scarcity associated with a tiny float fades.

Business model, moat, industry structure and geopolitics

What the company actually sells now

The H1 segment economics are unusually concentrated.

All H1 revenue was domestic. Direct sales generated CNY 1.172 billion (11.72 亿元), about 67.5%; distributors generated CNY 563.92 million (5.64 亿元), about 32.5%.

The cloud category includes GPUs, boards, integrated machines and KUAE cluster systems. The accounting policy matters. Revenue is recognized when the promised product or system is transferred and the performance obligation is satisfied, generally at a point in time. Where Moore Threads controls goods before transfer, it acts as principal and records the gross consideration; otherwise it records the agency economics. That policy makes cluster revenue much larger in absolute yuan than pure silicon revenue but also imports third-party hardware cost into cost of goods sold.

The CNY 660 million (6.60 亿元) KUAE order disclosed in March and delivered into H1 is the clearest evidence of commercial scale. It is also the clearest evidence of lumpiness: one contract equaled about 38% of the half-year’s entire company revenue. At June 30, disclosed unsatisfied performance obligations were only CNY 87.52 million (0.88 亿元), most expected to be recognized in 2026. The disclosed backlog offered very little forward coverage relative to the run rate.

Historical customer concentration reinforces that point. In H1 2025, the top five customers represented 98.28% of revenue and the largest customer alone represented 56.63%. The H1 2026 report does not provide an equivalent named top-five table that allows a like-for-like update, so I would not claim that concentration has already diversified. The CNY 660 million contract shows that large-ticket customer dependence is still economically material.

Operating leverage and the cash machine

Moore Threads has the cost structure of a semiconductor platform being commercialized at speed: large R&D and personnel costs are substantially fixed in the short term; foundry, packaging, memory, networking and cluster components vary with shipments; inventory and customer-credit needs can consume far more cash than the income statement suggests. H1 revenue grew 147%, while R&D rose 38% and administration declined slightly. That is positive operating leverage. Cost of sales rose much faster than revenue, which is why gross margin compressed.

The balance sheet is still well funded in gross terms. At June 30, cash was CNY 6.492 billion (64.92 亿元), supplemented by CNY 1.440 billion (14.40 亿元) of trading financial assets. Short-term borrowings were approximately CNY 443 million (4.43 亿元) and long-term borrowings CNY 2.825 billion (28.25 亿元). That leaves about CNY 4.66 billion (46.64 亿元) of net liquid financial resources if trading investments are included.

At the full H1 cash-consumption rate, that cushion is smaller than the gross cash number looks. Operating cash flow of negative CNY 2.169 billion (21.69 亿元) plus approximately CNY 769 million (7.69 亿元) of long-term-asset expenditure gives six-month free-cash consumption of roughly CNY 2.94 billion (29.38 亿元). Annualized mechanically, cash plus trading assets would cover about 1.35 years. Using operating cash flow alone gives about 1.83 years. Those are stress-run-rate calculations, not forecasts, because CNY 2.254 billion (22.54 亿元) of H1 operating cash consumption came from the inventory build and may not repeat at the same pace.

This is the financial logic behind the H-share proposal. The A-share IPO provided CNY 8 billion (79.996 亿元) of gross funding, but a company simultaneously funding new architectures, carrying large semiconductor inventories and shipping integrated clusters can absorb large amounts of working capital. An H-share raise extends that financing runway and diversifies funding. It also dilutes A-share holders and creates an external Hong Kong price reference, so it should not be valued as “free cash.”

What is real in the moat

The strongest candidate moat is the integrated MUSA software-and-hardware architecture. GPU competition is not decided only by theoretical arithmetic throughput. Customers need compilers, libraries, drivers, framework compatibility, cluster communication and debugging tools. Moore Threads’ aim of one unified architecture from graphics through AI gives it more software surface area than a narrow accelerator. The KUAE commercialization indicates that at least some customers are willing to deploy the stack at cluster scale.

The second candidate moat is integration know-how. Selling a large KUAE cluster involves networking, scheduling, system engineering and model deployment, not merely shipping chips. A successful installed cluster can generate switching friction because customers have already tuned software and operations around the stack. The drop in disclosed backlog means there is not yet enough public evidence to call that stickiness proven, but the transition toward clusters is economically more significant than a simple rise in GPU-card shipments.

The third is talent and organizational learning. Zhang Jianzhong’s Nvidia background and the scale of R&D investment gave Moore Threads a credible engineering starting point. H1 R&D still consumed CNY 769 million (7.69 亿元). The company has shown it can take multiple generations from architecture to commercial systems in a short period.

The moat is promising rather than proven. The company is only six years old, the commercial revenue surge is roughly two years old, and nearly all current sales come from the cloud-compute line. A full-function GPU architecture has strategic breadth, but customers are currently paying Moore Threads primarily for AI compute. The graphics and broad general-purpose capabilities have not yet produced a diversified profit pool large enough to prove that breadth translates into superior economics.

Entity List: the existential variable

BIS added Moore Threads-related entities to the Entity List in October 2023 in its advanced-computing-control package. Footnote 4 extends controls to certain foreign-produced items made using U.S. technology or software. BIS stated explicitly that foundries producing covered chips for such listed parties require a BIS license before shipment when the foreign-direct-product rule applies.

No responsible analyst can assume normal access to leading global foundry capacity. Reuters’ reporting on Moore Threads and MetaX similarly described U.S. restrictions as blocking access to leading global foundries such as TSMC. Moore Threads’ own risk disclosure says restrictions affect access to raw materials, U.S.-related technology/IP and R&D tools and can force supplier substitution at higher cost.

The precise current foundry, process-node allocation and committed wafer capacity for Moore Threads’ latest mass-production products are not disclosed in enough detail in the public H1 filing to support a quantified wafer forecast. I therefore do not assume an undisclosed TSMC allocation or a guaranteed domestic advanced-node allocation. The observable evidence is commercial output plus inventory: Moore Threads has clearly secured enough supply to generate CNY 1.736 billion (17.36 亿元) of H1 sales and accumulate CNY 3.550 billion (35.50 亿元) of inventory, but the composition and replenishment rate of that inventory are not public.

A useful sensitivity starts with a research base case of CNY 7.0 billion (70 亿元) 2027 revenue at 53% gross margin, producing CNY 3.71 billion (37.1 亿元) gross profit. In a tighter-control scenario where advanced silicon, memory, packaging or tool restrictions reduce deliverable 2027 revenue by 30% to CNY 4.9 billion (49 亿元) and mix/procurement pressure takes gross margin to 45%, gross profit falls to roughly CNY 2.21 billion (22.1 亿元), about CNY 1.50 billion (15 亿元) below the base assumption. That would almost certainly delay durable operating profitability.

Conversely, if domestic supply localization or regulatory loosening allows 30% more shipments than the base assumption, CNY 9.1 billion (91 亿元) of revenue at 55% gross margin would produce approximately CNY 5.01 billion (50.1 亿元) of gross profit, roughly CNY 1.30 billion (13 亿元) above the base. These are sensitivity cases, not management guidance. The asymmetry is important: policy can simultaneously create domestic demand for Moore Threads and constrain the manufacturing technology needed to supply it.

Horizontal competitor analysis

The relevant competitive field has changed since Moore Threads’ IPO. The scarcity value of being a public domestic GPU pure play is disappearing because investors can now compare several companies directly. The right framework is not “which one is China’s Nvidia?” but “which commercial model is becoming durable fastest?”

Cambricon became the profitable AI-accelerator incumbent

Cambricon is still the closest A-share valuation anchor even though its technical positioning differs. It focuses primarily on intelligent-computing accelerators rather than Moore Threads’ broader full-function-GPU mission. Its commercial breakthrough arrived earlier: H1 2026 revenue reached CNY 5.996 billion (59.96 亿元), and attributable net profit reached CNY 2.311 billion (23.11 亿元). Current data put TTM revenue around CNY 9.61 billion (96.1 亿元) and market capitalization around CNY 653 billion.

Moore price and market cap are as of 2026-09-11; Cambricon market data were current around the research date.

This comparison changes the prior thesis. Moore no longer commands a visibly higher sales multiple than Cambricon after updating both price and revenue. It instead receives roughly the same sales multiple despite being about one-quarter Cambricon’s H1 revenue, still consuming cash and still reporting a deducted loss. A parity multiple embeds an assumption that Moore’s faster growth and full-function-GPU optionality will compensate for materially weaker present earnings quality.

MetaX became the closest full-function data-center challenger

MetaX, or 沐曦股份, is structurally closer in high-performance GPU commercialization. H1 2026 revenue was approximately CNY 1.324 billion (13.24 亿元), 44.7% year-on-year growth. Headline attributable profit was approximately CNY 612 million (6.12 亿元), but around CNY 661 million (6.61 亿元) of non-recurring gains meant deducted profit remained a CNY 48.9 million (4,886 万元) loss for the half. Its second quarter reportedly reached positive deducted earnings, making its path toward core breakeven an important comparison for Moore.

Customers choose MetaX for a focused domestic high-performance GPU alternative and an increasingly mature data-center software stack. The competitive threat to Moore is direct: both companies want the same domestic training/inference deployments, and both need to reduce migration friction from incumbent CUDA-centric infrastructure. Moore’s advantage is broader GPU functionality and larger H1 revenue; MetaX’s near-core profitability raises the bar for how quickly that breadth needs to convert to cash.

Biren is a scale competitor without Moore’s broad positioning

Biren’s H1 revenue was about CNY 1.236 billion (12.36 亿元), putting it in the same commercial scale band as Moore and MetaX. Biren’s original design emphasis has been high-performance general-purpose computation and AI workloads rather than the entire graphics-to-compute spectrum. Its existence reduces the value of “domestic alternative” as a moat in itself. Customers increasingly have several local accelerator choices and can compare delivered cluster economics rather than merely asking whether a Chinese GPU exists.

Enflame illustrates both the opportunity and the concentration risk

Enflame, or 燧原科技, listed on STAR on 2026-09-11 and closed its debut with a market value around CNY 171 billion, remarkably close to Moore Threads’ CNY 168 billion market cap that same day. Enflame generated only about CNY 990 million (9.90 亿元) of 2025 revenue, and Tencent is both a major shareholder and customer. The debut itself, up 179%, shows that investors are still willing to assign extraordinary option value to domestic AI accelerators. It also shows why peer multiples cannot be treated as proof that Moore is cheap: the entire group carries a substantial policy, scarcity and AI-capex premium.

Hygon and Nvidia belong in different comparison boxes

Hygon provides an adjacent domestic-compute reference because it has established CPU/DCU scale and profitability, but it is not a clean product-equivalent GPU comparison. Its relevance is what mature domestic-compute economics can look like after commercialization, rather than a like-for-like Moore valuation anchor.

Nvidia is the technological benchmark and software-ecosystem benchmark, not an appropriate valuation comparable. Nvidia operates at vastly different revenue, margin, foundry-access and ecosystem scale. Using Nvidia’s sales multiple to value Moore would reward Moore as if the hard part of creating a CUDA-like installed base had already been accomplished. It has not. Reuters nevertheless reports that U.S. restrictions have opened meaningful room for local Chinese accelerators, which is precisely why Moore’s opportunity can be large despite the technology gap.

The ecological niche is clear. Moore is a challenger selling a broad domestic GPU architecture through increasingly large AI-compute systems. Its profit pool comes directly from data-center accelerator vendors, including Cambricon, MetaX, Biren, Enflame and Huawei’s Ascend ecosystem. Moore gains relative strength if customers value one architecture across graphics and compute or if KUAE integration materially lowers deployment friction. It weakens if the Chinese market converges on specialized AI accelerators where broad graphics functionality carries cost but little willingness to pay.

Current fundamentals, capital markets and valuation

What the latest half says underneath the headline

The income statement is much closer to breakeven than it was a year ago, but three lines matter more than the CNY 11.56 million (1,156.31 万元) headline loss.

First, deducted net loss is still CNY 150.82 million (1.51 亿元). Government subsidies and financial-asset gains produced most of the difference between that figure and the headline result. The prior concern about earnings quality remains factually correct.

Second, gross margin fell more than 12 percentage points year on year to 56.95%. That is an economically reasonable outcome if larger integrated-cluster contracts contain more bought-in components, but it means revenue cannot simply be extrapolated with the old product-level gross margin. The valuation model below therefore does not assume a return to 69% gross margin.

Third, working capital exploded. Accounts receivable rose to CNY 880 million (8.80 亿元) from about CNY 434 million (4.34 亿元) at year-end, while inventory rose to CNY 3.550 billion (35.50 亿元). Prepayments were CNY 1.344 billion (13.44 亿元). That is the balance-sheet signature of a business buying ahead of deliveries. It can unwind constructively if clusters ship and customers pay; it can become a source of impairments if demand or supply architecture changes.

Cash-flow passthrough and owner earnings

A five-year cash-conversion ratio is not economically meaningful for Moore because both earnings and operating cash flow have been negative over most of its short history. The prospectus shows persistent operating cash outflows during the pre-IPO years, while net losses were CNY 1 billion-plus annually. In H1 2026 the absolute operating cash loss, CNY 2.169 billion (21.69 亿元), was roughly 188 times the tiny CNY 11.56 million (1,156.31 万元) accounting net loss. A positive CFO/net-income ratio produced by dividing one negative number by another would be misleading.

Maintenance capex is not separately disclosed. H1 spending on fixed assets, intangibles and other long-term assets was about CNY 769 million (7.69 亿元), while depreciation and amortization were a fraction of that. Using depreciation/amortization as a rough floor suggests maintenance capital of perhaps CNY 120–200 million (1.2–2.0 亿元), which makes most H1 capital spending likely growth-related. That split is a research assumption, not company guidance.

Even on that more forgiving owner-earnings treatment, there is no useful P/E. Operating cash flow is deeply negative before maintenance capex. On the stricter free-cash-flow definition, H1 owner cash generation is roughly negative CNY 2.94 billion (29.38 亿元). Headline P/E and owner-earnings P/E are both economically unusable; sales and long-run cash-flow conversion are the relevant valuation anchors.

Recomputing the sales multiple

§ EV subtracts approximately CNY 4.66 billion (46.64 亿元) of June net liquid financial resources from the September market capitalization; it does not credit any future H-share proceeds.

Against the prior report’s approximately 192.6 times FY2025 P/S, price alone has cut the stale-denominator multiple to 111.7 times. Updating the revenue denominator cuts it again to 66.2 times TTM. This is the double compression the refresh needed to capture.

The plain answer to the prior “too expensive” conclusion is: the original reason is obsolete, but the absolute conclusion remains defensible. Moore no longer looks uniquely expensive against Cambricon on sales. Sixty-six times TTM sales remains an extraordinary multiple for a company whose current free cash flow is deeply negative and whose core earnings are still below breakeven.

H-share arithmetic

The H-share plan was approved by shareholders on August 28 and can be executed within 24 months. The maximum base issue is 10% of post-issue shares. Solving that against the existing 470.03 million A shares gives up to about 52.23 million initial H shares. A full 15% greenshoe could lift issuance to about 60.06 million shares and total shares to roughly 530.09 million.

That is a maximum dilution of roughly 11.33% to the ownership percentage of today’s shareholders. The cash received would partly finance R&D, inventory and working capital, so I do not add the full proceeds one-for-one to per-share fair value. Hong Kong pricing also introduces a second market-clearing reference. If H shares price at a material discount to the A-share, arbitrage expectations and investor comparisons can pull the A-share multiple lower even without direct fungibility.

Absolute valuation scenarios

The valuation uses a one-year research horizon, FY2027 revenue and EV/Sales, then discounts the implied FY2027 enterprise values back to the base date. I assume the full potential H-share dilution in share count and give no free valuation credit for H-share cash. That is deliberately stricter than valuing the company on today’s 470.03 million shares.

This is valuation-scenario analysis within a research framework, not investment advice.

The multiples are intentionally below today’s 66 times TTM sales because no company can compound revenue into a mature cash-generating state while retaining a startup scarcity multiple indefinitely. Even the base case still grants Moore a 22–27 times forward-sales valuation in 2027. That remains an unusually large premium and requires both rapid revenue scaling and visible movement toward cash profitability.

A reverse-valuation check is revealing. Current enterprise value is roughly CNY 163.5 billion (1,635 亿元) after June net liquidity. At a mature 30 times free-cash-flow multiple, that value requires about CNY 5.45 billion (54.5 亿元) of normalized annual FCF before any discounting for the years needed to reach maturity. A 20% FCF margin would require approximately CNY 27.3 billion (273 亿元) of annual revenue; even a 25% FCF margin requires about CNY 21.8 billion (218 亿元). Current TTM revenue is CNY 2.54 billion (25.4 亿元). The market is therefore still pre-spending a very large part of the future.

Expectation gap

The next decisive earnings question is no longer whether revenue can grow by triple digits from a tiny base. H1 answered that. Investors need evidence that CNY 6–10 billion annual revenue can be reached without simultaneously consuming CNY 4–6 billion of annual cash through inventory, prepayments and capex.

Gross margin is the second expectation gap. A stable 50–55% margin on much larger cluster revenue would support the operating-leverage thesis. A move below 45–50% would suggest that system-level commercialization scales revenue faster than economic value.

The third expectation gap is order visibility. June remaining performance obligations of CNY 87.52 million (0.88 亿元) are too small to underwrite the next year. A new multi-hundred-million-yuan cluster contract would materially alter near-term revenue confidence. Another half-year of large recognized revenue with little disclosed backlog would confirm a structurally lumpy sales model.

Margin-of-safety recheck

Current price is 117% above the high end of the CNY 120–165 conservative fair-value range and almost 200% above its low end. Margin of safety against the conservative scenario is zero.

The most fragile base assumption is the CNY 6.5–7.5 billion (65–75 亿元) FY2027 revenue range. Reducing that assumption to 70% while holding the sales multiple constant cuts the base valuation roughly linearly to around CNY 172–235 per share. At CNY 357.85, that would leave considerable downside.

If operating earnings are flat for the next three years, Moore produces no dividend yield and no positive owner earnings from which a shareholder return can compound. The fundamental annual return would be approximately zero before any multiple compression. There is no margin of safety at this buy price.

Margin-of-safety sufficiency verdict: none.

Cross-synthesis, risk, catalysts and final research conclusion

The permanent-loss risks

The first risk is semiconductor supply, with medium-to-high probability and high impact. Footnote 4 Entity List treatment directly reaches foreign-produced chips when the relevant U.S.-technology conditions are met. A further tightening can interrupt node access, memory, packaging, EDA or IP flows. The observable indicators are inventory composition and growth, prepayments, timing of new-generation launches and any new BIS rule affecting Footnote 4 entities. The transmission path is physical: fewer deliverable systems lower revenue, emergency sourcing lowers gross margin, redesign extends R&D cycles, and the valuation multiple contracts because the future revenue runway shortens.

The second is contract concentration, high probability and high impact. One CNY 660 million (6.60 亿元) KUAE contract represented about 38% of H1 revenue, while historical customer concentration has been extreme. The indicator is the share of revenue represented by newly announced large contracts and disclosed remaining obligations. Cancellation or delayed acceptance of a single large cluster can shift half-year growth materially, which in turn matters greatly to a stock valued at dozens of times sales.

The third is working-capital conversion, medium probability and high impact. Inventory of CNY 3.550 billion (35.50 亿元) now exceeds two years of annualized H1 cost of sales in value terms. The indicator is inventory versus trailing cost of sales and the path of operating cash flow. If inventory converts into shipped clusters, H1 burn can reverse quickly. If product obsolescence or sanctions trap it, accounting losses and cash losses can converge again through impairments.

The fourth is capital-market supply, high probability and medium-to-high impact. Approximately 185.91 million shares, 39.55% of the company, are scheduled to leave 12-month locks in December. The event is known in advance; the uncertainty is how much holders sell. It cannot destroy the underlying business, but it can permanently reset the A-share valuation center by removing scarcity. The H-share transaction can reinforce the same process if international investors demand a lower sales multiple.

The fifth is competitive catch-up, medium-to-high probability and high impact. Cambricon has already crossed into substantial profitability, MetaX is close to core breakeven, Biren has reached billion-yuan half-year scale, and Enflame has entered public markets. The indicator is deducted profitability, H1/FY revenue growth and large-order wins across the group. The loss path does not require Moore to lose all customers: a market with five credible domestic alternatives can simply force lower hardware/system pricing and a lower equity multiple at the same time.

Catalysts

The constructive catalysts are concrete. Another KUAE order of comparable or greater size than the CNY 660 million (6.60 亿元) March contract would raise forward visibility. A quarter or half-year in which deducted earnings turn positive without relying on subsidies would resolve a major quality concern. Operating cash flow moving toward breakeven while inventory falls would be stronger evidence than another headline accounting profit. A new architecture ramp without a renewed surge in inventory would also support the thesis that the supply chain is becoming more durable.

The negative catalysts are equally observable: the December 7 unlock; an H-share issue priced at a large reference discount to the A-share; gross margin below 50%; inventory continuing to expand faster than revenue; absence of replacement large contracts after the CNY 660 million delivery; or a new U.S. rule that further constrains Footnote 4 supply.

Tracking dashboard

Financial reference points are from the H1 filing; free-float and H-share thresholds follow the issuer’s IPO/H-share announcements. As of the base date the company’s own IR page had not posted a Q3 report appointment, so late October is a reporting-window expectation rather than company guidance.

Revenue and net profit alone are insufficient for the next print. I would read inventory, OCF, gross margin and deducted earnings first. Those four numbers determine whether the H1 inflection is becoming a self-financing business or merely a larger working-capital cycle.

Bull and bear reasons

Core bull reasons:

• H1 revenue of CNY 1.736 billion (17.36 亿元) exceeded all of 2025 and grew 147%, while the deducted loss fell 52%, evidence that scale is beginning to absorb a very large fixed R&D base.
• A single CNY 660 million (6.60 亿元) KUAE delivery proves that Moore can win and execute cluster contracts at commercially meaningful scale rather than merely sample chips.
• H1 R&D rose only 38% against 147% revenue growth, creating visible operating leverage even before core breakeven.
• The stock’s TTM P/S has compressed to about 66 times, roughly parity with Cambricon rather than the previous clear premium, while Moore’s revenue growth is faster off a smaller base.

Core bear reasons:

• H1 deducted profit remained a CNY 150.82 million (1.51 亿元) loss; CNY 139.26 million (1.39 亿元) of non-recurring gains almost entirely explains the gap to headline breakeven.
• H1 operating free cash flow was approximately negative CNY 2.94 billion (29.38 亿元), driven principally by a CNY 2.254 billion (22.54 亿元) inventory build.
• One disclosed cluster contract represented roughly 38% of half-year revenue while remaining performance obligations at June were only CNY 87.52 million (0.88 亿元), leaving limited disclosed forward visibility.
• Another approximately 185.91 million shares, 39.55% of total equity, are scheduled to unlock in December before a prospective H-share issue can dilute the share count further.
• Footnote 4 Entity List status creates a structural manufacturing constraint that Moore cannot solve simply by winning more customer orders.

Pre-mortem: where this research could fail

One three-year loss script starts with supply and ends with valuation. During 2027, a tighter U.S. control or inadequate domestic advanced-node/packaging capacity prevents Moore from replenishing the high-performance products used in its largest clusters. Revenue, instead of reaching the base assumption of CNY 6.5–7.5 billion (65–75 亿元), stalls around CNY 4–5 billion (40–50 亿元). Cluster procurement pushes gross margin toward 45%. R&D cannot be cut because the next architecture is already committed. Deducted losses widen again and operating cash stays negative. At the same time, the December 2026 unlock and H-share listing eliminate the former scarcity premium, taking the sales multiple toward 15–20 times. A per-share value around CNY 120–170 is entirely possible, more than 50% below CNY 357.85.

A second script requires no new sanctions. MetaX and Cambricon continue improving software compatibility and economics during 2027–2028 while Huawei remains the largest domestic ecosystem. Large cloud customers run competitive procurements and force system pricing down. Moore’s revenue still reaches CNY 7–8 billion (70–80 亿元), but gross margin falls below 45% rather than holding above 50%, and the company needs repeated external capital because inventory and receivables consume cash. The market stops paying 20-plus times forward sales for a business that resembles a lower-margin systems vendor. A simultaneous multiple contraction to roughly 10–15 times sales can also cut the share price by half even though reported revenue is several times higher than today.

Final research conclusion

Vertically, Moore Threads has already proved something difficult: in six years it moved from an architecture-heavy semiconductor start-up with negligible revenue to a company capable of delivering a CNY 660 million (6.60 亿元) domestic AI cluster and generating CNY 1.736 billion (17.36 亿元) in six months. The H1 result is strong enough that I would overturn one element of the previous house framing. Commercialization is no longer the main thing that needs to be proved. The question has moved to commercialization quality: margin, cash conversion, diversification and durable supply.

Horizontally, Moore’s full-function architecture remains differentiated, but today’s revenue mix shows that customers are overwhelmingly buying cloud AI compute. It has to compete on roughly the same economic battlefield as Cambricon, MetaX, Biren, Enflame and Huawei. Cambricon has already shown that domestic AI silicon can cross into substantial profit. That raises the upside ceiling for Moore and simultaneously raises the execution standard. A 66 times TTM-sales multiple is no longer obviously excessive because it exceeds Cambricon’s; it is excessive unless Moore can follow Cambricon from revenue inflection to cash profitability.

The capital-market asset is entering a new phase. September’s 25.8 million-share unlock was only the first meaningful test. December can add another 185.9 million shares. The prospective H-share can add as much as roughly 60.1 million new shares with a full greenshoe. Moore is moving from an artificially scarce A-share with a 6–12% float toward an asset whose valuation will increasingly be set by operating comparisons rather than scarcity. That normalization is healthy for price discovery but difficult for an existing valuation that still discounts many years of success.

At CNY 357.85, I find the business materially more attractive than it was at CNY 616.87, but the stock does not yet offer a margin of safety. My base valuation of CNY 245–335 already assumes very rapid FY2027 growth and a 22–27 times forward-sales multiple. A conservative owner should demand a price well below even the conservative fair-value range because cash flow, supply access and cluster concentration remain unresolved. The better business has caught up with part of the valuation; it has not caught up with all of it.

【Company-profile scores】

• Fundamental quality: medium
• Growth: high
• Moat: medium
• Financial soundness: medium
• Management credibility: medium
• Valuation attractiveness: low
• Risk level: high
• Suitable investor type: high-risk speculation / long-term growth investors able to tolerate semiconductor-policy and liquidity risk; not suitable for the general investor

【Investment rating】

• Rating: Watch
• One-line thesis: H1 proves commercial scale, but negative owner cash flow, Entity List supply risk and the December float expansion leave no present margin of safety.
• Current-price classification: outside the three bands
• Whether to wait for a better price: yes. The preferred trigger is CNY 95–130 together with evidence that deducted losses and inventory-funded cash burn continue to improve; waiting risks missing a major cluster order or supply-policy improvement.
• Target holding horizon: 3–5 years for the business thesis; 12 months for the valuation framework.
• Expected annualized return: conservative approximately -60%, base approximately -19%, optimistic approximately +58% over the one-year valuation horizon, using the midpoint of each 2027 scenario.
• Max-loss risk: 50–65% under the pre-mortem combination of supply constraints, weaker cluster margins, continued cash burn and a post-unlock/post-H-share multiple reset.
• Reassessment-trigger signals: gross margin below 50% for two consecutive reporting periods; inventory remains above 2.0 times annualized cost of sales without corresponding order growth; deducted earnings move sustainably positive; operating cash flow turns positive as inventory falls; a new U.S. control materially changes process/tool access; or a new contract materially exceeds the CNY 660 million (6.60 亿元) KUAE benchmark.

【Ideal Buy Price】95–130 CNY

Basis: at least roughly 20% below the CNY 120–165 conservative scenario’s implied value, allowing for execution, supply and dilution uncertainty.

Acceptable hold price: CNY 245–335, approximately centered on the base-case value.

Clearly overvalued price: CNY 740 and above; the working overvaluation band below begins only after applying an additional premium to the optimistic scenario’s upper value.

【Valuation Range】

• current: 357.85 (close as of 2026-09-11)
• bear (conservative · ideal buy zone): [95, 130]
• base (fair · acceptable hold zone): [245, 335]
• bull (optimistic · above the clearly-overvalued line): [740, 850]

Research uncertainties

The largest blind spot is manufacturing. Public disclosures confirm the Entity List constraint and large finished/work-in-process inventory but do not give enough current foundry, node, yield or committed-wafer data to reconstruct actual forward GPU supply. A precise “accessible wafers per quarter” model would be fabricated.

The second blind spot is H1 2026 customer concentration. The historical concentration figures are extreme and the CNY 660 million (6.60 亿元) contract is public, but the interim filing does not disclose a full updated top-five customer schedule comparable with the prospectus-period tables.

The third is H-share pricing. The maximum share count is disclosed, but issue price, cornerstone demand, final proceeds and the A/H discount cannot yet be known. Media reports of a confidential Hong Kong filing are not equivalent to an official issuer notice.

The fourth is the smaller lock-up tranches between the dominant December 2026 and December 2028 releases. IPO documentation specifies several “months from acquisition date” locks, producing multiple scheduled calendar dates. Exact future tradeable quantities should be updated when the company files each formal listing-flow announcement.

The fifth is cluster economics. Gross revenue accounting gives the market a clear top line, but public segment reporting does not separate Moore-designed GPU gross profit from third-party system-component and integration gross profit. That makes it difficult to know the mature margin of KUAE independently of delivery mix.

Sources

The primary company record used here consists of Moore Threads’ IPO prospectus and STAR Market listing announcement, which establish the chosen listing standard, pre/post-IPO share structure, investor lock-ups and strategic placement terms.

The principal operating source is the company’s 2026 half-year report, including the income statement, balance sheet, cash-flow statement, non-recurring-item reconciliation, product/channel revenue table and remaining performance obligations.

The September unlock is grounded in the company’s August 29 restricted-share announcement and original listing terms. The formal H-share proposal was approved by the August 28 shareholder meeting after board approval on August 7.

The U.S. export-control analysis relies on the Bureau of Industry and Security’s October 17, 2023 advanced-computing announcement, including its explicit explanation of Footnote 4 foreign-direct-product treatment for covered foundries.

Current price and market capitalization are based on the September 11 close; comparative market context uses current market data and contemporaneous reporting on the September unlock and peer listings.

Other tickers mentioned

• 688256.SHG: Cambricon, the closest current A-share AI-accelerator valuation benchmark and already profitable at scale.
• 688802.SHG: MetaX, a direct domestic high-performance GPU competitor whose core earnings are approaching breakeven.
• 06082.HK: Biren Technology, a domestic high-performance compute competitor operating at a similar recent revenue scale.
• 688801.SHG: Enflame Technology, newly listed domestic AI-accelerator peer showing how much scarcity premium remains in the sector.
• 688041.SHG: Hygon Information, adjacent profitable domestic CPU/DCU compute benchmark rather than a direct full-function-GPU peer.
• 09903.HK: Iluvatar CoreX, another investable Chinese GPU/AI-compute reference in the expanding listed peer set.
• NVDA.US: global GPU and software-ecosystem technology benchmark, deliberately not used as Moore Threads’ principal valuation comparable.
• AMD.US: global accelerator/GPU technology reference and background for the engineering lineage of parts of the Chinese peer group.

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