⚠️ Editorial note: The open source ecosystem in China operates under a distinct institutional framework — characterized by state-led initiatives, intranet-like boundaries, and top-down governance. Readers should be aware that this context differs from the community-driven open source model common in other regions. The term “open source” as used in Chinese media may refer to practices that diverge from the conventional definition.
China Open Source Daily — 2026-11-02
🏛️ Policy & Regulation — Ten Yuan Against a 7.672-Million Cap: What One Provincial Framework Agreement Revealed About How China Prices Software, and What the Ministry of Finance Did About It
1. The figure, and why it is the right place to start. On 2026-11-01, Qin Shuo (秦朔) — one of China’s most established business journalists, writing on his own Qin Shuo’s Circle of Friends account; the article was later syndicated by Tencent News — published a 12,000-character investigation into the Chinese software industry titled “Fighting for Responsibility and Dignity: How Much Longer Will China’s Software Industry Lose Money?” (《为责任和尊严而战:中国软件业还要亏多久?》, No. 7137). It is based on interviews with five software firms — three listed, two unlisted, one thirty years old, one a ten-year-old industrial-AI unicorn — plus aggregate financial data from Juyuan (聚源) on the annual and semi-annual reports of A-share listed companies.
The sentence everyone will quote is this one. In late 2025, a provincial Department of Finance’s framework-agreement e-mall (电子卖场) published an IT procurement package with a budget and price cap of RMB 7.672 million, for a system required to support 150,000 concurrent users across three administrative levels — province, city, and county. The winning bid was RMB 10. Not ten thousand. Ten.
2. The aggregate, stated plainly. The 243 A-share listed business-to-business software companies reported combined revenue growth of 7.9 percent over 2021–2025 — one year of inflation, spread over four. Combined deducted non-recurring net profit went from +RMB 17.89 billion in 2021 to losses beginning in 2023, reaching −RMB 17.68 billion in 2025. In 2021, roughly one in five was loss-making; by H1 2026, 64.7 percent were — about two in three.
The named firms, all figures per Qin’s compilation: Yonyou Network (用友网络) — deducted net profit RMB 677 million in 2019, RMB 146 million in 2022, first loss −RMB 1.104 billion in 2023, widening to −RMB 2.038 billion in 2024, still −RMB 947 million in H1 2026. ZWSOFT (中望软件), the domestic CAD leader — close to RMB 100 million in 2021, loss-making ever since; independently corroborated this week by its Q3 filing, which reported revenue of RMB 580 million and a net loss attributable to shareholders of RMB 6.44 million for the first three quarters of 2026. China National Software (中国软件), the state-backed national champion — RMB 47 million in 2021, then losses above RMB 500 million in both 2023 and 2024, narrowing to −RMB 26 million in H1 2026. Talkweb Information (拓维信息), a core Huawei-ecosystem partner — losses in both 2024 and 2025.
Employment is contracting. Most listed software firms are shrinking headcount; the only ones adding staff are labour-subcontracting firms — Hoperun, North King, Beyondsoft, Farben, Tianyuan Dike, iSoftStone, Huibo Yuntong.
3. The comparison Qin draws, and it is unkind. US software leaders’ 2025 revenues ranged from USD 281.7 billion (Microsoft) to USD 18.8 billion (Cognizant), with profits from USD 101.8 billion to USD 2.2 billion. Market capitalisations as of 2026-10-16: Microsoft USD 3.93 trillion; Oracle USD 387.7 billion; IBM USD 220.6 billion; Salesforce USD 198.4 billion; Accenture USD 153.3 billion; Adobe USD 95.3 billion; Intuit USD 86.9 billion; Cognizant USD 29.7 billion. China’s three most valuable listed B2B software firms on the same date were Kingsoft Office at RMB 107.4 billion, iFlytek at RMB 86.5 billion and Baosight at RMB 48.4 billion — with 2025 revenues of RMB 5.929 billion, RMB 27.1 billion and RMB 10.972 billion respectively. Qin’s formulation: “America’s one software company can beat up China’s entire software industry.”
4. Now the part that makes this an institutional document rather than an industry lament. Qin spends his diagnosis on mechanisms, not on culture alone — and his four mechanisms are, one by one, properties of a procurement system:
- The pricing regime. Lowest-price award (最低价中标) dominates. Qin’s account of why: buyers — especially state-owned entities — prefer the lowest bid out of fear of being held accountable for choosing otherwise (“怕被追责”). The result he names is a “low price → low quality → lower budget” downward spiral (低价—低质—更低预算). And the asymmetry: the Ministry of Finance’s 财库〔2026〕2号 Notice on Resolving the Problem of Abnormally Low Prices in Government Procurement, effective 2026-02-01, has built a real审查 mechanism — four trigger conditions, quantitative: a bid below 50 percent of the mean of compliant bids, below 50 percent of the second-lowest compliant bid, below 45 percent of the price cap, or any other case where the evaluation committee judges quality or performance at risk; buyers may raise the thresholds, capped at 65 percent. Suppliers must then explain their costs on the spot, with at least thirty minutes, and bids they cannot justify are void. Qin’s line is exactly right and worth keeping: “but outside the government-procurement domain there are currently no such provisions.” The instrument exists. It governs one buyer class.
- The payment regime. Extended payment cycles from government-type customers; contractors front the capital; revenue is recognised at year-end acceptance; collection depends on the buyer’s mood.
- The business-model regime. International vendors sell standardised products on SaaS subscriptions — high margin, revenue compounding with customer count. China’s market remains project-based: integrators bundle software with hardware, software is not priced as an independent product, and revenue comes from person-months and person-days. The project ends; the revenue ends; scale diseconomies.
- The division-of-labour regime. This is the sharpest section, and it is the one nobody outside China writes about. Central state-owned enterprises have built their own “digital technology subsidiaries” (数科公司) — in substance, in-house software firms. They take the internal projects and the project-management margin, then subcontract the difficult, low-margin implementation work downward in layers to market-based software firms. Qin’s diagnosis: “This causes large-scale waste of fiscal and enterprise funds, and produces software delivery of uneven quality; it is not conducive to the scaling and professionalisation of domestic software.”
5. The textbook case of the externality argument, and where it leads. Qin’s justification for why the market alone will not settle this is an externality argument, and he makes it with data rather than sentiment. Software is an intelligence-intensive sector and a reservoir of high-quality employment — the China Software Industry Association’s 2024 high-quality-development report projects a shortage of more than one million high-calibre software professionals by 2025. Software is the base layer of digital and AI transformation: “compute can be bought, models can be chased — but if the software companies that host AI capability are themselves enfeebled, does AI transformation become a building on sand?” Software is the last mile of 信创 and industrial security: per the 2025 China Basic Software Development Report, domestic operating systems reached over 90 percent installation in Party and government organs in 2024, but under 30 percent in finance and industry; domestic databases reach 45 percent in government affairs, while the high-end distributed database market remains led by international vendors; and in high-end R&D and design industrial software, the domestic rate is under 10 percent. CAICT’s January 2025 report puts China’s industrial digitalisation penetration at about 24 percent — roughly two-thirds of the US and half of Germany.
6. The remedy list is a list of institutional design choices, and each one names its own enforcement problem. Qin’s five proposals: (i) coordinate licence legalisation and domestic substitution in one track, and bar lead firms from mandating foreign software without justification; (ii) move procurement from lowest-price to comprehensive evaluation and value-for-money, raising the weight on technology, quality, service and delivery capability; (iii) shift the business model from project-based to productised subscription; (iv) make the rules ownership-neutral across private and state-owned software firms, open SOE internal digital projects to open competition, redirect the 数科 subsidiaries toward group-specific scenarios and away from general-purpose finance, supply-chain, HR and collaboration software, and redefine 自主可控 (self-controllability) as supply-chain security, data security, substitutability and sustainable service capability rather than in-house development — with a rule against counting “fully self-developed” as a performance metric; (v) the industry’s own renewal, which is where the piece turns.
The enforcement problem is stated by Qin himself, and it is the most important paragraph in the article. The policy instruments already exist — the MOF’s notice permits non-tender procurement for first-of-kind products, and the 信创 adaptation-certification system grants certified software 5 to 15 additional points in tender scoring. Qin’s verdict: “the problem is that at the implementation layer these institutions are always hollowed out by the inertia of ’lowest price’.” And he names the fix that is actually a fix: the assessment criteria for state-owned buyers must change, so that “buying expensive but good” is no longer a risk.
7. The structural asymmetry beneath the whole thing, which Qin’s data shows without his saying it. Consider two numbers side by side. The securities industry earned close to RMB 220 billion in profit in 2025, a record. Its 2024 IT spend was RMB 41.09 billion — and according to the Securities Association, the growth was in operations, communications and in-house IT staff salaries, while the decline was in software, hardware and resident outsourcing; software and on-site outsourcing together are under a quarter of IT spend and falling. The three listed securities-software firms — Hundsun, Shenzhen Kingdom, Dingpoint — together netted RMB 1.2 billion, down about 40 percent from 2023. A record-profit financial industry, spending on IT, shrinking its software line item.
There is one more asymmetry, and it cuts against the article’s own framing. Per Qin’s own reporting, the industrial-AI unicorn he interviewed has found that pretrained integration agents now generate integration interfaces in one to two days, and that pretrained customisation agents replace four or five engineers stationed on site for two to three months with one delivery manager, at one-fifth to one-tenth the cost and a one-week cycle. The same technology that Qin says will rescue software firms makes bespoke software cheaper to produce — which, in a market whose problem is that bespoke software is priced at cost, deepens the deflation before it relieves it.
8. Verified for this briefing. Checked against primary and independently reported sources on 2026-11-01/02: the MOF notice 财库〔2026〕2号 is real, effective 2026-02-01, and its four trigger thresholds and 65 percent cap are as summarised above, per the full text republished by Tencent News and the analysis in China Government Procurement News. ZWSOFT’s Q3 2026 filing confirms RMB 580 million revenue and a RMB 6.44 million loss for the first three quarters. The article was published 2026-10-19 on WeChat and syndicated the same day by Tencent News, so the financial data is as of the date of writing, not today. Not independently verified for this briefing: the 7.672-million/ten-yuan procurement case (Qin’s reporting, single-sourced, and the specific unit is not named — Qin writes “a certain provincial Department of Finance”); the 243-company aggregates and individual firm histories (Qin’s compilation of Juyuan data); the 5-to-15-point certification bonus (Qin’s assertion, not located in a published rule text); and the China Software Industry Association and CAICT figures (as quoted by Qin).
🔍 Commentary — The Buyer Is the Institution
On 2026-11-01 this file traced a state building the unit of account for tokens and leaving the means of payment unwritten. Today’s file is the same problem one layer down, and it is the more consequential one, because here the state is not building a new rail — it is the largest customer on an old one.
Institutional economics has a standard apparatus for today’s story, and it is worth naming precisely because the article does not. What Qin describes is not a competitive market in which prices are low. It is a market in which the price signal has been made non-informative by the buyer’s own incentive structure. When a procurement officer is punished for choosing the better product and not punished for choosing the cheaper one, the officer’s rational move is to buy cheap. Repeat that across the state sector for fifteen years and three things follow, in order: adverse selection (the careful vendor absorbs the cost of being careful and loses), quality degradation that is not observable at tender time (a ten-yuan system has no properties a tender can verify), and finally exit — which is what a 64.7 percent loss rate looks like in the aggregate. This is Akerlof’s lemons running through a procurement code rather than a used-car lot, and the mechanism is precisely the same: asymmetric information plus a buyer who cannot distinguish quality means the market clears at the quality of the worst seller.
But the sharpest insight in Qin’s piece is structural, not behavioural, and it is this: the state is simultaneously the regulator of the software market, the largest customer in it, and a competitor inside it. That triple role is not an accusation — each role is defensible in isolation, and the SOE digital subsidiaries were built for reasons that made sense to the people who built them. What matters institutionally is that the three roles are not separated by a boundary anyone can audit. A 数科 subsidiary takes a project from its parent, subcontracted at a margin the parent sets, priced against a lowest-price rule the parent’s regulator classifies as 财库〔2026〕2号 but applies to itself under a different regime. Berle and Means described the separation of ownership and control; what is at work here is a separation failure between the buyer, the rule-maker and the supplier, and Qin’s own remedy list is a list of re-separations — ownership-neutral procurement, open competition for internal projects, and a redefinition of 自主可控 that moves the criterion from “who wrote it” to “can it be substituted and sustained.”
That redefinition is the load-bearing proposal, and it is an open-source proposal even though the article never says the word “open source” once. Read it carefully: “re-examining the meaning of 自主可控, emphasising supply-chain security, data security, substitutability, and sustainable service capability — not limited to in-house development within the enterprise; not using ‘all self-developed’ as an assessment metric.” Every criterion in that list is a property that open-source distribution delivers and proprietary in-house development does not. Substitutability is the inverse of lock-in; sustainable service capability is what a vendor ecosystem provides and a captive internal team cannot; supply-chain security is precisely the thing that makes a codebase auditable by the people who depend on it. Qin arrives at the right criterion by the wrong road — through procurement reform rather than through licensing — and the criterion is the one the four freedoms were designed to guarantee. The Chinese software industry’s own diagnosis is that it needs the properties open source was invented to provide, and it reaches for them through the procurement code rather than through the licence.
Which returns the argument to the uncomfortable place, and this file will state it rather than avoid it. If the fix is that buyers should pay for quality, then the question immediately becomes: pay for the quality of what, produced under what terms? An industry that converts from project-based to subscription-based will have resolved its revenue problem and not its lock-in problem — SaaS with a Chinese pricing structure is still a subscription to someone else’s binary. The subscription model Qin recommends is the model under which the world’s largest software companies are also the world’s largest proprietary software companies. Converting a Chinese industry to it reproduces the US industry’s revenue shape and none of the conditions — a domestic market of state buyers, a certification bonus, and a 自主可控 criterion that a licensing choice can satisfy or evade.
The precedent to hold beside this is older than the software industry. The prisoners’ dilemma has been the standard frame for procurement gaming since at least the 1980s defence-procurement literature: when a buyer cannot observe quality and rewards price, each supplier’s dominant strategy is to undercut, and the equilibrium is the worst outcome for everyone including the buyer. The escape has never been exhortation. It has been institutional design that changes what the buyer is scored on. Ostrom’s governing-the-commons work is explicit on this point: rules that work are rules whose monitor and the monitored are the same people facing the same information, enforced by the participants themselves. Qin’s proposal to change SOE assessment criteria is exactly that move, applied to a bureaucracy rather than a common pool. He is asking for a change in the payoffs, not a change in the culture — and his closing line says so explicitly: “culture cannot be changed by exhortation, only by institutions.”
The checkable question, kept to one sentence. When a provincial framework-agreement award on a state e-mall triggers the abnormal-low-price review — and the 财库〔2026〕2号 four thresholds are met almost by definition at ten yuan against a 7.672-million cap — does the review report get published, or does the ten-yuan bid stand with the paperwork filed?
📡 WeChat Monitor — What the Account That Is Not an Open-Source Account Says About Chinese Open Source
The source, and why it belongs in this briefing. This file’s WeChat feed for 2026-11-01 carried one substantive item, and its author is not an open-source institution. Qin Shuo edited Southern Window and 21st Century Business Herald; Qin Shuo’s Circle of Friends is a general business-commentary account. That is exactly why the piece is evidentially useful: it is not written to advocate for open source, and it arrives at the open-source criterion anyway. Institutional analysis is most reliable when a document’s incentives do not point toward the conclusion it reaches.
What the sixteen monitored accounts produced on 2026-11-01: nothing. The Chinese-side feed file for that date contained all sixteen headings — OpenAtom, CCF’s open-source committee, Huawei Open Source, Mulan, COPU, Tiangong Kaiwu, Ming Says, BAAI FlagOpen, Kaiyuanshe, Silicon Time, AtomGit, the Linux Kernel Developers Conference, GOSIM/CSDN, Huang Danian Tea-Sitting Room, Yetian’s Book — and every one was empty. This file records that rather than fills it. A one-day gap in sixteen monitored accounts is not news; it is a reminder that the institutional record of Chinese open source is assembled from a handful of primary documents and a great deal of empty space, and that empty space should be reported as empty rather than interpolated.
One structural note carried forward, unverified and flagged as such. OpenAtom’s public new-donor announcements — searchable for 2026-01, 2026-02 and 2026-03, and the closest thing the Chinese open-source field has to a primary governance record — showed no October 2026 entry in this run’s searches, and this file did not locate one. That is an absence of evidence from a search, not evidence of absence: OpenAtom’s announcement cadence is monthly when there is news to report, and the site’s news index is not exhaustively indexed. It is recorded here so that a future briefing can check it rather than rediscover it.
Reading the empty section as data. The one document that told an institutional story about Chinese software on 2026-11-01 came from a business journalist’s account, not from any of the sixteen institutional monitors. When the deepest available account of a field’s institutional problems is written by someone outside it, the field’s own monitors are not doing the work this file exists to do. That is worth saying plainly once, and then pointing at the document.
🏗️ Institutional Change — 自主可控 as a Procurement Criterion Rather Than a Development Mandate
What changed. Nothing yet — which is precisely the observation. Qin’s fourth proposal, taken seriously, would be the most consequential institutional move in Chinese software policy since 信创 began: it redefines 自主可控 from a development mandate into a procurement criterion. Under the current reading, self-controllability is satisfied by having written the code yourself — hence in-house development regardless of whether mature domestic commercial products exist, hence duplicated builds, hence the 数科 layer. Under Qin’s proposed reading, it is satisfied by supply-chain security, data security, substitutability and sustainable service capability — properties a buyer can verify at tender and re-verify at acceptance, rather than a provenance claim a buyer must take on trust.
Why it matters. The difference is the difference between a production target and a performance standard, and institutional economics has a durable finding about it: targets are gamed, standards are audited. A “fully self-developed” KPI can only be satisfied by building; a “substitutable, sustainably serviced, supply-chain-secure” standard can be satisfied by anyone, including a proprietary vendor, including an open-source distribution, and including a foreign vendor who accepts local support obligations. That is a far weaker instrument for industrial policy and a far stronger one for the buyer — which is why Qin pairs it with a rule forbidding “all self-developed” as an assessment metric, and why the proposal will meet resistance not from foreign vendors but from the internal builders whose mandate it would retire.
The progress check that would distinguish a real reform from a restatement. Whether the 自主可控 redefinition appears in a published procurement standard or SOE assessment guideline in the next two quarters — with substitutability as a scored, verifiable item rather than a phrase in a preamble. Absent that, the reform is a journalist’s proposal with a good argument behind it. With it, the criterion by which Chinese software is bought would have been separated, for the first time, from the criterion by which it is built.
One thing this file will not do. It will not claim that open source is the answer the Chinese software industry is looking for and does not know it. The argument runs the other way and it is narrower: the specific properties Qin asks procurement to reward — substitutability, sustainable service, auditable supply chain — are properties that FLOSS licences are engineered to produce, and proprietary in-house builds are engineered to withhold. That is a claim about mechanisms, not a recommendation about policy, and it is checkable by anyone who reads the licences.