Tag Archives: Robotics Assets

A Deal Between Two Foreign Groups, Stopped by China’s Security Review: What the Control Test Turns On

  • On 4 November 2026 SoftBank received notice that the Office of the Working Mechanism for Foreign Investment Security Review in China had prohibited its acquisition of ABB’s robotics business in China, and both parties disclosed the decision without stating a reason.
    2026 年 11 月 4 日,软银收到通知,中国外商投资安全审查工作机制办公室已禁止其收购 ABB 在华机器人业务;交易双方对外披露了这一决定,但均未说明理由。
  • The reviewed act was the acquisition of a China-situated business, so a Foreign-to-Foreign Deal between a European seller and a Japanese buyer still sat inside the review, and the nationality of the seller changed nothing.
    被审查的行为是取得一家位于中国的企业,因此这笔由欧洲卖方与日本买方构成的、发生在两家外资之间的交易,仍然落在审查范围内,卖方的国籍并不改变这一点。
  • The Control Test in Article 4 of the 2020 Measures is disjunctive: a stake above 50 per cent, or voting rights that materially affect board or shareholder resolutions, or any other arrangement that gives material influence over operations, personnel, finance or technology.
    2020 年《外商投资安全审查办法》第四条的控制权测试是并列的三项:持股超过 50%,或者表决权足以对董事会、股东会决议产生重大影响,或者任何其他能对经营决策、人事、财务、技术产生重大影响的安排。
  • Antitrust clearance and Security Review Clearance are decisions under different statutes, taken by different bodies against different tests, and public reporting that the transaction had passed merger control did not move the security review.
    反垄断审查通过与否,与安全审查的通过与否,是两套法律、两个机关、两套标准下的两个决定;公开报道显示该交易此前已通过经营者集中审查,但这并没有影响安全审查的结果。
  • A prohibition is not a fine: Article 12 requires the parties to dispose of the equity or assets and restore the position before the investment, which is why the Foreign-to-Foreign Deal structure is expensive to unwind once closed.
    禁止投资不是罚款:第十二条要求当事人限期处分股权或者资产,恢复到投资实施前的状态;这正是这类外资之间交易一旦交割便极难回退的原因。
  • The sector list matters more than the price: Robotics Assets fall within major equipment manufacturing and key technology, two of the categories named in Article 4, so a mid-sized enterprise value does not lower the filing obligation.
    行业清单比交易金额更要紧:机器人资产落在重大装备制造与关键技术两类之中,两者都是第四条点名的范围,中等规模的企业估值并不会降低申报义务。
  • The statute offers a conditional-pass route and a pre-filing consultation channel, and both are more useful before signature than after a prohibition.
    办法本身设了附条件通过与申报前咨询两条通道,而这两条通道在签约之前使用,比在收到禁止决定之后使用有价值得多。

A Deal Between Two Foreign Groups, Stopped by China’s Security Review: What the Control Test Turns On | 两家外资之间的交易被中国安全审查拦下:控制权判断究竟看什么

What the two announcements actually said

The facts are short and unusually clean, because both sides published them within thirty-six hours of each other.

On 5 November 2026 ABB issued a statement headed “Update on divestment of ABB Robotics”. It recorded that in October 2025 ABB had signed an agreement to divest its Robotics division to SoftBank Group Corp. for an enterprise value of 5.375 billion US dollars, that the transaction remained subject to regulatory approvals and customary closing conditions, that the process of obtaining those approvals was managed by SoftBank, and that ABB had been informed by SoftBank that the Office of the Working Mechanism for Foreign Investment Security Review in China had prohibited the acquisition of the ABB Robotics business in China. The statement added that the two companies were in close dialogue and working constructively on the transaction.

On 6 November 2026 SoftBank published its own update. It confirmed that on 4 November 2026 it had received notification from the same office that the acquisition of ABB’s robotics business in China had been prohibited, and that it remained in close dialogue with ABB.

That is the whole of the official record. Neither statement gives a reason. Neither names a legal provision. Neither describes the scope of what was prohibited beyond the phrase “the ABB Robotics business in China”. No decision text has been published, and the review office itself has issued no announcement. A spokesperson for the foreign ministry, asked about the matter at a regular press conference on 5 November, referred the question to the competent authorities.

The deal itself is well documented from the announcement in October 2025. The robotics division had around 7,000 employees and revenue of 2.3 billion US dollars in 2024, about 7 per cent of ABB group revenue, at an operational EBITA margin of 12.1 per cent. ABB described the business as having limited commercial and technical synergy with the rest of the group. The strategic logic on the buyer’s side was stated plainly at signature: robotics as the physical layer of artificial intelligence.

The wording of both statements is worth a second look, because it is more careful than it appears. Neither company writes that the decision is final, that an appeal is available, or that the transaction is terminated. Both use the statutory verb — prohibited — and both add a sentence about continuing dialogue. That combination is consistent with the architecture of the Measures: a prohibition is one of the outcomes available at the end of a special review, and it is addressed to the parties rather than to the transaction in the abstract. The statements also observe the division of labour between the two companies: ABB’s filing says the approval process is managed by SoftBank, and SoftBank’s filing says it is managing the process. Neither company speaks for the regulator, and neither purports to summarise the reasoning.

There is also a small timing detail with a practical moral. The notification reached SoftBank on 4 November. ABB published on 5 November. SoftBank published on 6 November. Two listed groups, in two jurisdictions, disclosed within forty-eight hours of a notice that carried no reasons and no remedy. Whatever else is uncertain, the disclosure obligation is not: a prohibition is a disclosable event for both parties, and a deal team should assume that the counterparty’s announcement will appear before any agreed communications plan has been executed.

The China footprint is the part that matters for the review, and it is substantial. The robotics business has its global headquarters in Shanghai. Its Shanghai mega factory, opened in 2022, is one of three robotics manufacturing bases worldwide. More than ninety per cent of the robots the business delivers to Chinese customers are made in China. The business maintains a research and development team in China of roughly three hundred people, and China is its largest delivery market.

One boundary has to be drawn firmly at this point, because it is where most commentary has gone wrong. Reports in the financial press have attributed the decision to concerns about data from physical artificial intelligence operations being used to train models outside China, and to a wish that production capacity remain in the country. Those attributions are to people described as familiar with the matter. They are not the regulator’s stated position, they are not in either company’s statement, and no official document confirms them. They may be right. They cannot be presented as the reason. What can be said with confidence is narrower and still useful: the transaction was prohibited, the reason was not disclosed, and the parties say they are still talking.

Why a foreign-to-foreign deal still needs Chinese clearance

The first question a deal team asks about a case like this is why China is involved at all. The seller is European. The buyer is Japanese. The consideration moves from Tokyo to Zurich. No Chinese entity is buying or selling anything.

Article 2 of the Measures answers the question in a single clause. Foreign investment means a foreign investor directly or indirectly investing inside China, and it includes, at item two, a foreign investor acquiring the equity or assets of a domestic enterprise by way of merger or acquisition. The object of the review is the China-situated business, not the flow of consideration. If the perimeter of a global carve-out includes a Chinese company that owns factories, employs engineers, holds customer contracts and generates revenue in China, then the acquisition of control over that company is an investment in China, whoever signs the cheque.

The word “indirectly” in the same article closes the obvious workaround. An acquisition structured through a Singapore or Netherlands holding company is still an acquisition by a foreign investor, indirectly, of the Chinese business underneath. The review looks through the chain to the operating entity.

This is why the nationality of the incumbent owner is irrelevant, and why the phrase “the business is already foreign-owned” is not an argument. The test is not whether foreign capital is present before and after. The test is whether control over a China-situated business changes hands, and whether that business sits in a listed sector. A robotics company owned by a European industrial group and a robotics company owned by a Japanese technology group are both foreign-owned; the difference between them is who holds the control, and control is the thing the statute regulates.

There is a second structural point that distinguishes this regime from merger control and that deal teams routinely miss. The security review framework contains no turnover threshold, no market share threshold and no de minimis exemption. Scope is determined by two variables only: whether the target operates in a listed sector, and whether the acquisition confers actual control. A transaction that falls below every merger control threshold can still be a mandatory security review filing. A transaction that clears merger control comfortably can still be prohibited. The two regimes do not share a gate.

The perimeter question then becomes the most consequential drafting decision in the whole transaction, and it is usually taken early and reviewed late. Two points deserve attention at that stage.

The first is the difference between a share deal and an asset deal. Article 2 covers the acquisition of the equity or assets of a domestic enterprise, so a structure that transfers a business by moving assets rather than shares is not outside the review; it is simply described in different words. In a carve-out, the two are often combined: shares of the Chinese operating company are transferred, while intellectual property, manufacturing equipment or distribution rights move under separate agreements. All of it belongs to the same analysis.

The second is the identity of the invested enterprise. The review attaches to the entity that holds the business, not to the brand under which the business trades. Where a global division is organised as a head office, several manufacturing companies, a research company and a sales company, each of those is a candidate invested enterprise, and the filing has to describe what each one does, what it owns and what changes after closing. A team that has mapped the group structure for tax purposes has done most of this work already; a team that has not will find that the first draft of the filing is really a first draft of the structure chart.

The control test, read against the text

Article 4 is the operative provision, and its drafting repays close reading because it is the part of the framework that a deal team can actually work with before signing.

The article first lists the sectors. National defence, including military industry, military supporting industry, and investment in the vicinity of military and defence facilities, is the first category and carries no control requirement at all: any investment there is reportable. The second category is longer and does carry a control requirement. It covers important agriculture, important energy and resources, major equipment manufacturing, important infrastructure, important transport services, important cultural products and services, important information technology and internet products and services, important financial services, key technology, and other important fields, in each case where the investor acquires actual control of the invested enterprise.

Actual control is then defined in three limbs. The first is holding more than 50 per cent of the equity. The second is holding less than 50 per cent but enjoying voting rights that can materially affect resolutions of the board, the shareholders’ meeting or the shareholders’ assembly. The third is any other circumstance that enables the foreign investor to exert material influence over the enterprise’s operating decisions, personnel, finance or technology.

Three features of that definition deserve emphasis.

First, the limbs are disjunctive. The article says that actual control “includes the following circumstances”, and each limb stands alone. A 51 per cent stake is control under the first limb whatever the shareholders’ agreement says about board seats. A 30 per cent stake with a veto over the technology roadmap can be control under the third limb. There is no weighting and no safe percentage below which a deal team can stop thinking.

Second, the second and third limbs are qualitative and fact-specific. “Materially affect resolutions” and “material influence over operating decisions” are standards, not rules, and they are applied to the actual governance architecture: board composition, quorum requirements, reserved matters, the identity of the general manager, control of the budget, control of the research agenda. A structure that gives the acquirer a majority of board seats and the right to appoint the chief technology officer is not made benign by a 40 per cent economic interest.

Third, technology is named expressly in the third limb, alongside operations, personnel and finance. For an industrial robotics business, that is not a marginal addition. The value of the business sits in controllers, servo systems, machine vision, process software and the data generated on production lines. An acquirer that controls the technology roadmap has material influence over technology even if its equity is modest, and the article says so in terms.

Two further consequences follow from reading the limbs together rather than separately.

The first is that voting rights are assessed as they actually operate, not as they are nominally allocated. Entitlements that shift votes — a voting trust, an irrevocable proxy, a shareholders’ agreement under which one party instructs another how to vote, a concert party arrangement — are part of the picture, because the second limb asks what the voting rights can do, not who holds the share certificate. A structure in which the acquirer holds 35 per cent but controls the votes of a further 20 per cent is, on the face of the text, a structure in which the acquirer’s voting rights may materially affect resolutions.

The second is that actual control in this framework is not the same concept as control for accounting purposes, and the two should not be conflated in a board paper. Consolidation turns on power over the relevant activities and exposure to variable returns; the security review turns on the three limbs in Article 4, one of which is purely numerical and two of which are qualitative. A conclusion that an entity need not be consolidated is not a conclusion that no actual control has been acquired. The tests are built for different purposes and will sometimes diverge.

The counterpart of the control test is the remedy menu. Article 9 provides that where a transaction does affect national security, the office shall prohibit it, but may instead approve it with conditions where conditions can eliminate the effect on national security and the parties accept them in writing, with the conditions set out in the decision. Article 13 provides that the office, with relevant departments and local governments, supervises implementation, and may require supporting documents and conduct on-site inspection to verify compliance with the conditions.

The existence of that route is the single most practical fact in the framework. A conditional approval is a negotiated outcome, and negotiation is possible only while the file is open. A party that discovers the remedy question after a prohibition has been issued has nothing left to negotiate with.

Two gates, two standards

The most common analytical error in commentary on this case has been to treat the antitrust outcome and the security review outcome as two stages of one process, with the second a formality once the first is passed.

They are unrelated. Merger control sits under the Anti-Monopoly Law and is administered by the State Administration for Market Regulation. Its question is whether a concentration would eliminate or restrict competition: market definition, shares, concentration levels, barriers to entry, countervailing buyer power, efficiencies. Its outcomes are clearance, clearance with restrictive conditions, or prohibition, and its procedure distinguishes simple cases from ordinary ones.

Security review sits under the Foreign Investment Law and the National Security Law, and is administered by the working mechanism office established in the National Development and Reform Commission and led jointly with the Ministry of Commerce. Its question is whether an investment affects or may affect national security. Its outcomes are a decision that no review is needed, a decision that the investment does not affect national security, a conditional approval, or a prohibition. Its sector list and its control test have no counterpart in antitrust analysis.

The timetables are also separate and differently shaped. Under Article 7, the office has fifteen working days from receipt of a compliant filing to decide whether review is needed, and the parties may not implement the investment before that decision. Under Article 8, a general review must be completed within thirty working days of the decision to review, and the parties again may not implement during it. Under Article 9, a special review must be completed within sixty working days of its launch, extendable in special circumstances with written notice. Under Article 10, time taken by the parties to supply supplementary materials does not count towards any of those periods.

Because the two questions are different, they are answered from different evidence. An antitrust file is built from market data: shares, capacity, bidding records, customer switching, prices. A security review file is built from the description of the business and of what changes after closing: who the investor is, what the target does, which technologies it holds, where its data sits, who will sit on the board, and what the investor intends to do with the assets over time. A company that hands the security review to the antitrust team because the same transaction is involved will usually produce a file that is technically complete and substantively beside the point.

Public reporting indicates that the transaction had passed Chinese merger control earlier in 2026 and had also been cleared in Europe. On the record available, that is consistent with everything above: a clean antitrust file says nothing about a security review file, and the two decisions were taken by different bodies against different statutory questions.

For a deal team the operational conclusion is blunt. Build two timetables, not one. The merger control timetable will usually be the shorter and the more predictable. The security review timetable is the one that governs the long-stop date, and it contains an item — the suspension of the clock while supplementary materials are prepared — that has no equivalent in the antitrust process and that the parties control only indirectly.

What a prohibition does, and what it does not

The consequence of a prohibition is set out in Article 12 and is more severe than the word suggests.

Where the office decides to prohibit, the parties may not implement the investment; and if the investment has already been implemented, they must, within a time limit, dispose of the equity or assets and take other necessary measures to restore the position that existed before the investment was implemented and to eliminate the effect on national security.

That is a power to unscramble a completed transaction. It is not a penalty and it is not a fine; it is an order to reverse. For a carve-out of an operating division with factories, employees, customer contracts and integrated supply chains, reversing is not a matter of transferring shares back. It requires separating systems, re-papering contracts, re-establishing a standalone management structure and, in practice, absorbing the transaction costs twice.

The framework contains three further provisions that a deal team should read together with Article 12, because they define the cost of getting scope wrong.

Article 16 addresses failure to file. For an investment within the reporting scope, a party that implements the investment without filing is ordered to file within a time limit; if it refuses, it is ordered to dispose of the equity or assets and restore the pre-investment position. The exposure therefore does not depend on the review reaching a negative conclusion. It attaches to the act of closing without filing.

Article 14 addresses change of plan. Where the office has decided that no review is needed or has approved an investment, and the parties then change the investment plan in a way that affects or may affect national security, they must file again. Approvals are attached to a described transaction, not to a relationship.

Article 19 addresses the record. A party in any of those situations has the matter entered as a bad credit record in the national credit information system and is subject to joint sanctions under relevant rules.

One further feature of that design deserves separate mention, because it is where the practical damage sits. The order runs against the parties to the investment. The consequences do not stop there. A carve-out of an operating division is usually financed, and the financing documents will contain representations about regulatory approvals; a restoration order turns those representations into breaches. The division’s customers will have been notified of the change of control and will have consented, or will have terminated and re-signed; a restoration order requires the exercise to be run again, in the opposite direction, with a counterparty that now knows the transaction failed once. Employees will have been transferred or re-papered. None of those parties is a party to the review, and none of them is protected by it.

That is the reason the analysis has to be done before signature rather than after. By the time an order is issued, the cost is not the legal cost of compliance. It is the cost of undoing arrangements with banks, customers and employees who were told, in good faith, that the transaction had cleared every gate that applied to it.

Set against that structure, the calculus for a mid-market cross-border carve-out is straightforward. The cost of filing is measured in months and in professional fees. The cost of not filing, if the transaction turns out to be in scope, is measured in unwinding a closed deal. The asymmetry is large enough that the only defensible default is to file where the sector and control tests are arguably met.

Three things to do before signature

The useful output of this case for a deal team is not an opinion about the outcome. It is a set of steps that change the shape of the risk before the agreement is signed.

The first is a security review feasibility assessment and a filing-path design, completed before signature rather than after. That means identifying, entity by entity, what sits inside the transaction perimeter and where it is organised; testing each China entity against the sector list; and testing the post-closing governance architecture against the three control limbs, including reserved matters, board composition and technology decision rights. It also means using the consultation channel in Article 5, which allows a party to consult the office on relevant questions before filing. That channel is in the statute, it is expressly available pre-filing, and it is under-used relative to its value, because a consultation converts an open question about scope into a documented position that the board can rely on.

The second is to make security review a named closing condition, separate from merger control and separate from any other regulatory approval. The drafting points that matter are unglamorous and decisive. The condition should name the review expressly rather than sitting inside a general “regulatory approvals” definition, so that the allocation of risk is unambiguous. The long-stop date should be built from the statutory clock — fifteen plus thirty plus sixty working days, plus an allowance for extension, plus an allowance for the period during which the clock is suspended while supplementary materials are prepared — rather than from a commercial estimate. And the agreement should say in advance who bears what if the outcome is a prohibition: who pays the costs, whether a reverse termination fee applies, and how a remedy negotiation is funded and controlled.

The third is to assess the indivisibility of the China assets before deciding the perimeter. This is the question that decides whether the transaction is a China control transaction in substance.

If the China manufacturing base, the China research team and the China customer relationships can be separated from the rest of the global business without destroying its value, the parties have a genuine choice: exclude China from the perimeter and proceed with the remainder, or include it and accept the review risk. If they cannot be separated — because the product is designed in one place and made in another, because the supply chain is integrated, because the customers are served from a single global platform — then the transaction is, in economic substance, an acquisition of control over the China business, and the review outcome is binary rather than incremental.

Where the answer is indivisibility, the preparation that pays is a remedy package designed in advance: commitments on production capacity, on supply to the domestic market, on continuity of research and development, on data handling, and on the standing of the local management team. Article 9 allows a conditional approval where conditions can eliminate the effect on national security and the parties accept them in writing. A package that has been thought through, costed and cleared internally before filing is a different instrument from a package improvised in the final weeks of a special review.

Underneath those three steps sits a document that costs little and is worth a great deal: a written scope memorandum, dated before signature, recording the sector analysis, the control analysis and the perimeter decision, with the reasons for each. Its purpose is not to persuade a regulator. Its purpose is to be produced, years later, to a board, an auditor, a lender or a court, in answer to the question of whether the company considered the filing obligation at the time. Companies that file have the memorandum and do not need it. Companies that decide not to file need it and often do not have it.

One caution applies to all three steps. They are preparatory, and preparation is not a substitute for the local conversation. The office operates through a process that involves relevant departments and, where appropriate, local government, and the practical questions a deal team will have — whether a particular description of the business is adequate, whether a particular change of governance is material, what a remedy package would need to contain — are answered in that process rather than in the text. The three steps are what allows a company to have that conversation usefully instead of arriving at it with a structure that cannot be changed.

The open questions

Three things remain unknown, and it is worth being explicit about them rather than filling the gaps with inference.

The reason for the prohibition is not public. The Measures do not require the office to publish decisions or reasons, and no decision text has appeared. Any account of the reasoning is therefore an account of someone’s understanding, not of the record. That is unsatisfying for advisers and it is a feature of the regime rather than an oversight: a review system built around national security will not, in any jurisdiction, publish the analysis that supports a prohibition.

Whether the parties can revive the transaction is also unknown. Article 11 allows a party to modify the investment plan during review, with the clock restarting from receipt of the modified plan, and to withdraw, in which case the review terminates. Both mechanisms operate inside an open file. After a prohibition, the route back would be a new filing on a materially different transaction, and neither company has said it intends to take it. Their statements say only that dialogue continues.

What a conditional approval would look like is the third gap, and it is the one that matters most for planning. The Measures say only that conditions must be capable of eliminating the effect on national security, must be accepted in writing by the parties, and must be listed in the decision; Article 13 adds that compliance is supervised and can be verified by requiring documents and conducting on-site inspection. Nothing in the text limits the subject matter of a condition. That breadth is deliberate — a condition has to fit the transaction — and it means the working assumption for a deal team should be that anything capable of addressing the specific concern is on the table, and that the parties are expected to propose it rather than wait to be told.

And the significance of the case is not yet established. This is reported to be the second prohibition under the current framework in 2026; the earlier matter, in April, is reported to have involved a requirement to restore the pre-closing position after completion. Two data points do not make a pattern, and neither decision has been explained. What can be said is that the framework has now been used at the prohibition end of its range twice in a single year, which is a different fact from the framework merely existing.

For a company planning a cross-border carve-out with Chinese operations, the practical reading is narrower than the headlines and more useful. The review is a scope question and a schedule question. Both are answerable before signature, both are answerable with the statute in hand rather than with market rumour, and the answers determine whether the transaction should be signed at all, on what conditions, and with what walk-away rights.

Sources

  • ABB·Update on divestment of ABB Robotics(2026-11-05 声明原文):https://resources.news.e.abb.com/attachments/published/140154/en-US/0A15DBC92198/20261105_Update_on_divestment_of_ABB_Robotics_EN.pdf
  • 软银集团·Update on Acquisition of ABB Ltd’s Robotics Business(2026-11-06):https://group.softbank/en/news/press/20261106
  • ABB·ABB 与软银签约拟出售机器人业务(2025-10-08,交易企业估值 53.75 亿美元):https://new.abb.com/news/zh-CHS/detail/129790/abb-to-divest-robotics-division-to-softbank-group
  • 中国政府网·《外商投资安全审查办法》(发展改革委、商务部令第 37 号,2020-12-19 公布,2021-01-18 施行):https://www.gov.cn/gongbao/content/2021/content_5582626.htm
  • 商务部·国家发展改革委、商务部发布《外商投资安全审查办法》(2020-12-19):https://www.mofcom.gov.cn/xwfb/rcxwfb/art/2020/art_593675063f2d45af8c8ab673f2836ef2.html
  • 澎湃新闻·日本软银以 54 亿美元收购 ABB 机器人的计划受挫(2026-11-05,媒体转述,非监管机关披露):https://www.thepaper.cn/newsDetail_forward_34413380

两家外资之间的交易被中国安全审查拦下:控制权判断究竟看什么

两份声明究竟说了什么

事实很短,而且罕见地干净,因为交易双方在三十六小时之内各自公开了同一件事。

2026 年 11 月 5 日,ABB 发布题为《ABB 机器人业务出售进展更新》的声明。声明记载:2025 年 10 月,ABB 已签署协议,将其机器人业务出售给软银集团,企业估值 53.75 亿美元;该交易尚需获得监管机构批准并满足惯例成交条件;监管审批流程由软银负责推进;ABB 已接到软银通报,中国外商投资安全审查工作机制办公室已禁止软银收购 ABB 机器人在华业务。声明最后一句是,双方正保持密切沟通,并就该交易开展建设性磋商。

2026 年 11 月 6 日,软银发布自己的更新公告,确认已于 2026 年 11 月 4 日收到同一办公室的通知,其收购 ABB 机器人业务在华部分已被禁止,并表示与 ABB 保持密切沟通。

官方的全部记录就是这些。两份声明都没有给出理由,都没有援引具体条款,对”被禁止的是什么”的描述也只到”ABB 机器人在华业务”这一短语为止。没有任何决定文本公开,审查机关本身也没有发布消息。11 月 5 日外交部例行记者会上,发言人被问及此事,答复是建议向中方主管部门询问。

交易本身在 2025 年 10 月的公告里记录得很完整。机器人业务约有七千名员工,2024 年销售收入 23 亿美元,约占 ABB 集团总收入的 7%,运营息税摊销前利润率 12.1%。ABB 对该业务的描述是,与集团其他业务单元之间的商业与技术协同有限。买方在签约时把战略逻辑讲得很直白:机器人是人工智能的物理层。

两份声明的措辞值得再看一眼,因为它们比表面上更谨慎。两家公司都没有写这个决定是终局的,没有写存在何种救济途径,也没有写交易已经终止。两家都用了条文里的那个动词——禁止——并且都补了一句沟通仍在继续。这个组合与办法的架构是吻合的:禁止是特别审查走到终点时可以作出的一种结果,而它是向当事人作出的,不是向抽象的交易作出的。两份声明还恰好印证了两家公司的分工:ABB 的声明写监管审批流程由软银负责推进,软银的公告写自己正在管理这一流程。两家都没有代表监管机关说话,也都没有试图概括理由。

还有一处时间细节,带出的教训很实用。通知在 11 月 4 日到达软银,ABB 在 11 月 5 日发布,软银在 11 月 6 日发布。两家在两个法域上市的公司,在一份既无理由也无救济的通知到达后四十八小时内完成了披露。无论其他事情多么不确定,披露义务本身是确定的:禁止对双方都是应当披露的事件,交易团队应当假定对方的公告会早于任何商定的沟通方案落地。

对审查而言真正要紧的是中国部分的体量,而它相当可观。该业务的全球总部设在上海;2022 年投产的上海超级工厂,是集团全球三大机器人制造基地之一;交付中国客户的机器人,九成以上在中国本地生产;在华研发团队近三百人;中国是其最大的交付市场。

这里必须划一条硬边界,因为绝大多数评论都在这一点上出了偏差。财经媒体把这一决定归因于对物理人工智能业务产生的数据被用于在境外训练模型的顾虑,以及希望产能留在中国境内。这些归因都来自”知情人士”。它们不是监管机关的表态,不在任何一家公司的声明里,也没有任何官方文件予以确认。它们可能是对的,但不能被当作理由来陈述。可以确定地说的要窄得多,但仍然有用:交易被禁止,理由未披露,双方称仍在沟通。

为什么两家外资之间的交易仍要中国批准

面对这样的案子,交易团队的第一个问题总是:中国为什么会介入。卖方是欧洲企业,买方是日本企业,对价从东京流向苏黎世,没有任何中国主体在买或在卖。

第二条一个条款就把这个问题回答了。该条把外商投资定义为外国投资者直接或者间接在境内进行的投资活动,并在第二项明确列出:外国投资者通过并购方式取得境内企业的股权或者资产。审查的对象是位于中国的企业,不是对价的流向。如果一项全球剥离的交易边界内包含一家中国公司,而这家公司拥有工厂、雇有工程师、持有客户合同、在中国产生收入,那么取得对这家公司的控制权,就是对中国的投资——不论支票由谁签。

同一条里的”间接”二字,堵住了最明显的绕道设计。通过新加坡或者荷兰控股公司实施的收购,仍然是外国投资者间接取得下层中国业务的收购。审查会穿透股权链条,看向实际经营的实体。

这也解释了为什么原持有人的国籍无关紧要,以及为什么”这块业务本来就是外资的”不构成理由。判断的不是交易前后是否存在外资,而是对一家位于中国的企业的控制权是否易手,以及这家企业是否落在清单所列行业之内。一家由欧洲工业集团持有的机器人公司,和一家由日本科技集团持有的机器人公司,都是外资;两者的区别在于谁持有控制权,而控制权正是这套办法规范的对象。

还有一处结构性差别,是交易团队经常忽略的:安全审查框架里没有营业额门槛、没有市场份额门槛、也没有微量豁免。范围只由两个变量决定——标的是否经营清单所列行业,以及收购是否取得实际控制权。一笔远低于任何经营者集中申报标准的交易,仍可能是强制性的安全审查申报;一笔顺利通过经营者集中审查的交易,仍可能被禁止。两套制度不共用一个闸门。

于是,交易边界的划定就成了整笔交易里最有分量的起草决定,而它通常在早期被草率定下、在后期才被重新审视。这个阶段有两点值得盯住。

第一点是股权交易与资产交易的区别。第二条覆盖的是取得境内企业的股权或者资产,因此以转移资产而非转移股份的方式剥离一项业务,并不在审查之外,只是被描述成了另一套词。在业务剥离中,两者常常并用:中国运营公司的股份被转让,而知识产权、生产设备或者经销权则依据另外的协议转移。它们属于同一个分析。

第二点是被投资企业的识别。审查附着于持有业务的那个实体,而不附着于业务所使用的品牌。当一项全球业务被组织成总部、若干制造公司、一家研发公司和一家销售公司时,每一个都是潜在的被投资企业,申报材料必须逐一说明它做什么、拥有什么、交割后发生什么变化。为税务目的梳理过集团架构的团队,这项工作已经完成大半;没有梳理过的团队会发现,申报材料的第一稿,其实就是架构图的第一稿。

控制权测试:回到条文本身

第四条是操作条款,它的起草方式值得逐字读,因为这是框架中交易团队在签约前真正能用的部分。

该条先列行业。第一类是国防,包括军工、军工配套,以及在军事设施和军工设施周边地域投资;这一类不附加控制权要求,任何投资都需申报。第二类更长,且附有控制权要求,涵盖关系国家安全的重要农产品、重要能源和资源、重大装备制造、重要基础设施、重要运输服务、重要文化产品与服务、重要信息技术和互联网产品与服务、重要金融服务、关键技术以及其他重要领域,且以取得所投资企业的实际控制权为条件。

实际控制权随后由三项情形界定。第一项是持有企业 50% 以上股权。第二项是持股不足 50%,但享有的表决权能够对董事会、股东会或者股东大会的决议产生重大影响。第三项是其他导致外国投资者能够对企业的经营决策、人事、财务、技术产生重大影响的情形。

这个定义有三处需要强调。

第一,三项是并列的。条文说实际控制权”包括下列情形”,每一项各自独立成立。持股 51% 就是第一项意义上的控制,无论股东协议如何约定董事席位;持股 30% 但对技术路线拥有否决权,可能构成第三项意义上的控制。这里没有权重,也没有一个可以让人停止思考的安全比例。

第二,第二项与第三项是定性判断,取决于事实。”产生重大影响”与”能够产生重大影响”是标准,不是规则,适用的对象是实际的治理架构:董事会构成、表决机制与出席比例、保留事项、总经理人选、预算控制权、研发议程控制权。一个让收购方取得多数董事席位并有权任命首席技术官的结构,不会因为经济权益只有四成而变得无害。

第三,技术被明确写进第三项,与经营决策、人事、财务并列。对一家工业机器人企业来说,这不是可有可无的补充。这项业务的价值在于控制器、伺服系统、机器视觉、工艺软件,以及产线上生成的数据。控制了技术路线图的收购方,即便股权比例不高,也对技术有重大影响——条文就是这么写的。

把三项放在一起读,而不分开读,还能得出两个推论。

第一个推论是,表决权按它实际如何运作来评价,而不是按名义上如何分配。能够移动投票权的安排——表决权信托、不可撤销的委托、约定一方按另一方指示投票的股东协议、一致行动安排——都属于判断范围,因为第二项问的是表决权能起什么作用,而不是股票凭证在谁手上。收购方持股三成五、同时控制另外两成的投票权,从条文表面看,就是一个其表决权可能对决议产生重大影响的结构。

第二个推论是,这套框架里的实际控制权,与会计意义上的控制不是同一个概念,两者不应在董事会材料里混用。合并报表看的是对被投资方相关活动的权力与可变回报的风险敞口;安全审查看的是第四条的三项情形,其中一项纯属数字,另两项是定性判断。得出”不必合并”的结论,并不等于得出”没有取得实际控制权”的结论。两套测试服务于不同目的,有时会分道扬镳。

与控制权测试相对应的,是救济菜单。第九条规定,申报的交易影响国家安全的,应当作出禁止投资的决定;但通过附加条件能够消除对国家安全的影响,且当事人书面承诺接受附加条件的,可以作出附条件通过的决定,并在决定中列明附加条件。第十三条规定,审查决定由工作机制办公室会同有关部门、地方人民政府监督实施;对附条件通过的,可以要求提供证明材料、现场检查等方式核实附加条件的实施情况。

这条通道的存在,是整套框架里最实用的一个事实。附条件通过是一个谈出来的结果,而谈判只在案卷还开着的时候可能。等禁止决定已经作出,才发现救济方案还没准备的人,手上已经没有可交换的东西。

两个闸门,两套标准

围绕本案最常见的分析错误,是把反垄断结果与安全审查结果当成同一流程的两个阶段,并认为前者通过之后,后者只是形式。

两者并不相关。经营者集中审查依据反垄断法,由国家市场监督管理总局主管,它问的是这项集中是否会排除、限制竞争:相关市场界定、份额、集中度、进入壁垒、买方力量、效率抗辩。它的结果是通过、附加限制性条件通过或者禁止,程序上区分简易案件与普通案件。

安全审查依据外商投资法与国家安全法,由设在国家发展改革委、由国家发展改革委与商务部牵头的工作机制办公室主管。它问的是这项投资是否影响或者可能影响国家安全。它的结果是不需要审查、通过、附条件通过或者禁止。它的行业清单与控制权测试,在反垄断分析里没有对应物。

时间表也是分开的,形状还不一样。按第七条,办公室自收到符合规定的材料之日起十五个工作日内,作出是否需要进行安全审查的决定,作出决定前当事人不得实施投资。按第八条,决定审查的,应当自决定之日起三十个工作日内完成一般审查,审查期间同样不得实施投资。按第九条,启动特别审查的,应当自启动之日起六十个工作日内完成,特殊情况下可以延长,延长须书面通知。按第十条,当事人补充提供材料的时间,不计入上述审查期限。

因为两个问题不同,支撑它们的证据也不同。反垄断案卷由市场数据搭建:份额、产能、投标记录、客户转换、价格。安全审查案卷由业务描述与交割后变化的描述搭建:投资者是谁,标的是做什么的,持有哪些技术,数据放在哪里,谁将进入董事会,以及投资者打算在更长的时间里如何处置这些资产。一家公司如果因为”是同一笔交易”就把安全审查交给反垄断团队去办,通常交出的会是一份技术上完整、实质上跑偏的材料。

公开报道显示,该交易已于 2026 年早些时候通过中国经营者集中审查,同年亦获欧盟放行。就已公开的记录看,这与上述分析完全一致:一份干净的反垄断案卷,对安全审查案卷什么也说明不了;两个决定由不同机关、针对不同法律问题分别作出。

对交易团队来说,操作层面的结论很直接:排两张时间表,而不是一张。经营者集中那张通常更短、更可预测;安全审查那张才是决定最后交割日的,而且里面有一项在反垄断流程中没有对应物的安排——准备补充材料期间时钟暂停,而暂停多久,当事人只能间接控制。

禁止决定做了什么,没做什么

禁止的后果规定在第十二条,比这个词听起来更重。

办公室作出禁止投资决定的,当事人不得实施投资;已经实施的,应当限期处分股权或者资产以及采取其他必要措施,恢复到投资实施前的状态,消除对国家安全的影响。

这是一项把已完成的交易拆回去的权力。它不是处罚,不是罚款,而是恢复原状的命令。对一项包含工厂、员工、客户合同与一体化供应链的业务剥离来说,恢复原状不是把股份转回去那么简单。它要求拆分系统、重签合同、重建独立的管理架构,并且在实践中把交易成本付两遍。

框架里还有三条,应当与第十二条放在一起读,因为它们界定了”范围判断错了”的代价。

第十六条管的是未申报。属于申报范围的投资,当事人未申报即实施的,责令限期申报;拒不申报的,责令限期处分股权或者资产,恢复到投资实施前的状态。也就是说,风险并不取决于审查最终得出否定结论,而附着于”没申报就交割”这个动作本身。

第十四条管的是方案变更。办公室作出不需要审查或者通过审查的决定后,当事人变更投资方案,影响或者可能影响国家安全的,应当重新申报。批准附于一项被描述过的交易,而不附于一段关系。

第十九条管的是记录。出现第十六条、第十七条、第十八条情形的,应当作为不良信用记录纳入国家有关信用信息系统,并按规定实施联合惩戒。

这套设计里还有一处值得单独提,因为实际损害恰恰落在那里。命令是针对投资当事人的,后果却不止于当事人。一项经营性业务的剥离通常伴随融资,融资文件里会有关于监管批准的陈述与保证;恢复原状的命令会把那些陈述变成违约。被剥离业务的客户此前已被告知控制权变更,并且已经同意,或者已经解约重签;恢复原状意味着整个流程要反向再做一遍,而对手方此时已经知道这笔交易失败过一次。员工也已经被转移或者重签合同。这些主体都不是审查的当事人,也都不在审查的保护范围内。

这就是分析必须在签约前完成、而不是签约后补做的原因。等到命令作出,代价已经不是履行命令的法律成本,而是向银行、客户与员工——他们曾被告知这笔交易已通过所有适用于它的闸门——收回安排的成本。

把这些放在一起,一笔中型跨境剥离的账其实很好算。申报的成本以月和专业费用计;该申报而未申报、且事后被认定属于范围的成本,以拆解一笔已交割交易计。这个不对称足够大,以至于唯一站得住脚的默认做法是:只要行业与控制权两项测试存在可争辩的成立空间,就申报。

签约之前该做的三件事

这个案子对交易团队真正有用的产出,不是对结果的评论,而是一组能在协议签署前改变风险形状的动作。

第一件,是在签约前完成安全审查可行性评估与申报路径设计,而不是在签约后。这意味着逐主体梳理交易边界内有什么、注册在哪里;把每一家中国主体对照行业清单过一遍;把交割后的治理架构对照三项控制权情形过一遍,包括保留事项、董事会构成与技术决策权。这也意味着用上第五条给出的咨询通道——当事人向工作机制办公室申报前,可以就有关问题向其咨询。这条通道写在办法里,明确可用于申报前,而相对于它的价值,用的人偏少。一次咨询,能把一个关于范围的开放问题,转成董事会可以依赖的有记录立场。

第二件,是把安全审查写成一项具名的交割先决条件,与经营者集中、与其他任何监管批准分开列明。真正起作用的起草细节都不起眼,却具有决定性。条件应当直接点名安全审查,而不是塞进一个笼统的”监管批准”定义里,这样风险分配才不含糊。最后交割日应当按法定时钟搭建——十五加三十加六十个工作日,加上延长的余量,加上准备补充材料期间时钟暂停的余量——而不是按商业估算。协议还应当事先写清:如果结果是禁止,谁承担什么,费用由谁支付,是否适用反向终止费,救济方案的谈判由谁主导、如何出资。

第三件,是在确定交易边界之前,先评估中国资产的不可分割性。这个问题决定这笔交易在实质上是不是一笔中国控制权交易。

如果中国的制造基地、中国的研发团队与中国的客户关系,可以从全球业务中分离出去而不摧毁其价值,双方就有真正的选择:把中国排除在边界之外、推进其余部分,或者把中国包进来、承担审查风险。如果分不开——因为产品在一地设计、在另一地制造,因为供应链是一体的,因为客户由同一个全球平台服务——那么这笔交易在经济实质上就是取得对中国业务的控制权,审查结果是非此即彼,而不是可以逐步退让的。

答案若是”分不开”,真正划算的准备是提前设计一套救济方案:关于产能的承诺、关于境内市场供应的承诺、关于研发连续性的承诺、关于数据处理的承诺、关于本地管理团队地位的承诺。第九条允许在附加条件能够消除对国家安全影响、且当事人书面承诺接受的情况下作出附条件通过。一套在申报前已经想清楚、算过成本、内部通过的方案,与一套在特别审查最后几周临时拼出来的方案,是完全不同的两件工具。

这三件事之下,还有一份成本很低、价值很高的文件:一份落款在签约之前的范围备忘,记录行业分析、控制权分析与边界决定,并写明各自的理由。它的用途不是说服监管机关,而是在若干年之后,被交给董事会、审计师、贷款人或者法院,用来回答”公司在当时是否考虑过申报义务”这个问题。申报的公司有这份备忘,但用不上;决定不申报的公司用得上,却往往没有。

三件事都有一条共同的注意点。它们是准备工作,而准备不能替代当面的沟通。办公室运行的流程会涉及有关部门,并在适当情形下涉及地方政府;交易团队真正会有的那些问题——对业务的某一种描述是否足够、治理架构的某一处变化是否重大、一套救济方案需要包含什么——是在这个流程里得到回答的,而不是在条文里。这三件事的意义,是让公司能够有效地进行那场对话,而不是带着一个已经改不动的结构走进对话。

尚未回答的问题

有三件事仍属未知,把它们说清楚,比用推断去填补空白更有价值。

禁止的理由不公开。办法并未要求办公室公布决定或理由,也没有任何决定文本出现。因此,任何关于推理过程的叙述,都是某个人理解的叙述,而不是记录的叙述。这对顾问来说不舒服,但它是这套制度的一个特征,而不是疏漏:任何法域里,围绕国家安全搭建的审查体系都不会公布支撑禁止决定的分析。

双方能否让这笔交易复活,同样未知。第十一条允许当事人在审查期间修改投资方案,审查期限自收到修改后方案之日起重新计算;也允许撤销投资,撤销则审查终止。两种机制都发生在案卷还开着的时候。禁止决定之后,回来的路只能是以一项实质不同的交易重新申报,而两家公司都没有表示打算这么做。它们的声明只说沟通在继续。

第三处空白,是附条件通过会长成什么样,而它对规划最要紧。办法只说了三件事:附加条件要能够消除对国家安全的影响,要经当事人书面承诺接受,要在决定中列明;第十三条又补充,附加条件由办公室会同有关部门、地方人民政府监督实施,可以通过要求提供证明材料、现场检查的方式核实。条文没有对附加条件的题材作任何限制。这种宽是有意的——条件必须贴合具体的交易——它意味着交易团队的作业假设应当是:凡是能够回应那项具体关切的手段,都摆在桌上,而且当事人被期待主动提出,而不是等着被告知。

这个案子的意义也还没有定型。据报道,这是 2026 年现行框架下的第二起禁止;此前 4 月那一起,据报道涉及在交割后被要求恢复原状。两个数据点不构成规律,两个决定都没有被解释。可以说的只是:这套框架在一年之内,两次被用到了它区间的禁止一端——这与”框架存在”本身,是两个不同的事实。

对正在筹划带中国业务的跨境剥离的企业来说,务实的读法比标题窄,也比标题有用。审查是一个范围问题加一个时间表问题。两者都能在签约前回答,都能拿着条文而不是拿着市场传闻回答;而答案决定这笔交易该不该签、在什么条件下签、以及退出权怎么写。

来源

  • ABB·Update on divestment of ABB Robotics(2026-11-05 声明原文):https://resources.news.e.abb.com/attachments/published/140154/en-US/0A15DBC92198/20261105_Update_on_divestment_of_ABB_Robotics_EN.pdf
  • 软银集团·Update on Acquisition of ABB Ltd’s Robotics Business(2026-11-06):https://group.softbank/en/news/press/20261106
  • ABB·ABB 与软银签约拟出售机器人业务(2025-10-08,交易企业估值 53.75 亿美元):https://new.abb.com/news/zh-CHS/detail/129790/abb-to-divest-robotics-division-to-softbank-group
  • 中国政府网·《外商投资安全审查办法》(发展改革委、商务部令第 37 号,2020-12-19 公布,2021-01-18 施行):https://www.gov.cn/gongbao/content/2021/content_5582626.htm
  • 商务部·国家发展改革委、商务部发布《外商投资安全审查办法》(2020-12-19):https://www.mofcom.gov.cn/xwfb/rcxwfb/art/2020/art_593675063f2d45af8c8ab673f2836ef2.html
  • 澎湃新闻·日本软银以 54 亿美元收购 ABB 机器人的计划受挫(2026-11-05,媒体转述,非监管机关披露):https://www.thepaper.cn/newsDetail_forward_34413380