M&A Chan

陈明键 James Chen。中国买方那一侧的并购谈判桌，我坐了三十年。这里写从这一侧看到的交易。东建国际（HKEX: 00329）。

[M&A Chan @OCIP_MAChen]

标的去年净利润 509 万元。业绩承诺第一年 4000 万元。

菲林格尔（603226）9 月 18 日盘后公告，拟以约 2.14 亿元现金、分期付款，收购南京克锐斯自动化科技 51% 股权。菲林格尔主业是木地板和全屋定制家居，克锐斯做的是 PCB 压合工序的自动化设备。

业绩承诺：克锐斯 2026 至 2029 年扣非后归母净利润分别不低于 4000 万、4500 万、5000 万、5000 万元。

标的的实际数据（未审计）：2025 年营收 5387.46 万元、净利润 509.22 万元；2026 年上半年营收 4494.90 万元、净利润 972.10 万元。公告预计本次交易形成商誉约 2 亿元，占净资产比例可能超过 20%。

买方自己这半年：营收 9629.76 万元、同比降 30.42%，归母净利润 -3708.64 万元，经营性现金流 -5393.90 万元。

4000 万对 509 万，接近 8 倍；上半年已经做了 972 万，下半年还要再做三千万出头才够。这两个数是我按公告数字算的，不是公告写的。

这里有一条卖方应该记住的机制：业绩承诺不是历史业绩的延长线，是从估值倒推回来的。想要这个估值，就得承诺这个利润。

所以你在谈判桌上多争到的那部分价格，会以承诺的形式原样回到你身上，而且后面跟着补偿条款。争估值和签承诺是同一件事的两面，不能只谈前一半。

你愿意为多出来的那部分估值，签几年的承诺？

[M&A Chan @OCIP_MAChen]

同一个价格，换了一个买家。

奥联电子（300585）9 月 23 日公告：控股股东广西瑞盈资产管理拟把 16.20% 股份转让给骏图天泽（无锡）投资合伙企业，另把 5% 转让给衢州市信安广合产业并购股权基金，每股均为 19.29 元，合计约 7 亿元。实际控制��拟由钱明飞变更为陈扬。9 月 24 日复牌，还要经深交所合规性确认、办理过户。

上一次是 2025 年 12 月 26 日，同一个卖方和天津潮成创新签约：3266.67 万股、19.09%、对价 6.3 亿元。潮成创新没有按约支付首期对价 1.26 亿元，被认定违约，协议在 2026 年 2 月解除。

6.3 亿元除以 3266.67 万股，上一次的每股价格也是约 19.29 元。这个数是我算的，公告没有这样写。

价格一分没变，变的是价格以外的东西：

骏图天泽承诺所获股份 60 个月内不转让、36 个月内不质押；衢州信安广合锁定 36 个月；卖方瑞盈承诺 36 个月内不转让剩余股份，也不再谋求控制权。

据 21 世纪经济报道，骏图天泽的资金 92.59% 是自有资金，约 3960 万元拟申请并购贷款；陈扬方面出资约 2.71 亿元，公告称已经全部到位。

上一次的教训是：价格谈成了，钱没到。这一次，买方在公告里先回答了「钱在哪」。

对卖方来说，第二次卖同一样东西，能多要的往往不是价格，是确定性：资金到位的证明、锁定期，以及对方违约时你手里有什么。

如果你的上一个买家没付首期款，第二个买家你会先要什么？

[M&A Chan @OCIP_MAChen]

对价现金 51%，股份 49%。那 49% 的作用不在付款。

星帅尔（002860）9 月 21 日晚间公告，拟收购东莞市梵宇自动化科技和东莞市湘鹰新材料科技两家公司 100% 股权，基础整体合计估值暂定 8.4 亿元。支付方式是现金收购 51%、发行股份收购 49%，同时募集配套资金。标的做 PCB 刀具设备和半成品耗材，实际控制人陈勇。9 月 22 日开市起停牌，预计不超过 10 个交易日内披露方案。公告称预计不构成重大资产重组。

业绩承诺：2026 至 2028 年扣非净利润分别不低于 5500 万元、7500 万元、8000 万元，三年累计不低于 2.1 亿元。

把两个数放一起。股份对价按 8.4 亿元的 49% 算，约 4.12 亿元；三年承诺利润合计 2.1 亿元。股份这一半，接近承诺总额的两倍。

这个比例通常不是随手切的。现金给出去就收不回来，股份给出去还在买方自己的登记簿上，锁定期、解锁节奏、补偿时怎么回收，都能写进协议。所以带业绩承诺的交易里，现金和股份的比例，往往先照「万一没做到要拿回多少」排，再回头看卖方想拿多少现金。

具体的补偿条款还没有。公司在停牌，方案要等不超过 10 个交易日。

卖方这边对应的动作只有一个：谈现金股份比例的时候，把补偿的上限、顺序、以及先扣股份还是先扣现金一起谈完。分开谈，第二次坐下来时你已经没有筹码。

现金和股份，你会怎么切这个比例？

[M&A Chan @OCIP_MAChen]

@vesmallu "出海"说的是去海外市场卖东西、找客户，跟谁家模型更强关系不大。国内团队做应用、做硬件、做服务，用国外的模型一样能做，赚的是海外客户的钱。还有一条路是直接买一家当地公司，渠道和客户一次到手，代价是贵，整合也难。

[M&A Chan @OCIP_MAChen]

@youngstockuser 买一只股票，这十个够用了。买整家公司的人还会多翻一样：10-K 附件里的重大合同，专看控制权变更条款。公司一换主人，大客户能不能解约、银行能不能提前收贷，往往就写在那里。估值再便宜，这一条踩中了，买到的东西可能跟你以为的不一样。

[M&A Chan @OCIP_MAChen]

@LabSpeculation 算得很细。从买方角度补一句：B 类一卖就自动转 A，意思是控制权根本不在市场上流通。二级市场买再多，也只买到分红，买不到话语权。想谈 SpaceX 的控制权，对面只有一个人。这和"持股过半却管不住公司"正好是两个极端：这里 46% 就是全部控制。

[M&A Chan @OCIP_MAChen]

@ChineseWSJ 借钱一贵，PE 的出价就得往下压。这时候卖方会回头看那些不靠杠杆的产业买家：价格未必最高，但融资这一项不会在交割前出问题。对手里有现金的战略买方，这是一段不用硬拼价格的窗口。

[M&A Chan @OCIP_MAChen]

第（4）条最值得看。同一年，安永按"宣布"口径统计的中企海外并购是436亿美元，增长近四成；公报这边是153.4亿，降了四成。两个数差了将近三倍（这是我算的）。一个统计宣布，一个统计实际完成。口径不同是一部分原因，签了没落地、最后没做成的交易也在这个差里。签约到付款之间死掉一批，在买方桌上是常态。

[M&A Chan @OCIP_MAChen]

股份不发了，配套募资取消了，100% 砍成 60%。

观想科技（301213）9 月 21 日公告，收购辽晶电子的方案做了重大调整。原方案是发行股份及支付现金购买 100% 股权，并募集配套资金。新方案是全现金购买 60%，不涉及发行股份，配套募资取消。交易总价款 4.82 亿元。本次构成重大资产重组及关联交易，不构成重组上市。

业绩承诺：辽晶电子 2026 至 2028 年净利润分别不低于 6900 万元、7300 万元、7700 万元，三年累计 2.19 亿元。按 4.82 亿元对应 60% 倒推，100% 作价约 8.03 亿元。这个 8.03 亿是我算的，公告给的是 4.82 亿和 60%。

三件事一起改，指向同一个东西：时间。发行股份购买资产要走交易所审核和证监会注册这一道，配套募资再加一道。改成全现金买 60%，这两道就都不用走了。

代价很具体。4.82 亿元现金实打实出去，而原本能靠配��募资补回来的那一块，现在没有了。剩下的 40% 留在原股东手里，以后要么再买一次，要么长期共处。公告没有说调整的理由。

���卖方来说，这个案子的信息量在于：买方愿意为「快」付多少。这笔钱不体现在溢价上，体现在它自己的现金头寸、股权结构的完整度，和一次以后还要重新坐下来的谈判。

买方提出把对价从股份换成现金、同时把比例从 100% 砍到 60%，你会先问什么？

[M&A Chan @OCIP_MAChen]

@pand_lin 同时！

[M&A Chan @OCIP_MAChen]

标的净资产是负的 1.5 亿。所以钱不能付给卖方。

逸豪新材（301176）9 月 22 日披露预案，拟向特定对象发行 A 股募集资金不超过 18 亿元，扣除发行费用后全部用于增资收购铭丰电子不低于 78% 的股权。公告用的词是「增资收购」，不是「受让」。

铭丰电子做高性能铜箔，电子电路铜箔和锂电铜箔。按预案披露的 2026 年 6 月末未经审计数：资产 31.48 亿元，负债 32.98 亿元，净资产负 1.50 亿元，资产负债率 104.77%。上半年营收 11.28 亿元，净利润负 0.78 亿元。设计年产能 7 万吨，整体开工率不足 50%。

这张表把结构定死了。

如果走受让老股：18 亿元进的是现有股东的口袋，铭丰的负债还是 32.98 亿元，停着的产能还是停着，买方接手的是一家原封不动的公司，自己还多背一笔支出。

增资是把钱打进标的公司本身。资本金上去，负债率下来，闲置的线才有钱重启。

对卖方来说，这里有一条不太舒服的机制：当标的净资产为负，交易结构就轮不到双方偏好来定，资产负债表先把它定了。你想套现，对方给不了，因为钱一旦给了你，这笔交易本身就不成立。你能拿到的是稀释之后的剩余股权，和一家被修好的公司。

真正值得等的数字也不在那个 78%。在这 18 亿元里，多少去还债，多少去开机器。这个拆分预案里还没有。前者买的是一张干净的报表，后者买的是产量。

如果你的公司净资产为负，你愿意接受「钱进公司不进你口袋」这个前提吗？

[M&A Chan @OCIP_MAChen]

I stopped trying to be interesting and started being useful.

Nobody follows an account. They follow a feed that tells them something they didn't know.

I write about cross-border M&A: deal structures, who's buying whom, and where the money actually moves. That's the whole account. One niche, no compromise.

[M&A Chan @OCIP_MAChen]

@drfeifei "The threat is within ourselves" is the right frame — and it's also the reason AI governance can't be solved by the labs themselves. If the risk is human, then the oversight has to sit with someone who isn't selling the model.

[M&A Chan @OCIP_MAChen]

Anthropic is doing something no AI lab has done: it's putting its safety evaluators inside its own building.

Anthropic and Accenture are creating an embedded evaluation team. They'll sit alongside employees, watch model-training decisions in real time, and issue independent safety assessments.

Each side commits at least $1 billion over five years.

Why this matters more than it sounds: last week, the "we must pace the frontier" calls became an antitrust lawsuit — the argument being that synchronized safety warnings were really incumbents locking the door. The critique wasn't that safety is fake. It was that self-regulation is unverifiable.

Embedded evaluators are an attempt to make it verifiable.

The context: Anthropic's annualized revenue is expected to top $100 billion this year, up from $65 billion in July, targeting $200 billion by 2028 — while it prepares a Nasdaq listing at a reported $2 trillion.

So here's the tension worth naming: the company about to ask public markets for the largest raise in history is also building the institution that will grade its own safety.

Independent verification only means something if the verifier can say no.

$1 billion a year is the budget. The question is what happens the first time they find something they don't like.

#Anthropic #AI #Governance

[M&A Chan @OCIP_MAChen]

There's a $300 billion layer in the AI buildout that almost nobody is pricing.

It's called a residual-value guarantee.

How it works:

A tech company promises a chip or data center will retain a minimum future value
A special-purpose vehicle owns the asset and issues the debt
The guarantor never fully books the liability
In under a year, big tech has issued up to $300 billion of these commitments while recording little of it on balance sheets.

Morgan Stanley counted $3.1 TRILLION in off-balance-sheet commitments and credit support across just seven hyperscalers and chipmakers.

Meta deployed it first on a data center project. Broadcom used it in Anthropic's chip financing. Nvidia has offered similar support to OpenAI.

Bankers call it "balance-sheet efficient." That phrase tells you both the appeal and the risk.

The mechanism only works if utilization holds. If AI infrastructure demand disappoints — oversupply, slower adoption, business models that never turn — those guarantees convert from footnotes into real liabilities, all at once.

Read the structure honestly: the AI buildout is being financed on assumptions about demand that hasn't arrived yet, using vehicles designed so the risk doesn't show up where you'd look for it.

Watch the disclosure changes in the next set of filings. That's where this story moves.

#AI #Finance #Credit

[M&A Chan @OCIP_MAChen]

Alibaba just answered the chip question — with hardware, not a press release.

At its Yunqi conference, Alibaba's T-Head division unveiled Zhenwu V900.

3x the compute of the previous M890
216 GB memory, 1,200 GB/s inter-chip bandwidth
Native FP8 and FP4 support
Up to 500,000 chips in a single cluster
Mass production: Q1 2027
And the bigger number: Alibaba Cloud is targeting over 20 gigawatts of global data center capacity by 2032.

The part that matters most isn't the chip. It's that M890 supernodes are already live, already running Qwen3.8 and Kimi K3 at 2-trillion-parameter scale. Alibaba says 650+ enterprise customers are on the Zhenwu line.

Alibaba's stock rose over 4% in Hong Kong on the announcement.

Here's the strategic read: the US export controls didn't stop China's AI buildout — they forced it vertical. Alibaba now designs the chip, the interconnect, the network card, the SSD controller, the supernode, the cloud, AND the model.

That's not a workaround. That's a full stack.

When you can't buy the best chip, you build the whole factory.

#Alibaba #AI #Semiconductors

[M&A Chan @OCIP_MAChen]

The biggest Hollywood deal in a decade just cleared its last hurdle — and the price was a list of promises.

Paramount settled with California and 11 other states over its $110 billion Warner Bros. Discovery acquisition. WBD jumped 11%.

What Paramount gave up to avoid selling CNN:

At least 30 films a year for two years, then 32 a year for three
At least 4 indie films a year, and 20% must be blockbusters — miss the quota, pay $30M per film
$300M MORE per year in US production spending, $1.5B total
No theater-rate increases for three years
No selling either studio lot
An independent news editorial board for CBS and CNN
The company also agreed not to cut certain jobs for five years and to pay $17.5M into the Writers Guild health fund.

Closing expected early October. And note what it avoids: a $7 million per day ticking fee that would have started October 1.

Read the structure of this settlement. The remedy wasn't divestiture — it was behavioral commitments, monitored by a board, backed by financial penalties.

That's the new template for media consolidation: you don't have to sell assets, you just have to promise to behave, with a price tag attached to breaking your word.

$6B in promised synergies. $80B in debt. And a compliance department that now has real teeth.

#Paramount #WarnerBros #MergersAndAcquisitions

[M&A Chan @OCIP_MAChen]

A Hong Kong-listed shell just bought "an AI compute company" for HK$3.19 billion. Read the structure.

Wah Da Holdings (https://t.co/btAJ2ncKhV) is issuing 1.18 billion new shares at HK$2.70 to acquire Apex Realm Ventures.

The price signals:

10.3% DISCOUNT to market. In a control transaction, that's unusual.
335% PREMIUM to the company's audited net asset value per share of HK$0.62.
Now the part that matters. What does it actually own?

Apex Realm → holds 99% of Haikou Guangzhi → which holds 50.10% of the Chinese operating company (Terminus Shanghai Intelligent Technology).

Net indirect economic interest: roughly 49.6%. Not control. A minority stake, wrapped in three layers of holding companies, valued by an appraiser at HK$10.49 billion for the operating company.

And the operating company's financials? 2025 revenue of ¥239 million, with a net loss of ¥84 million.

So: 3.19 billion Hong Kong dollars for a 49.6% indirect stake in a loss-making company, paid in discounted new shares, at a 335% premium to book.

Bought for the AI narrative, priced on a valuation report, paid with paper.

Watch the lock-up. Two years. That tells you what the seller thinks about the price.

#MergersAndAcquisitions #HongKong #AI

[M&A Chan @OCIP_MAChen]

Nuclear medicine just had its consolidating deal — and it's a supply-chain play, not a science play.

Telix is acquiring Germany's ITM (Isotope Technologies Munich) for $1.65 billion.

The structure:

$1.25B in Telix stock (Nasdaq ADRs at $11.84/share, escrowed)
Telix assumes $302M of net debt
Up to $700M contingent — tied to ITM-11's FDA approvals and 2030 global sales
Now here's why this is really interesting:

ITM is the world's only commercial-scale supplier of lutetium-177. Also actinium-225 and terbium-161. If you make targeted radioligand therapies, you buy your isotope from ITM — or you don't make them at all.

Telix already had the drug pipeline. What it didn't have was guaranteed access to the isotope.

That's what $1.65 billion buys: a chokehold on the raw material in a market headed to $41 billion by 2034.

This is what vertical integration looks like in pharma. You don't buy the competitor. You buy the thing the competitor can't operate without.

#Telix #Pharma #MergersAndAcquisitions

[M&A Chan @OCIP_MAChen]

The safety story just got subpoenaed.

Anthropic, OpenAI, SpaceXAI, and Google are now facing an antitrust suit alleging that their synchronized calls to "pace the frontier" amount to collusion — incumbents using safety as cover to lock the door behind them.

Follow the sequence:

1、Three of the biggest labs publicly call for a coordinated slowdown.
2、They propose a third-party standards body — a FINRA for AI.
3、Then the market realized: in every industry, "let's agree to go slower" is the oldest cartel playbook there is.

This was obvious weeks ago. When the three companies with the largest lead all agree the race should slow down, the question isn't whether they believe it. It's who benefits.

And now the courts get to decide.

The uncomfortable irony: if safety arguments can be litigated as anticompetitive conduct, every future lab has a reason to never say a word about risk again.

#AI #Antitrust #OpenAI