Steven Jia

Investor · Entrepreneur · Economist | Frontier Tech × Global Macro × Capital Markets

[Steven Jia @JUNXINJIA]

PhD, Visiting Professor, and VC Founder.

I’ve invested in hundreds of companies across AI, embodied intelligence, robotics, biotech, advanced manufacturing, and semiconductors. My research covers frontier tech—from aerospace and quantum computing to new energy, advanced materials, and biopharma.

Beyond deep tech, I closely track geopolitics, global conflicts, macro trends, and global capital markets.

My core lens: I cut through the tech hype, stress-test commercial viability, and analyze how geopolitics and macro shape valuations, supply chains, and venture returns. I separate what’s actually technically feasible from overblown market BS.

Always open to sharp debates with founders, researchers, and fellow investors.

Note: To avoid spam and scams, I don't take cold calls, and I only engage in DMs after basic verification. Thanks for understanding.

[Steven Jia @JUNXINJIA]

Anthropic在IPO前夕甩出Claude Sheets、Whiteboards和Data这套Office组合拳，微软盘中一度跌超3%。市场这个反应不算没道理，但得看清楚它到底在打哪儿。

Claude Sheets现在什么水平？Anthropic自己的产品负责人说得很直白——“相当基础”，不面向财务建模这种复杂活儿。所以它压根不是想做一个更好的Excel，它想干的是让你不用打开Excel就能把活干了。你从提示词出发生成表格、改公式、导出.xlsx，工作流的起点从Office应用挪到了Claude的对话框里。Excel就从“干活的地方”变成了“一个文件格式”。这才是微软股价承压的真实逻辑——威胁不在功能替代，在入口迁移。

但微软手里也不是没牌。第一，Anthropic同时是微软的大客户，承诺买300亿美元Azure算力，微软还打算在Anthropic的IPO里私人配售10亿美元股份。竞争归竞争，Anthropic越成功，微软的云收入和股权收益也跟着涨，这关系挺微妙的。第二，微软之前已经允许Claude以插件形式进驻Word、Excel、PowerPoint和Outlook。你要是不让它进，可能招来反垄断审视；让它进，反倒加固了Office作为工作平台的不可替代性。微软真正的护城河从来不是某个功能，而是企业客户的数据、权限和工作流惯性，这些东西不是换个对话框就能搬走的。

所以一句话：Claude Sheets的beta版好不好用根本不重要，真正值得盯的是企业客户会不会开始把“从零生成分析”这件事从Excel里挪到Claude里。会，按席位收费的模式才会松动；不会，这就是一次IPO前的产品秀。微软跌的那3%，是情绪，不是基本面。

以��仅代表个人对公开信息的理解，不构成任何投资建议。

[Steven Jia @JUNXINJIA]

Helen lit Troy’s decade‑long war. Now this reported scandal hands Nvidia its very own Trojan‑horse moment. Inside: the board’s rumored leadership shake‑up. The market hauled it right in, and the stock got walloped. Nothing’s set in stone. Just my take — not investment advice.

[Steven Jia @JUNXINJIA]

光通信这波反弹，底层逻辑到底在哪？中美产业结构的两侧卡位

光通信这波过山车，说白了就是同一件事被市场用两套故事先后讲了一遍。11月5号那天跌得那么惨，表面上看是中期选举之后大家开始担心数据中心监管收紧，但真正把跌幅放大的，是筹���太挤了。光模块板块三季度跌了快46%，中际旭创股价从年内高点回撤六成多，可同期净利润涨了170%多。业绩在涨，股价在跌，这个背离本身就在攒反弹的劲儿。到了11号暴力拉升，导火索是微软拉着几家科技公司去白宫游说，想让政府放弃对中国光收发器的进口限制，再加上砍单传闻被公司自己出来辟谣。所以你看，跌的是“脱钩真要发生”的恐惧，涨的是“脱钩短期根本做不到”的现实。

美国那边其实自己先把底牌亮了。云厂商手里攥着一堆AI芯片，可没有光模块把这些芯片连起来，芯片就是一堆摆设。摩根士丹利说中国企业在800G和1.6T高速光模块上占了75%到80%的份额，短期这个缺口填不上。而且禁令的时间窗口设在3.2T那一代，3.2T要到2028年才放量，政策的节奏和产业的节奏根本对不上。这个错位一出来，压在估值上的政策折价自然就松了。

具体到公司，大家卡的位置不一样。Coherent是把磷化铟产能提前翻倍，还拿下了AI巨头五年200亿美元的长协，它卖的不是模块，是激光器和光学整合平台，下游谁做模块都得找它。Marvell赌的是并购Lumentum之后，把ASIC、DSP、硅光子和激光器串成一条链，从芯片供应商变成光互连系统供应商。AAOI是同一个逻辑的轻量版，手里有自有激光器产能，1.6T订单多到供不上货。POET最虚，它的Optical Interposer平台还没被量产验证过，反弹的弹性就来自“万一平台故事真成了”的那点期权价值。康宁最钝，但也最底层，不管光模块技术路线怎么变，数据中心里光纤总得用。

中国这边，反弹的深层原因还是筹码结构，跌透了，禁令预期松了，业绩也兑现了。光迅科技三季度营收利润都创了历史新高，1.6T已经批量交付。但软肋也很清楚：产能优势没变成定价权。买家是全球云厂商，激光器卡在上游，政策风险来自美国监管。中际旭创H股比发行价跌了快四成，这说明市场愿意为中国的产能付钱，但不愿意为中国的产能付溢价。

把两边拼一起看，底层逻辑就一条：美国需要中国的光模块让已经买好的芯片转起来，中国需要美国云厂商的资本开支来消化扩出来的产能。这是相互挟持，不是谁单方面占优。但话说回来，微软去游说白宫，说明产业界是在争取时间，不是在放弃脱钩。FCC的限制计划没撤销，只是推到了3.2T那一代。中国厂商在这个窗口期能做的，就是把1.6T和CPO的交付纪录做实，把上游激光器的自给率往上推。做不到的话，下一轮政策冲击来的时候，反弹会更小，跌得会更狠。

最后说一句，以上内容仅代表个人对产业结构和市场叙事的理解与梳理，不构成任何投资建议。文中提到的公司、订单、产能和股价数据都来自公开信息，但市场变化快，信息也可能有滞后或偏差。光通信板块波动剧烈，政策风险、技术路线变化和筹码结构都可能让判断失效。投资决策请结合自身风险承受能力，独立判断，必要时咨询持牌专业人士。股市有风险，入市需谨慎。

[Steven Jia @JUNXINJIA]

Disclaimer: This is personal commentary and humor only. It does not constitute investment advice of any kind.

Let me be upfront: I'm a loyal NVIDIA investor. I hold the stock, I believe in the company, and it's made me money. So I'm not writing this to rubberneck — I'm writing it because it genuinely hurts.

The U.S. market's approach boils down to one sentence: Do whatever you want in your private life — just don't drag the company into a ditch.
1. The U.S. market doesn't judge character — it judges "contagion."
You cheat, you flirt, you text at midnight — the market has zero interest in playing morality police. It asks one question: will this screw shareholders, blow up, or create new risk? If yes, it investigates. If no, carry on. But I'm a shareholder. I don't care about the morality — I care about where that line of "will it contaminate the company" actually gets drawn.
2. Private conflicts of interest? Put them all in the sunlight.
Executives don't rely on self-discipline; they rely on hard rules. Personal relationships, hidden connections, potential conflicts — all disclosed. Want an exemption? Say it publicly and let the whole market watch. No backroom deals. I bought a company, not a mystery box.
3. When something breaks, management steps aside.
This Huang investigation isn't "insiders investigating insiders." It's outside independent directors only, plus brand-new outside counsel, with their own budget and full authority. The CEO can't touch it and can't bury it. The investigation is as neutral as the neighbor auditing your books. As an investor, I'd rather they investigate hard than feed me a snow job.
4. Directors don't slack off — because they're genuinely afraid of paying out of pocket.
U.S. directors aren't just ceremonial titles. See a risk signal and ignore it? Shareholders can sue you personally for damages. So the board has to do its job proactively instead of waiting for regulators to nag them. I'm all for this — I'm one of those shareholders who can sue.
5. The constraints are a three-dimensional siege, not a single pressure point.
SEC oversight, class actions, shareholder accountability, internal whistleblowers, exchange rules — locked down layer by layer. Anyone trying to fudge or cover up gets hit from multiple angles at once, like whack-a-mole.
6. A-shares' biggest problem: insiders investigating insiders.
When something goes wrong, it's basically this: management self-audits, its own law firm reviews, its own people judge its own people. Independent directors mostly have no real power, no independent counsel, and no independent budget. Check on the controlling shareholder? Mostly through eye contact.
7. Domestic regulators only chase violations, not "hidden risk spillover."
A-shares currently punishes crimes, violations, and pledge blowups — the obvious stuff. Executives' personal relationships, foreign connections, hidden benefit transfers, confidentiality risks — institutionally, these are blank spots. Nobody systematically checks. It's a blind spot under the lamp.
8. The core lesson worth learning: regulators should mind their boundaries and focus on the essence.
Good capital market governance doesn't act as a morality cop — it acts as a risk firewall. No administrative crackdowns; instead, institutional design that makes companies proactively self-audit and self-correct, plugging risks before they blow. You do your thing — just don't blow up my position. I'm a shareholder. That's not a joke. I mean it.

[Steven Jia @JUNXINJIA]

出轨不用下台，但风险必须查清～从英伟达风波看懂上市公司治理的底层逻辑

1. 美股根本不管你私生活对错
美国这套制度最聪明的地方：不审判CEO人品、不抓道德问题。它只盯一件事：你的个人问题会不会连累公司、坑股东、带出新风险。

2. 高管所有私人利益冲突，必须晒在阳光下
美股对大佬不是靠自觉，是靠硬规矩。任何私人关系、隐秘关联、潜在冲突，都得备案。真要破例豁免，必须公开说清楚，市场所有人都看得见，没法暗地操作。

3. 一旦出事，调查权彻底和管理层切割
像黄仁勋这次，不是公司自己人自查自圆。直接上纯外部独��董事+全新外部律师，自己有预算、自己说了算，CEO管不到、捂不住，调查完全中立。

4. 董事不敢摆烂，是真的怕担责赔钱
美股董事不是挂名混头衔。只要有风险苗头你视而不见、不查不报，股东真的能起诉你个人赔钱，所以董事会必须主动尽责，不是等监管来催。

5. 约束是多层立体的，不靠单一行政压力
SEC监管、集体诉讼、股民追责、内部吹哨人、交易所规则，层层卡死。谁想糊弄、隐瞒，随时会被从不同角度击穿。

6. A股最大问题：缺少真正独立的自查纠错机制
我们A股出事基本都是：管理层自查、自己律所核查、自己人审自己。独董大多没有实权、没有独立律师、没有独立预算，根本制衡不了实控人和高管。

7. 东大国内监管只盯违法，不管“隐性风险外溢”
���在A股只罚犯罪、违规、质押爆雷这些明牌。但高管私人关系、涉外关联、隐秘利益输送、涉密隐患，制度上是空白，没人系统核查。

8. 最值得学的核心经验：监管要守边界、抓本质
好的资本市场治理，不做道德警察，只做风险防火墙。不用行政打压，而是靠制度设计，让公司自己主动排查、自己主动纠错，把风险提前堵住。

[Steven Jia @JUNXINJIA]

Friends, stop forcing a link between party politics and tech stock moves. That's reversing cause and effect.

The cleanest proof: the internet bubble under Clinton, the mobile internet bull run under Obama, the AI capex cycle under Biden—three tech rallies under Democratic administrations. None of them rallied because of Democratic policy. They rallied because technology penetration inflected and liquidity loosened. Flip it around: tech stocks fell under Bush, and tech stocks rallied just fine under Trump's first term. Party affiliation explains close to zero of tech stock pricing.

Wall Street banks do have an incentive to fan the flames here. Political narrative is the best panic amplifier—simple, emotional, easy to sell, far easier to market than hard variables like capex slope and the rate path.

The only stable rule in this market: tech rallies happen when technology penetration inflects and liquidity loosens. Who's in power is background noise.

What investors should watch now isn't who controls Congress. It's two hard variables: the slope of AI capex and the Fed's rate path. When those two turn, tech turns. Party rotation doesn't change that pricing anchor.

The above represents my personal views only and does not constitute any investment advice.

[Steven Jia @JUNXINJIA]

朋友们，别把党派和科技股涨跌强行挂钩，那是把因果搞反了。

最直接的证明是在克林顿时期的互联网泡沫、奥巴马时期的移动互联网长牛、拜登时期的AI资本开支周期——三次民主党执政下的科技股大行情，涨的原因从来不是民主党政策，而是技术渗透率爆发叠加流动性宽松。反过来，小布什任内科技股是跌的，特朗普第一任期科技股照样涨得很好。党派变量在科技股定价里，解释力几乎接近于零。感觉那些华尔街投行故意���风点火，制造恐慌！

其实股市真正稳定的规律只有一条：科技股的大行情，永远发生在“技术渗透率快速提升+流动性宽松”的组合下。 谁执政，什么党派执政都只是背景噪音。

投资者现在要盯的不是国会谁说了算，而是两个硬变量：AI资本开支的斜率，和美联储的利率路径。 这两个东西转向，科技股才转向。党派轮替，改变不了这个定价锚。

[Steven Jia @JUNXINJIA]

Adding to optical connectivity, optical chips, and Nvidia. That's my personal play.

Optical got absolutely hammered this week. Lumentum down 30% in two sessions. Coherent down 6% in a day. Corning down 6%. Marvell down 30% in three days. The trigger is clear: Marvell's all-stock acquisition of Lumentum, $115 billion, with closing not until 2028. Both shareholder bases voted with their feet.

But here's the thing—what's getting sold is the deal structure, not the thesis.

Top five cloud providers' 2026E capex is approaching $800 billion, with 2027 guidance still moving higher. So where does the optical interconnect demand gap get filled from?

NVDA Q3 revenue came in at $57 billion, up 62% YoY. Data center $51.2 billion, up 66%. Blackwell's shipment slope is steepening, not flattening. The top five buyers have already deployed $556 billion, and guidance is still going up. You call that a bubble?

The optical interconnect logic is more rigid than GPU. GPUs can wait. Optical interconnect cannot. 800G is ramping, 1.6T is climbing, 3.2T validation is done. Every rate step-up drives non-linear expansion in optical chip volume and ASP. Lumentum shipments are running 25–30% below demand. EML, CW light sources, indium phosphide substrates—long expansion cycles, narrow windows.

NVDA at 35x TTM. Revenue growth 65%, DC 66%, GM 73%+. 35x. Expensive? That's a definitional question. NVDA defines the architecture: scale-out penetrating into scale-up, CPO/NPO/OIO converging. Copper-to-optics is irreversible. Optical chip value content moves monotonically higher.

My personal play is simple: short-term noise prices panic, long-term certainty provides the chips. Cloud providers' FCF pressure is real. Financing structure changes are real. Alphabet's CFO said it verbatim: 2027 capex will increase significantly. You can disagree with my call, but you have to respect their budget. There's no brake option in the 2027 arms race.

Optical connectivity sits at the most rigid bottleneck: slow supply expansion, high barriers, slow qualification, demand queuing up. That mismatch is where excess returns come from.

NVDA held. Optical chips added. That's my personal play. I can live with the volatility—the anchor is the physical limit of AI infrastructure over the next three years.

On valuation: the selloff is done, and it's cheap. That's math, not feeling.

Lumentum at $680, 26x forward versus a five-year average of 30x. Mean reversion implies roughly 17% of repair. CEO said it verbatim: products are essentially sold out through early 2029, EML gap exceeding 30% of demand. Orders locked in, 30% shortfall, 26x forward—you call that expensive?

Coherent at $195, down by half from $440. Forward P/E compressed from 57x to around 24x, hugging the five-year average of 23.6x. Analyst mean target $387.8, implying roughly 100% upside. Scale-up CPO ramps in 2H 2027, with optical content value per system at $15,000. The ramp hasn't changed—the stock just broke first.

Corning at 28.5x forward, PEG 0.96. Mean target $180 versus $118 now, implying 51% upside. Marvell was cut by BofA from $400 to $225—what got cut was the deal structure and regulatory uncertainty, not the optical interconnect fundamentals.

The valuation compression is over. What got sold was sentiment and deal structure, not orders. Lumentum's 30% gap, Coherent's order book locked through 2028, Corning tied to AI data centers. The valuation anchor has moved from AI narrative premium back to a reasonable manufacturing range, while the fundamental slope hasn't been revised down.

Nvidia at 35x TTM actually looks like the most "expensive" name in this basket. But it defines the architecture; it isn't defined by it. The valuation optical chips have sold off to is my direct reason for adding.

My personal play: panic pricing gives you chips, the valuation anchor gives you conviction.

Personal trading record. Not investment advice.

[Steven Jia @JUNXINJIA]

买入了光连接、光芯片、英伟达。我的个人做法。

光通信这周被血洗。Lumentum两天跌30%，Coherent单日跌6%，康宁跌6%，迈威尔三天跌30%。触发点：Marvell全股票收购Lumentum，1150亿美元，2028年才交割。两边股东用脚投票。

但杀的是交易结构，不是逻辑。

五大云厂商2026E capex逼近8000亿美元，2027指引上修。光互联需求缺口从哪补？

NVDA Q3营收570亿，+62%。��据中心512亿，+66%。Blackwell斜率在陡峭化，不是平缓化。前五大买家已投5560亿，指引还在上修。这叫泡沫？

光互联逻辑比GPU更刚性。GPU可等，光互联不可等。 800G放量，1.6T爬坡，3.2T验证完成。速率迭代驱动光芯片用量和ASP非线性扩张。Lumentum出货低于需求25–30%。EML、CW光源、磷化铟衬底，扩产周期长，窗口窄。

NVDA 35倍TTM。营收增速65%，DC 66%，GM 73%+。35倍。贵？定义问题。NVDA定义架构：Scale-out向Scale-up渗透，CPO/NPO/OIO收敛。光进铜退不可逆，光芯片价值量单调上行。

我的个人做法：短期噪音定价恐慌，长期确定性给筹码。 云厂商FCF压力真实，融资结构变化真实。Alphabet CFO原话：2027 capex显著增加。可以不认我，得认他们的budget。2027军备竞赛无刹车选项。

光连接卡在最刚性瓶颈：供给扩产慢、壁垒高、验证慢，需求排队。错配即超额收益。

NVDA持有，光芯片加仓，我的个人做法。波动扛得住，锚是未来三年AI基础设施物理极限。

估值：跌完了，确实便宜。算出来的，不是感觉。

Lumentum 680美元，26倍forward，五年均值30倍。均值回归约17%修复空间。CEO原话：产品基本卖到2029年初，EML缺口超需求30%。订单锁死、缺口30%、26倍forward，贵？

Coherent 195美元，从440腰斩。forward P/E从57倍压到24倍，贴五年均值23.6倍。分析师均值目标387.8，隐含翻倍。Scale-up CPO 2027下半年放量，单系统光学价值量1.5万美元。ramp没变，股价先崩。

康宁28.5倍forward，PEG 0.96。均值目标180，现价118，隐含51%上行。迈威尔被BofA从400砍到225，砍的是交易结构和监管不确定性，不是光互联基本面。

杀估值杀完了。杀的是情绪和交易结构，不是订单。 Lumentum缺口30%，Coherent订单锁到2028，康宁跟着AI数据中心走。估值锚从AI叙事溢价回到制造业合理区间，基本面斜率没下修。

NVDA 35倍TTM在这个组合里反而最“贵”。但它定义架构，不被架构定义。光芯片跌出来的估值，是我加仓的直接理由。

我���个人做法：恐慌定价给筹码，估值锚给底气。

个人交易记录，不构成投资建议。

[Steven Jia @JUNXINJIA]

@mcf0310 是啊

[Steven Jia @JUNXINJIA]

美股在资本开支这件事上，奉行一套耐人寻味的双重标准。一旦云厂商资本开支指引边际走弱，Marvell、Lumentum、Coherent、康宁瞬间被重新定价，被视作周期下行标的；它们高额的研发与产能投入，几���全部被当成周期风险。

与此同时，特斯拉与 SpaceX 常年保持巨量资本支出，商业化落地周期更长、不确定性更高。投资者给予它们稳定的期权溢价，把每一分投入都解读为护城河建设。

这并不是预判哪家企业终将胜出，而是指出华尔街根深蒂固的叙事偏见：终���平台拥有耐心，不可替代的硬件基建，却几乎没有容纳短期节奏波动的余地。

市场情绪会给周期错价。拉长时间，真实需求与技术壁垒往往会胜出。

[Steven Jia @JUNXINJIA]

@EdwardCassas "Love the old-school hustle, Edward! That’s a great throwback. 💯 Times have definitely changed though—the internet is wild these days, so a little extra verification just helps me filter out the actual scammers. Appreciate you reaching out. Cheers!"

[Steven Jia @JUNXINJIA]

PhD, Visiting Professor, and VC Founder.

I’ve invested in hundreds of companies across AI, embodied intelligence, robotics, biotech, advanced manufacturing, and semiconductors. My research covers frontier tech—from aerospace and quantum computing to new energy, advanced materials, and biopharma.

Beyond deep tech, I closely track geopolitics, global conflicts, macro trends, and global capital markets.

My core lens: I cut through the tech hype, stress-test commercial viability, and analyze how geopolitics and macro shape valuations, supply chains, and venture returns. I separate what’s actually technically feasible from overblown market BS.

Always open to sharp debates with founders, researchers, and fellow investors.

Note: To avoid spam and scams, I don't take cold calls, and I only engage in DMs after basic verification. Thanks for understanding.

[Steven Jia @JUNXINJIA]

The U.S. stock market operates under a curious double standard when it comes to capex.
When cloud capex guidance softens at the margin, Marvell, Lumentum, Coherent and Corning are instantly repriced as cyclical losers. Their heavy R&D and capacity spending is treated almost entirely as cycle risk.

Meanwhile, Tesla and SpaceX maintain years of massive capital outlays, with far longer and more uncertain commercialization timelines. Investors award them lasting option premiums, framing every dollar spent as moat building.

This is not a prediction about which businesses will succeed. It points to Wall Street’s ingrained narrative bias: terminal platforms get patience; irreplaceable hardware infrastructure gets almost no room for temporary rhythm shifts.

Sentiment can misprice cycles. Over time, genuine demand and technical barriers tend to prevail.

Disclaimer: This is my personal market observation, not investment advice. All views draw only from publicly available information. For full details, please see my article on Substack. https://t.co/crRTy2B7Ev

[Steven Jia @JUNXINJIA]

The views expressed in this article are solely my own and do not constitute investment advice.

[Steven Jia @JUNXINJIA]

Optical Communications Didn't Crash. The Way It's Being Paid For Did.

The market is in a panic. Chinese-language commentary is saturated with the intuition that "someone is systematically shorting this sector." But if you break down the decline logic of the four core names, the conclusion is harder to swallow: not a single one of them was killed by an outside force. Each one walked into a pit of its own making—they just happened to fall in at the same time.

Marvell + Lumentum: A self-inflicted squeeze via all-stock acquisition. Lumentum shareholders receive 4.5 Marvell shares per share, implying a deal value of roughly $1,132 at announcement. After Marvell fell to $196.40, those same 4.5 shares were worth only about $884—a 20% haircut. The buyer falls on dilution, the seller falls because "the money it's receiving is depreciating." One mechanism drives both directions. This isn't a short-selling campaign; it's a structural flaw of all-stock deals in a volatile market.

Coherent: Dragged down by its paymasters' cash-flow anxiety. No structural flaw in its own operations—what's falling is its customers. Three of the five largest tech giants reported negative free cash flow in their latest quarter, and Alphabet plunged 12.6% in a single day. Optical communications revenue is highly dependent on cloud giants' capex. When the "paymasters" go cash-flow negative, the market is repricing "the people paying may no longer be able to afford it," not "the product isn't selling."

Corning: One leg running fast, three legs dragging. Enterprise optical grew 84%, but the solar business posted a net loss and glass revenue and profit both declined. The earnings beat rests almost entirely on one optical leg. When sector valuations compress, a single-leg earnings structure is the most fragile. Short interest is just 2.22%—what's falling is structural imbalance, not a broken story. The repair window is 2027.

These three death paths converge on the same macro variable: AI infrastructure expansion is shifting from a "technical feasibility" problem to a "financing sustainability" problem. Google and Oracle have both turned free-cash-flow negative, and optical communications is the most sensitive link in the chain—revenue is recognized when orders land, but valuation is anchored to expectations of how much longer customers can keep burning cash.

Investors should stop treating these four as one sector. Marvell/Lumentum hinge on whether Marvell's stock can stabilize—the all-stock consideration is floating, the deal doesn't close until 2028, and there's plenty of time for repricing in between. Coherent tracks hyperscaler cash-flow expectations; no structural flaw of its own. Corning has the clearest repair path, but it needs solar capacity ramp-up to play out—2027 is the observation window.

The industry logic of optical communications isn't broken. What's broken is the "paying with stock" deal structure, the "paymasters' cash-flow crunch" financing anxiety, and the "one leg running fast" earnings structure. The short-selling forces are market-driven and structural, not conspiratorial. Treating them as a sector that rises and falls together is itself the mistake.

[Steven Jia @JUNXINJIA]

The views expressed in this article are solely my own and do not constitute investment advice.

[Steven Jia @JUNXINJIA]

Optical Communications Didn't Crash. The Way It's Being Paid For Did.

The market is in a panic. Chinese-language commentary is saturated with the intuition that "someone is systematically shorting this sector." But if you break down the decline logic of the four core names, the conclusion is harder to swallow: not a single one of them was killed by an outside force. Each one walked into a pit of its own making—they just happened to fall in at the same time.

Marvell + Lumentum: A self-inflicted squeeze via all-stock acquisition. Lumentum shareholders receive 4.5 Marvell shares per share, implying a deal value of roughly $1,132 at announcement. After Marvell fell to $196.40, those same 4.5 shares were worth only about $884—a 20% haircut. The buyer falls on dilution, the seller falls because "the money it's receiving is depreciating." One mechanism drives both directions. This isn't a short-selling campaign; it's a structural flaw of all-stock deals in a volatile market.

Coherent: Dragged down by its paymasters' cash-flow anxiety. No structural flaw in its own operations—what's falling is its customers. Three of the five largest tech giants reported negative free cash flow in their latest quarter, and Alphabet plunged 12.6% in a single day. Optical communications revenue is highly dependent on cloud giants' capex. When the "paymasters" go cash-flow negative, the market is repricing "the people paying may no longer be able to afford it," not "the product isn't selling."

Corning: One leg running fast, three legs dragging. Enterprise optical grew 84%, but the solar business posted a net loss and glass revenue and profit both declined. The earnings beat rests almost entirely on one optical leg. When sector valuations compress, a single-leg earnings structure is the most fragile. Short interest is just 2.22%—what's falling is structural imbalance, not a broken story. The repair window is 2027.

These three death paths converge on the same macro variable: AI infrastructure expansion is shifting from a "technical feasibility" problem to a "financing sustainability" problem. Google and Oracle have both turned free-cash-flow negative, and optical communications is the most sensitive link in the chain—revenue is recognized when orders land, but valuation is anchored to expectations of how much longer customers can keep burning cash.

Investors should stop treating these four as one sector. Marvell/Lumentum hinge on whether Marvell's stock can stabilize—the all-stock consideration is floating, the deal doesn't close until 2028, and there's plenty of time for repricing in between. Coherent tracks hyperscaler cash-flow expectations; no structural flaw of its own. Corning has the clearest repair path, but it needs solar capacity ramp-up to play out—2027 is the observation window.

The industry logic of optical communications isn't broken. What's broken is the "paying with stock" deal structure, the "paymasters' cash-flow crunch" financing anxiety, and the "one leg running fast" earnings structure. The short-selling forces are market-driven and structural, not conspiratorial. Treating them as a sector that rises and falls together is itself the mistake.

[Steven Jia @JUNXINJIA]

光通信没崩，崩的是买单方式！

市场在恐慌，中文舆论场弥漫着“有人系统性做空”的直觉。但是如果你把四只核心标的的下跌逻辑拆开，结论更难堪：没有一只股票是被外部势力杀死的，是它们各自踩进自己挖的坑，只是恰好同时掉了进去。

Marvell + Lumentum：全股票收购的自我��杀。 Lumentum股东每股换4.5股Marvell股票，信息宣布时隐含对价约1132美元。Marvell跌到196.40美元后，同样4.5股只值约884美元，缩水了五分之一。买方跌是稀释，卖方跌是“收的钱在贬值”，同一机制驱动两个方向。这不是做空势力，是交易结构在波动市里的固有缺陷。

Coherent：被金主的现金流焦虑拖下水。 自身经营没有硬伤，跌的是客户。五大科技巨头最新季报三家自由现金流为负，Alphabet单日暴跌12.6%。光通信收入高度依赖云巨头的资本开支，当“金主”现金流转负，市场重新定价的是“付钱的人可能付不起了”，不是“产品卖不动”。

康宁：一条腿跑得很快，三条腿在拖后腿。 光通信企业端增速84%，但太阳能业务净亏损、玻璃业务营收利���双降。业绩超预期几乎全靠光通信一条腿，板块估值一压缩，单腿支撑的盈利结构最脆弱。做空比例仅2.22%，跌的是结构不均衡，不是故事破产。修复窗口在2027年。

这三条死路交汇于同一个宏观变量：AI基础设施扩张，正从“技术可行性”变成“融资可持续性”问题。 谷歌、Oracle自由现金流转负，光通信是链条上最敏感的一环——收入确认在订单落地时，估值锚定在客户还能烧多久的预期上。

投资人未来别把这四只当同一个板块。Marvell/Lumentum看Marvell股价能否企稳，全股票对价浮动，2028年才交割，中间有大量重新定价的时间；Coherent跟hyperscaler现金流预期走，自身无结构性硬伤；康宁修复路径最清晰，但要等太阳能产能爬坡，2027年是观察窗口。

光通信的产业逻辑没坏。坏的是“用股票买单”的交易结构、“金主现金流告急”的融资焦虑、“一条腿跑得快”的盈利结构。那些做空力量是市场化的、结构性的，不是阴谋性的。把它们当成同涨同跌的板块来交易，本身就是错的。

[Steven Jia @JUNXINJIA]

救市救错了层级？央行托不住盈利，减税降费才是真药方！

市场吵“哪种救市更管用”，却没人问对问题：每类工具打的是哪个层级、药效撑多久。把央行买信贷、免印花税、减税降费摆一张桌上比，本身就是层次错配。

① 央行/汇金买信贷、ETF——守下���，托不住企业端盈利
这种操作本质是防系统性风险、补流动性、管预期。2025年4月央行通过再贷款给汇金兜底，理论上打开资金上限，形成类平准基金闭环。但历史反复打脸边际效应：2008年汇金首次增持四大行，沪指单日暴涨9.45%；2011、2012、2013年后续增持，市场只剩微涨甚至下跌。学术实证同样：短期提振信心，中长期成交量“冲高—下行”，机构趋于谨慎。
它不改变企业盈利，也不改变居民持久收入。用流动性工具搏盈利修复，是工具与目标错配——这就是“有效性不行”的根源。错不在工具，在期待层级。

② 免印花税——交易端短效脉冲，效果猛但不改趋势
降交易成本、提换手率、抬风险偏好，历史复盘当周成交量升幅70%–139%，���绪冲击够猛。但2008年9月单边征收后，股市4个交易日见顶回落。它触不到资产端盈利，注定短效。极端时刻稳情绪可以，指望它逆转趋势是误判。

③ 财政减税降费——长期结构性工具，有时滞但是效应持续
最易被误归类的一类。它不是短期刺激，是长期结构性改善。
企业端：降税改善现金流、降内部融资成本、抬升扩大再生产的边际收益率。企业加不加信贷，取决于投资回报率与利率之差，减税改善的正是这个差——供给端长期效应，不是发一次钱就结束。
居民端：降税增加可支配收入，改变持久收入预期。持久收入的边际消费倾向远高于暂时性收入，消费拉动是趋势性的，不是一次性的。
减费：压低制度性交易成本、改善营商环境，影响长期投资意愿与进入退出决策，属产业经济学意义上的长期成本冲击。
发消费��是最即时但是可能是一次性、有替代效应，是短兵器；减税降费有时滞但效应持续，是长兵器。用“立竿见影”要求减税降费，是以短兵器标准衡量长兵器。

所以说短效守下限，长效搏上限，是组合不是替代
央行兜底防崩盘，守下限；印花税提情绪，做短效脉冲；减税降费修企业现金流与居民持久收入，搏上限。把减税降费当短期刺激用是误判，当长期结构性工具用才是正解。真正有效的救市从来不是二选一，而是短效守下限＋长效搏上限的组合拳——流动性工具不让市场崩，财税工具让微观主体重新有能力和意愿扩张。层次错配，才是讨论救市有效性时最该规避的错误。

[Steven Jia @JUNXINJIA]

奇怪

[Steven Jia @JUNXINJIA]

22.4万亿蒸发：谁在流血，谁在装死？

5月13日，A股总市值137.33万亿。今天，114.9万亿。五个月，22.4万亿灰飞烟灭。

这钱不是均匀消失的。它落在不同人头上，痛感完全不同。
公募基金，亏得最体面。 三季度整体亏损1.53万亿，股票型亏7463亿，混合型亏7770亿。基金经理排队写道歉信——“深表歉意”“非常应当反思”。但亏的是基民的钱，道歉的是基金经理，管理费照收。大基金不是扛住了，是痛感延迟。
私募和杠杆资金，痛得最真实。 同犇投资从浮盈80%到亏损超10%，游资1.7倍杠杆梭哈东山精密，股价腰���，净值从4跌到1.2。杠杆是放大器：涨时放大贪婪，跌时放大死亡。
散户，数据最残忍。 上半年79%以上账户亏损，人均浮亏2.1万。10万以内小散，亏损率98%以上。盈利超30%的散户只有3.1%，翻倍的不足0.3%。散户贡献六七成交易额，只拿三成市值——在一个水位下降的池子里拼命游泳。
外资，跑得最聪明。 三季度北向净流出902亿，长线外资流出869亿。嘴上说长期看好，身体在调仓：撤科技，躲银行医药。
所以22.4万亿是谁的？
散户贡献了最大的人数损失和情绪损失——最没有缓冲垫的一层。公募贡献了最大的账面损失——但痛感是间接的。杠杆资金贡献了最惨烈的个体死亡——爆仓不出现在统计里。外资贡献了最聪明的撤退——没硬扛，换了方向。
市场从不消灭财富，它只转��财富。问题永远是：从谁，到谁。

你的账户，是被转移的那一边，还是接收的那一边？