The Underwater Private-Market World: From SpaceX to ByteDance Secondary Shares

Summary

• 庄明浩给一级半市场的定义直指其成因：一级市场基金“到期一定要结账”，但滴滴事件后美股上市基本停滞、港股低迷、A股一度堵塞，退出环境没有给大家到期结账的空间，于是老股交易从水下江湖被逼成一个有买方、卖方、标的价格和层层中间方的准市场。 The same trend has emerged in China and the US over the past few years, and it is moving toward formalization and tighter oversight.
• The US private-company market has formed a “real Magnificent Seven”: SpaceX, OpenAI, xAI, Anduril, Databricks, Anthropic and Stripe, representing the three dominant consensus themes of AI, defense and chips. Raymond says SpaceX is already a $1.8T company and expects OpenAI and Anthropic to reach valuations above $1T when they go public; even Stripe, ranked last, has “so much momentum that it can hold a conference alongside AWS, Google Cloud or even Nvidia,” and has just acquired OpenRouter.
• SpaceX went from a deal available only to VCs willing to commit tens of millions to hundreds of millions of dollars to an investment accessible for $5,000, enabled by layered SPVs and financial infrastructure left behind by Web3 — “putting the hype aside, Web3 did help.” Dollar capital with Chinese links is participating in SpaceX and Neuralink through layers of SPVs; buyers “don’t know whether they are touching the second layer or the seventh,” but once a deal closes, they can claim, “I’m already an investor in it.” Raymond says Chinese nationals cannot invest in US AI companies under the current US regulatory framework, yet these structures continue to appear.
• On pricing high-growth assets, Raymond asked how much weight formal factors carry relative to fundamentals, and 庄明浩 answered that “the vast majority is being assigned to formal factors.” When Zhipu secondary-share SPVs appeared at the end of last year, they were still using 2024 financial data, with revenue “roughly equivalent to zero,” yet trading at RMB16B–18B valuations. The rules are “willing buyer, willing seller; you take the bet, you live with the outcome,” and “when the rocket is about to launch, don’t worry about your seat.” For stagnant assets such as Canva and Discord, however, the information set is “already not much different from a prospectus,” bringing the decision back to traditional investment craft.
• ByteDance is the world’s singular private-secondary asset: it appeared more than 500 times across 180-plus 36Kr message-board issues over 5 years, almost every issue; its annual employee-share buyback price creates self-reinforcing price discovery, and it has the cash and willingness to buy back secondary shares into its own Cap Table, functioning as a liquidity pool. 庄明浩 thinks ByteDance cannot go public in the short term, while Raymond sees it as a market-proof company: “ByteDance itself may be bigger than Hong Kong; wherever ByteDance goes, it will lift that market,” and the capex required for the AI war demands the financing tools available to public companies.
• More than 5,000 pieces of information on the 36Kr message board over 5 years form a statistically meaningful sample of China’s private-secondary market: Raymond notes that 87% were sell orders in 2021, while 57% are buy orders this year, with the market rotating from internet and consumer assets to Web3, hard tech and AI, then to robotics and commercial space — a “leading indicator of sector heat.” But the board is only a message board, not a trading venue: “the distance from displaying and listing information to an actual closing is enormous,” which is also why Chinese technology media have struggled to make FA businesses work in practice.
• SHEIN is set to list in a few days, becoming the most sobering case: its peak valuation was $98B, versus a $26B valuation in its IPO prospectus. Raymond asks why capital-markets execution reached this point when almost no company matches SHEIN in Chinese supply-chain management. 庄明浩’s answer is that individual actors — investors, banks, founders and others — may all have offered seemingly rational advice, but “something went wrong in the group decision,” while timing, external conditions and luck still exert enormous influence over an IPO.
• One overlooked opportunity: the message board carried 321 “sell-the-company” notices seeking acquisitions in 2024, a figure that fell sharply in 2025, which 庄明浩 attributes to more companies being able to pursue Hong Kong listings. The biggest obstacle to Chinese M&A is a generation of founders who “never leave the office,” but those 321 owners willing to step away are worth systematic screening. Raymond’s verdict: “This could make a lot of money.”

Deep dive

1. The Private-Secondary Market: Funds Must Settle, but There Is No Exit

• 庄明浩’s primer: the primary market is VC financing for unlisted companies; the secondary market is the public market after an IPO, with price discovery, trading hours and financial instruments. The private-secondary market is the middle state: the company is still private, but early shareholders need to exit, shares need to circulate, “there is a buyer, a seller, an asset and a price for that asset,” plus a range of intermediaries built to make the trade happen.
• The core contradiction is that “all funds in the primary market are essentially vehicles with finite lives … they have to settle up, while the secondary market can exist in perpetuity.” The reality is that “the exit environment did not give everyone room to settle at maturity,” forcing participants to find a route through the middle. Market turmoil in China and the US over the past few years has pushed what was once a “below-the-surface, highly volatile market” toward greater formalization, more regulation and visibly faster growth.

2. ByteDance Has Not Listed in 14 Years; Angel Investors May Be Due for Their Second Exit

• Raymond highlights the structural mismatch: ByteDance was founded in 2012 and is now 14 years old, “far beyond the life of most funds.” Angel investors “may now be due for their second exit”; in his view, the company is market-proof but has stayed private for various reasons, leaving investors stuck.
• 庄明浩 adds a phenomenon that “basically did not exist” earlier in his career: unlisted companies actively buying back employee options at an annual price. In China, only ByteDance, Xiaohongshu and Kimi are now widely recognized and relatively well understood examples. Raymond’s view is that “ByteDance is because it really makes a lot of money; the other two companies may each have their own reasons.”

3. SpaceX: From a $100M-Plus Ticket to a $5,000 Retail Asset

• 庄明浩’s explanation: SpaceX carries the full halo of Musk’s success. Retail investors “actually don’t care that much about its financial data … what people really believe in is the story and the grand promise Musk is building.” Participation originally required becoming an LP of a top-tier fund, with too high a threshold. Demand from both sides pushed platforms and “trading methods” in quotation marks to emerge, eventually breaking the asset down to $5,000-sized tickets.
• A frequently overlooked enabler was Web3: “putting aside the hype around Web3, purely from the perspective of financial-industry infrastructure, Web3 really did provide help.” Even if “there may be legal defects in the middle of the chain,” the narrative and the hook were at least connected at the right moment. Raymond linked this to his episode “How to Buy SpaceX Stock at Half Price Before Its IPO”: after the IPO news emerged, the platform price doubled.

4. Chinese Money in SpaceX and Neuralink: Nobody Knows Whether It Is Layer Two or Layer Seven

• 庄明浩 was surprised when compiling the message-board data that Neuralink ranked in the top 5, ahead of Discord, Cainiao and DJI. The reason is that dollar capital with Chinese links needs assets: internet-company co-founders after an IPO and retired industry veterans are not wealthy enough to establish family offices, yet have difficulty becoming LPs of top-tier dollar funds. They can only participate through SPV stacked on SPV. “We don’t know whether they are touching the second layer or the seventh,” but once the deal closes, “in theory they can say, ‘I’m already an investor in it.’”
• Raymond’s anecdotal test: he often sees people on WeChat public accounts or Xiaohongshu claiming to be SpaceX investors, but “I also don’t know whether he invested $5,000 or $50M.” The two worked through a hypothetical: invest $100, with $50 in OpenAI and $50 in Anthropic, and “I can claim that I’ve invested in both companies.” 庄明浩’s response: “In fact, that’s not wrong.” Raymond repeatedly stressed: “This is absolutely not investment advice.”

5. Robinhood Tokenization: You Are Buying an IOU and Trust in the Platform

• In the second half of last year, Robinhood launched tokenized versions of OpenAI and Anthropic pre-IPO shares. Both companies quickly said the products were illegal and unrecognized. Robinhood’s logic was: “The stock in front is a stock issue; the token I list is a token issue.” The two broke down the substance: buyers are “buying a certificate, an IOU from Robinhood,” and betting that it “won’t run away and won’t refuse to honor it,” not buying ownership of the shares. Even when a large, public platform such as Robinhood is involved, the structure may still present legal problems.
• The regulatory wall is higher still. Raymond says Chinese nationals cannot invest in US AI companies under the current US regulatory framework. He also mentioned the $380B round in the first half of this year, where many different structures, briefs and documents imposed requirements concerning the buyer’s Chinese identity. That raises the question of why people would insist on buying something that is not legal.

6. Takedown Requests: How Sensitive Chinese Capital Is Inside SpaceX’s Cap Table

• 庄明浩 described a personal experience: a Web3-oriented media outlet examined how many China-related funds were participating in SpaceX’s IPO through SPVs and obtained the Cap Table. After the article was published, “many people came to me asking whether I knew the journalist and whether the article could be deleted.” The amounts were not especially large — $20M or $30M invested in the early days was “not a particularly large amount” for Web3 funds at their peak — but the parties did not want to be exposed. The article may have used only pinyin, yet people who knew the details could still identify them.

7. The Real Magnificent Seven: AI, Defense and Chips Consume the Entire Consensus

• The list itself is the thesis: SpaceX, OpenAI, xAI, Anduril, Databricks, Anthropic and Stripe “represent the huge consensus in the entire US private market over the past 1-2 years: AI, defense and chips.” Raymond says SpaceX is already a $1.8T company and expects OpenAI and Anthropic to be valued above $1T when they list. SpaceX is being discussed as part of a ranking of unlisted companies, not described as already public.
• 庄明浩 compares this type of ranking with the S&P 500: “It keeps kicking out the companies that perform poorly and adding new ones, so the index can keep rising.” Companies such as Canva that are not appreciating are excluded. Even the tail-end name, Stripe, recently acquired OpenRouter and has “so much momentum that it can hold a conference alongside AWS, Google Cloud or even Nvidia.” With no clear listing timetable, its private-secondary trading is extremely active, second only to OpenAI, which may have ranked No. 1 the previous year.

8. Raymond’s Read on Sam Altman: Helping Employees Sell Their Secondary Shares

• Raymond offers one angle on Sam Altman: after being pushed out and then brought back, “one of the most important things he may do every year is help many OpenAI employees sell their secondary shares.” He says OpenAI sold to SoftBank last year at a very high price, and this year’s transaction may already be complete, at more than $70 per share. Distributing the proceeds of such a large company and valuation in a compliant, legal way “is not something just anyone can do.”
• Raymond adds that Anthropic also conducted an employee-share tender in April or May this year, but he “doesn’t know who Anthropic was selling with.”

9. Formal Pricing: Zhipu’s Secondary Shares Had Almost No Fundamentals

• Raymond’s question was about normal information transparency. An investor buying Pinduoduo can review 8 quarters of financial statements and infer from earnings calls that management is “extremely indifferent” to the stock price. By contrast, nobody knows whether Anthropic’s ARR is net or gross, what proportion is recurring revenue, or whether users churn after token maxing. So how much of the price is formal and how much is fundamentals? 庄明浩’s answer: “The vast majority is formal. There’s no alternative.”
• Zhipu is the cleanest example. It listed early this year, but by the end of last year China already had a batch of Zhipu secondary-share SPVs using 2024 financial data, with revenue or fundamentals “basically equivalent to zero,” yet trading at RMB16B–18B valuations. It was “a potential transaction possibility being pushed forward by enormous emotion.” The rules are “willing buyer, willing seller; you take the bet, you live with the outcome”: “When the rocket is about to launch, don’t worry about your seat. Get on first.”
• The counterexample is the stock of companies that are not rising. Canva and Discord “are still growing revenue, are not small, and still hold their positions,” but “the richness of information in the private-secondary market may already be no different from a prospectus.” Buying into them brings the decision back to relatively traditional investment craft.

10. Vanished Unicorns and S Funds: A Liquidation Logic Like Used-Phone Recycling

• Raymond observes that many names outside the top 10 on Hurun- or 任泽平-style unicorn lists “have been on there since 2015 and have not been mentioned for many years,” yet remain listed. 庄明浩 connects this to IT Juzi’s 2015 “vanished unicorns” list. These companies appeared countless times in LP letters and once looked spectacular in IRR calculations, but when a fund reaches maturity and settles its books, those positions have to be cleared.
• 庄明浩 says some Chinese institutions specializing in private-secondary funds value fund interests at cost. Raymond adds that some do not mark them up at all; it depends on asset quality. Why take the 9 dead projects along with the one good project? Because “if you take it separately, it becomes a true single-asset purchase.” Fund liquidation is a portfolio-level audit: the 10 portfolio companies have different statuses, costs and reasonable valuations; “most should be written off,” while only a few merit discussion over a discount, par value or modest premium. Raymond’s analogy is used-phone recycling: “An iPhone originally sold for RMB6,000; I buy it back for RMB1,000, take it apart, pull out the memory and sell it, and throw everything else away on the spot.” 庄明浩 says that is only one possible outcome; every situation occurs.

11. Xiaohongshu Pricing: 3 Deals on the Same Day, 3 Different Prices

• Xiaohongshu first appeared on the list at roughly $12B. The last price that could be found “should be” $38B, above the $32B valuation Zhu Xiaohu was rumored to have cited at the time; the round recently bought by DST “may have been in the low-$20Bs.” 庄明浩 ranks the pricing factors as follows: the underlying business comes first, followed by trading sentiment, the distance to a listing timetable, the China-US environment and seller concentration. Funds that entered near the peak tend to mature at roughly the same time, creating a wave of supply over a short period and making a strong price impossible.
• The highly non-standard reality is that China lacks the kind of online trading venue available in the US, so much of the process is negotiation. Within the same time window, 3 Xiaohongshu secondary-share trades may clear at 3 entirely different prices, depending on whether the buyer enters the Cap Table directly, how many structural layers are nested, whether the consideration is in RMB or dollars, whether the buyer is an individual or an institution, the payment terms and the buyer’s identity. Sophisticated buyers prepare different tools for different transaction formats and quote accordingly.

12. ByteDance Is Its Own Pool: It Has a Price and the Cash to Support Liquidity

• ByteDance appeared more than 500 times across 180-plus 36Kr message-board issues over 5 years, “almost every issue, with both buys and sells.” Its special feature is the closed loop: it publishes an annual employee-share buyback price, creating its own price discovery; in first-layer transactions involving shares that actually enter the Cap Table, ByteDance is willing to buy when investors need to exit because it has enough cash. It can serve as a liquidity pool, so “for shareholders, whether it lists or not makes less difference.”
• Raymond adds 2 secondary-share pricing mechanisms. For companies with clear financing needs, such as OpenAI and Anthropic, the benchmark is typically the new round’s pre-money valuation, less a discount. 庄明浩 cautions that “sometimes there is no discount,” although Raymond’s past experience includes 70% and 75% pricing. For ByteDance, which has not raised serious money from the market since 2018, prices are also driven by each fund’s maturity waterfall: “Someone at SIG is retiring and needs to sell; a Sequoia fund has exited and needs to sell; an angel investor wants to buy a house and needs to sell.”

13. The 36Kr Message Board: Statistical Value in 5,000 Pieces of Data Over 5 Years

• When 36Kr launched the database in 2021, 庄明浩 had a serious conversation with its CEO and worried that “suddenly there would be shares circulating in the market at a price far below your financing valuation, while you had no idea who the seller was or why it was that price.” He is now willing to use the data because more than 5 years, 180-plus issues and 5,000-plus pieces of information provide enough history to screen out some outliers. 36Kr also did not receive a major direct benefit from the exercise. “The trends produced through AI do match the events we remember from the past few years, so they can be brought out for discussion.”
• Why did 36Kr not turn the board into a trading platform? 庄明浩 sees a structural reason: “Information is only the first step in a transaction, and may even carry very little weight in the entire complex process.” Secondary-share trades are far more complex than financing, involving instrument selection, information being passed and reshaped, coordination with company shareholders and existing investors, and various above- and below-the-surface arrangements. “Many Chinese technology media outlets have tried to do FA, and it seems none has done it particularly well,” because the distance between displaying and listing information and actually closing a deal is enormous. 36Kr remains a message board responsible only for displaying information.

14. Friction in Non-Standard Trades: Pooling Money, Broken Chains and Luxury-Home Brokers

• Raymond described pooling money with friends to buy secondary shares: “We wanted to buy RMB1M; he put up RMB300,000 and I put up RMB500,000.” Legal structures, nominee holdings and trust issues multiply. “You keep creating group chats, keep talking to lawyers, and keep dealing with all kinds of problems,” while the labor required and the potential return are highly unpredictable; the outcome may even be nothing.
• 庄明浩 was cautious on the worst cases. He has not personally heard of a very large, obvious case of someone absconding or refusing to honor a deal, but “it definitely happens; we just may not know about it.” Every additional SPV layer, or every inclusion of individuals, companies, fund nominees and RMB-dollar conversions, creates another point where a hook can break and the transaction chain can collapse. The more complex the chain, the higher the probability of failure.
• The ecosystem is small: there are not many FA firms or intermediaries dedicated to these trades, and participants may all know the same people. They close only a few deals a year, “just like luxury-home brokers.” The surrounding landscape includes people liking a post on Xiaohongshu and receiving a DM offering fund interests, as well as the joke that shares in a leading company can be bought at a commercial KTV in Hangzhou. Raymond also says the DeepSeek financing round helped popularize many of these concepts.

15. SHEIN: From $98B to $26B, a Breakdown in Group Decision-Making

• SHEIN is set to list in a few days, with a peak valuation of $98B versus a $26B IPO prospectus valuation. 庄明浩’s framework is that “valuation is an outcome, compressing every factor into a single dimension.” The company considered the US, tried becoming a Singapore company, explored the UK and Hong Kong; these issues were not entirely of SHEIN’s own making and were also shaped by China-US trade tensions and other external factors.
• Raymond unusually interrupted to say he did not fully agree. SHEIN’s business is exceptionally solid: “In terms of Chinese supply-chain management, there is almost no company that compares.” He has visited the factories and understands the fast-response supply chain. His question was why capital-markets execution reached this point: did investors, lawyers, investment banks and the founder not influence the outcome?
• 庄明浩’s response was that the final decision still belonged to the founder. Investors, banks, founders and other decision-makers may all have offered seemingly rational advice and feedback, but something went wrong when the group made its decision, compounded by luck and external conditions. The conclusion is that an IPO still contains many metaphysical factors and is difficult to explain purely through rationality and long-termism. Most SHEIN messages on the board were sells rather than buys, but appearing on the list means holders believed it was worth trading, and that it had at least once reached a moment of glory.
• The list’s anonymous aesthetic is that most trades appear as “a certain company.” Tuhu appeared many times, but was always described as “a certain automotive aftermarket-services company,” because it may be the only company in the field worth discussing. “A certain autonomous-driving company” cannot be identified with confidence.

16. Cainiao: A Club Deal from Day One, and the Chef Has Gone Home

• 庄明浩’s verdict on Cainiao is that it represents a classic form of Chinese club deal: “It was a club deal on the first day it was born.” Every participant wanted the person who organized the club to handle everything, but that organizer faced other issues over the past year and could no longer carry the load, while Cainiao was too peripheral to its system.
• His analogy: “Everyone wants to come to my house for dinner, but the chef has gone home. There is no food, so we can only order some takeout and make do.” The holders are left to solve the problem themselves. “There is no one to take it. Who would take it?”

17. From 87% Sells to 57% Buys: Leading Indicators and Sector Rotation

• The data begins in June 2021, when the Didi incident occurred. 庄明浩 says, “The world before Didi and the world today are 2 different worlds.” US listings effectively disappeared, Hong Kong weakened and A-shares were blocked; in the “VC is dead” era, everyone used every means available to find an exit, creating the first wave of private-secondary-market activity. Raymond notes that 87% of information was sell-side in 2021; as Hong Kong reopened, A-shares supported hard tech and the US showed signs of thawing, buy-side information reached 57% this year.
• The rotation is easy to read: internet platforms and consumer names, including Heytea at a peak valuation of RMB60B — “what is it worth now? Nobody knows; there have been no trades for several years” — then Web3 and the metaverse, hard tech, AI, and finally robotics, commercial space and rockets. When a new cycle emerges, there are often no sellers in the new names because holders still see upside. Take Unitree: many people want to buy, but that does not mean they can.
• 庄明浩 accepts Raymond’s methodological warning: a message board is not the same as a closing. In the extreme case, it is possible that not a single one of the 5,000 pieces of information resulted in a trade. The board can sense market sentiment, but it “cannot actually solve the transaction itself and certainly cannot prove that a deal closed.” Their consensus is that sufficient volume, a long enough time series and alignment between the assets and real-world experience create meaningful credibility; the board is a “leading indicator of market sentiment.”
• Raymond also says he recently used AI to organize more than 20,000 resumes from OpenAI, Anthropic and DeepMind. More than 1,000 and nearly 2,000 names were written in Chinese pinyin, with roughly 300 clearly linked to China. The inspiration came from 汉洋’s 2023 piece “Who Is Building Binance.” He believes the work would have been difficult without AI. 庄明浩 says both matches and non-matches can validate a view or produce unexpected discoveries. Information transmission is also accelerating: a chain that once passed through traders, analysts, institutions, retail investors and media may now complete in 1 day, 2 days or even 24 hours.

18. 321 Sell-the-Company Notices: Founders Willing to Step Away Are Scarce Assets

• The message board also carries acquisition-seeking notices, potentially from industrial buyers, A-share listed companies or their affiliates. 庄明浩’s context: “China has no M&A platform. There is no venue for matching assets; M&A is all about fate.” These requests account for only a single-digit percentage of the total, but they were relatively numerous in 2024 and fell sharply in 2025, which he attributes to more companies being able to list in Hong Kong.
• Raymond sees the opportunity: Chinese M&A is difficult because company founders are not professional managers. A first-generation entrepreneur “never clocks out — do you think Jack Ma has clocked out? Of course not.” The 321 sell-the-company notices in 2024 therefore represent 321 founders willing to step away. “I think this is very worth discussing … this could make a lot of money.” 庄明浩 admits he has not yet examined the group closely, but says it is worth going back and mining.

19. The Next ByteDance? “Wherever ByteDance Goes, It Lifts That Market”

• 庄明浩’s view is that “it will be difficult; ByteDance is too special.” SpaceX is already on the list, and OpenAI and Anthropic will soon list, while new US companies will continue entering the so-called Magnificent Seven. But ByteDance cannot go public in the short term; he jokes that one would have to ask “how good China-US relations need to become before ByteDance lists.” In his view, Hong Kong would struggle to absorb ByteDance, and 5 years from now it may still rank first.
• Raymond takes the opposite view: ByteDance is market-proof and could list in either a bull or bear market. “ByteDance itself may be bigger than Hong Kong.” Whether Hong Kong can absorb it is not the issue; wherever ByteDance goes, it will lift that market. Raymond also argues that the AI war requires massive capex — buying GPUs, building data centers and constructing steel-and-concrete “steel monsters.” Public companies have financing tools, including borrowing and issuing convertible bonds, that private companies struggle to access. There have also been recent rumors that ByteDance may go to the STAR Market.
• 庄明浩 retains his uncertainty: “It may also suddenly list one day. Who knows?” Raymond thinks Zhang Yiming may feel, “I have enough for myself; but if I need to, I can also list.”