Where Did the Money Go? Who’s Making Moves? A Review with 刘旌 of VCs’ New Consensus and Undercurrents in 2025
Summary
- The most consequential change in China’s primary market in 2025 was the rapid strengthening of VC consensus after AI became “the hope of the whole village.” The market was still deeply pessimistic at the end of 2024, but DeepSeek R1 and Manus landed like two consecutive shots of adrenaline, sending fund managers rushing back after the Spring Festival to ask: “Why did we miss it? Where do we find the next 梁文锋?” Research windows compressed and FOMO intensified; some funds may have deployed into more than 10 AI hardware companies within months, even as major disagreements remained over specific teams and entry points.
- “PMF” faded from the conversation not because the problem disappeared, but because AI Coding, Manus, Genspark, Lovart and Plaud finally delivered positive signals in users, revenue or sales. The discussion argued that these cases weakened the assumption that “Chinese software inevitably loses to the US”; 刘旌 said Manus showed that Chinese software products are not inherently worse than those from American startups, and that both young founders who started companies straight out of school and high-ranking Big Tech executives can achieve results at this stage. The market has moved from “giving each other shots of adrenaline” to debating bubbles, validating the line that “critics always look right, while builders often look clumsy.”
- Chinese foundation-model assets may be undervalued, but free competition and resource consumption make that discount understandable. Kimi’s latest round valued it at roughly $4B, versus about $500B for OpenAI; the episode estimated that the “Six Tigers” together should be worth around $20B, below Thinking Machines at roughly $50B. Meanwhile, Doubao had likely surpassed 100M DAUs but remained free, as did Yuanbao, Qwen and Lingguang; Big Tech’s AI spending may already exceed its investment in legacy businesses, and the capital demands of Scaling Law mean a startup’s focus advantage may not be enough.
- The factors truly constraining the next leg of valuation may include a Growth funding gap arriving at the same time as model commoditization. In 2011, DST and similar investors could put $300M into JD.com in a single round; today, a $130M check is enough to shake the market. After the large-scale retreat of Tiger, Coatue, Vision Fund and DST-style Mega Funds, CVCs can fund robotics but may not replace financially driven Growth Funds. More troubling, model prices could fall from RMB10 to RMB1 and then to zero revenue, making it necessary for each financing round to occupy a new narrative: “Even if it does not directly lead to commercialization, you only need to monopolize that narrative.”
- AI hardware and embodied intelligence have become China’s most crowded advantage narratives, supported by supply chains, exit precedents and waves of talent. Insta360’s market capitalization reached roughly RMB150B after listing; P1S, founded about 5 years ago, scaled rapidly; Plaud’s global sales exceeded 1M units, with a high software contribution. The DJI ecosystem has continued to produce founders behind firms including EcoFlow, P1S and Strut. Yet robotics also resembles the “108 Arhats,” with as many as more than 200,000 robotics companies by CCTV’s count; the truly scarce asset remains a product that makes users say, “I want to buy that.”
- Foreign LP interest in Chinese GPs is recovering, but capital is more likely to concentrate in a new generation of firms that can prove their ability to source winners. Monoris raised roughly $288M-$289M plus RMB1.2B, or about $500M in total; Source Code completed a new fund of roughly $600M, with the episode saying its fund life runs 25 years. LPs may use star assets such as Kimi to understand a GP, while European, Middle Eastern and Southeast Asian capital could return; even so, it remains exceptionally rare for a Solo GP to raise $100M for a first fund.
- No mainstream fund is “not investing in AI”; the differences are portfolio concentration, founder preference and the ability to keep following on. Sequoia has maintained its “invest whenever possible” approach, reportedly backing more than 80 companies over the year; Hillhouse has added heavily to robotics and hardware, Matrix has executed aggressively across robotics, energy and consumer electronics, and Qiming has built a strong presence through high-frequency, cross-stage investing. Ant, Baidu, JD.com, Meituan and Xiaohongshu are also writing large checks, but outside JD.com and Meituan’s robotics synergies, most strategic investments have yet to form a clear ecosystem loop.
- A VC’s long-term moat is not a single sector call, but a combination of talent, luck, organizational capability and generational renewal. 刘旌’s standard is to “embrace change while staying true to yourself”; after joining ZhenFund as a Venture Partner, he also downgraded judgment from “the only thing that matters” to one factor among culture, brand, organization and founder relationships. AI will let founders use the same $1M to try 5 times, but it will also make failure more frequent, leaving “resilience” as the quality that ultimately matters; the communications baseline is: “The worst thing is for nobody to know what you are doing.”
Deep dive
1. 2025’s Keyword Was Not Non-Consensus, but Consensus Forming Faster
刘旌’s read on the year’s mood: “Much hotter than last year, even a little too hot.” AI became “the hope of the whole village,” pulling capital, attention and talent toward a high-slope industry and reducing non-consensus at the directional level.
Koji’s question clarified the boundary: consensus around AI, hardware and embodied intelligence does not mean funds agree on specific teams, product forms or entry paths; aesthetic disagreement on individual deals remains normal for VCs.
DeepSeek R1 and Manus jolted funds out of their late-2024 pessimism. Managers called long meetings after the Spring Festival to ask, “Why did we miss it?” But the time available for research and judgment was short, making it increasingly difficult for institutions to invest calmly on their own timetable.
2. PMF Is No Longer a Slogan Because Applications Are Finally Producing Data
The core question in 2024 was: if models keep improving, why have they not produced successful commercialization or strong user demand? By 2025, AI Coding had “given everyone a 100% shot of adrenaline,” while Manus, Genspark and Lovart were also showing traction.
刘旌 saw the change in public sentiment as a shift in phase: last year, people still needed the line “critics always look right, while builders often look clumsy” to maintain conviction; this year, the debate had moved to bubbles, with Zhu Xiaohu even saying publicly that “there will be no bubble within 3 years.”
Koji’s summary: PMF has not become less important; the assumption that it was “impossible to solve” has been broken. The market can now see user data, revenue and global products at the same time, so it no longer needs to repeatedly prove that AI applications can work.
3. Manus and Insta360 Put a Decade-Long Marathon and a Generational Cycle on the Same Chart
What moved 刘旌 most was the nearly 10-year investment relationship between 刘源 and 肖弘: from Nightingale Technology, Butterfly Effect and Monica to Manus, a young investor stayed alongside an even younger founder and made 5 bets, eventually reaching what can still only be called “stage-one success.”
Another coincidence unfolded at Liangzhu’s “Post-00s Night”: the day 李丰 took the stage, Insta360—backed by him during IDG’s “Post-90s Fund” era—listed. Koji recalled that, if memory serves, 李丰 barely mentioned Insta360 at all. The exit of a Post-90s project from 10 years ago and the arrival of a new generation of Post-00s founders created a powerful sense of succession.
Generational competition is also spreading between LPs and GPs. At an overseas event, an LP rushed up to Monolith’s 曹曦 and said, “I’m willing to do anything.” Behind it was a positioning battle among investors seeking a slot with “China’s top 3 new-generation GPs.”
4. The Big-Model Capital Story Is Not Over; the Valuation Logic Has Simply Become More Brutal
Looking back at the “Six Tigers,” new updates from 01.AI and Baichuan have thinned out, MiniMax and Zhipu are pursuing listings, while Kimi—once thought to be nearing the end of its story—completed another financing round. 刘旌 inferred from related reports that the capital narrative still has more chapters to come.
The valuation gap is enormous: Kimi is worth roughly $4B, OpenAI about $500B, leaving the former at less than 1% of the latter. The episode estimated that China’s “Six Tigers” together should be worth around $20B, still below Thinking Machines at roughly $50B and not directly comparable with Mistral.
Koji offered “Chinese assets are undervalued” as one explanation. 刘旌 retained a layer of caution: “One of my fallacies—though I may be wrong.” The discount reflects both the supply and pricing difficulties of China’s primary market and the diversion of capital into embodied intelligence, hard tech and other sectors.
Unlike US VCs, which may have to concentrate almost entirely on AI, Chinese investors have other sectors in which to allocate capital. ByteDance and other Big Tech companies are also present, so scarce capital and competition from giants are jointly compressing startups’ pricing power.
5. Free Big-Tech Products Have Put Chinese Model Companies in a Structural Commercialization Trap
刘旌 highlighted a frequently overlooked fact: “Doubao should have surpassed 100M DAUs by today, but it is free software,” and may remain free for a long time. Yuanbao, Qwen and Lingguang are also free, making it difficult for the Six Tigers to charge users against that reference point.
A startup’s traditional advantage is having fewer resources but greater focus. In this cycle, however, Big Tech founders may already be betting more on AI than on their existing businesses. Add in the compute and capital required by Scaling Law, and the lead that focus can buy deserves to be reassessed.
Alibaba once invested in 5 of the 6 major model companies, but is now going all in on the Qwen model and Qwen App at the same time; Ant is also pushing Lingguang aggressively. The Big Tech firms that once supplied startups with capital have entered the competition themselves, blurring the identities of capital provider and competitor.
6. The Growth Funding Gap and Model Depreciation Are Squeezing Valuations at the Same Time
The expansion and competition of the mobile internet era could keep burning cash because of Mega Funds such as Tiger, Coatue, Vision Fund and DST—not early-stage VCs. DST and similar investors putting $300M into JD.com in a single round in 2011 reflects a level of funding that is now rare.
CVCs such as Meituan and JD.com can continue funding robotics companies, but 刘旌 worries that without market-based Growth Funds driven by financial returns, model, software and hardware companies will struggle to sustain higher valuations. Foreign capital may return if it sees predictable returns, but the exit path still has to work.
Models are “seemingly among the fastest-depreciating products in history”: shortly after o1 launched, DeepSeek R1 caught up and open-sourced. A capability might sell for RMB10, then RMB1 6 months later, and possibly generate no revenue 6 months after that. Financing premiums therefore depend on a continuous stream of new narratives rather than stable technology rents.
7. Chinese Software Is Proving Itself Through Products, Not Sentiment
The longstanding assumption was that “Chinese software companies can never beat American ones.” Products such as Manus, Genspark and Lovart have shown the market that products built by Chinese teams are “not necessarily worse than those from American startups.” Manus’s breakout also generated national confidence: “So China can actually make a product this polished and international.”
MAX and JSPAR have similar product forms, visual logos and even investors, but their founders represent different archetypes: 肖弘 is the “young prodigy” who kept building after graduation, while 景鲲 is a Big Tech veteran at a high P level. Both paths have produced results at this stage, weakening the case for a single ideal founder profile.
Technology teams still often need a “composite”: alongside scientists, they need someone like 张予彤 to take on COO responsibilities, fundraising and external communications. Funds disagree sharply on whether such combinations can work, making this a genuine project-level judgment beneath the directional consensus.
8. “Post-97s and Post-98s” Is Both Talent Thesis and Fund Marketing Language
Yunqi launched the Y Transformer program to find Post-98s, BlueRun recently emphasized Post-97s in its startup camp, and ZhenFund continues to run Post-00s programs. The logic is that these people spent several years training at Big Tech before encountering AI while still young, making them potentially the most AI-native cohort.
刘旌 remains skeptical about whether narrowing the cutoff to 1997 or 1998 is scientifically meaningful, asking: “When IDG ran the Post-90s Fund, was that really the victory of Post-90s founders, or the victory of the mobile internet?” Koji’s response: “Each generation encounters its own opportunity.”
Even if birth year is not the decisive variable, the framing still has value for funds: in a fully competitive GP market, a convergent marketing message lets founders and LPs quickly understand “who you are looking for and what you believe.”
9. Behind the AI-Hardware Consensus Are Exit Precedents and Talent Spillover from DJI
By the second half of 2025, AI hardware was close to universal VC consensus. One A16Z China employee even said bluntly that he would focus on Chinese hardware companies because “this is what China should be doing,” although whether the institution can invest is a separate question.
Confidence comes from a string of precedents: Insta360 reached roughly RMB150B in market capitalization after listing; P1S grew rapidly within about 5 years of founding; Plaud still looked like a “make a quick buck” story last year, but global sales have now exceeded 1M units, with a high software contribution.
Talent supply is equally important. People trained in the smartphone supply chain and the DJI ecosystem are starting companies in waves. P1S founder 陶冶 said DJI worried that new graduates might have to choose between DJI and P1S; that competition itself shows the new organization has become an attractive destination.
Founders carrying the “DJI label” are highly popular in the capital markets, from P1S and EcoFlow to Strut’s automated wheelchair. Continued talent movement could produce more Insta360s and P1Ss. 刘旌’s view: “Talent will appear in groups.”
10. Overseas LP Interest Is Recovering, but It Is Chasing Only Recognizable Star Assets
Monoris’s new fund consisted of roughly $288M-$289M plus RMB1.2B, or approximately $500M in total; Source Code announced a new fund of about $600M, with the episode saying its fund life reaches 25 years. These are rare pieces of good news in a difficult fundraising environment.
An overseas report did not headline the story as “Monoris raises $500M,” but as “Kimi-heavy fund raises $500M.” 刘旌 inferred that overseas LPs still understand a Chinese GP through its most prominent assets.
Pure US-dollar capital may find China harder to enter, but capital from Europe, the Middle East and Southeast Asia remains plentiful. LPs are not merely waiting; they are actively looking for deployment routes. If the countries where AI is truly investable are mainly the US and China, China remains difficult to exclude from the allocation.
11. Embodied Intelligence Is at Its Most Prosperous—and Most Vulnerable to Becoming the “108 Arhats”
Koji once showed a city-by-city map of robotics companies: 7 or 8 each in Beijing and Shanghai, roughly 40 or 50 nationwide, visually resembling the “108 Arhats.” The spectacle was striking, but the underlying problem was blurry faces, homogeneous products and weak commercialization—“not one that makes you want to own it.”
SUN’s launch created a different reaction. HONI ZHAO and 陈驰 demonstrated products that produced a clear “WOW moment”: “This is the kind of thing I want to buy.” It brought the technology showcase back to user desire.
刘旌 stressed that the robotics boom has a strong policy and ideological thrust, similar to the low-altitude economy, though it may be more sustainable. Koji said he had seen a CCTV News headline a few days earlier claiming China already had more than 200,000 robotics companies. Regulators have begun calling for less homogeneity, but both men agreed that it is “too hard to avoid.”
Even if real-world deployment still takes time, capital markets may continue to provide positive feedback in the short term. Moore Threads’ IPO performance was extremely strong, and if Unitree lists successfully, 刘旌 expects it to “very likely be a pretty good result as well.”
12. Sequoia, Hillhouse, IDG, Qiming and BlueRun Chose Different Degrees of Aggression
Sequoia has maintained the internet-era approach of investing whenever possible, reportedly backing more than 80 companies over the year. Miles invested early, and J S Park later followed on. Hillhouse has also been highly active, with extensive robotics and hardware exposure; 张磊 has personally met very young founders.
刘旌 believes IDG’s brand has been undervalued over the past 5 to 10 years. It is no longer merely a VC, but a capital-management platform, with Insta360, Chipone Technology and P1S continuing to produce returns. Its approach is not indiscriminate; it remains selective.
Qiming has always been rooted in healthcare and technology, while also managing the Beijing Artificial Intelligence Fund, which may give it a different lens when evaluating companies. Its overall investment pace is also highly active.
BlueRun entered China in 2005. 刘旌 had not previously felt a strong presence from it, but over the past 2 years he has described it as “extremely ambitious and hardworking.” Its new RMB fund exceeds RMB2B, and its brand, projects and events have all reached the top tier.
13. Matrix, Qiming Venture Partners and Early-Stage Funds Showed Another Set of Allocation Methods
Matrix invests beyond AI and is one of the few US-dollar VCs to have genuinely executed a hard-tech strategy and received feedback from it. Manner is approaching an IPO, its energy investments began early, and it has been highly active in robotics and consumer electronics. By contrast, it has invested very little in foundation models—only MiniMax.
Qiming Venture Partners’ defining feature is that it is “very loud,” with its brand everywhere and investments extending from first rounds into growth. Its portfolio is broad and frequent. Gaorong is relatively understated, but it continues to invest from Kimi during the foundation-model era through newer projects.
ZhenFund’s significance lies not in fund size, but in the thermometer effect it may have as one of China’s earliest early-stage or angel funds. 刘源 said the past 1 or 2 years may have been the best period in ZhenFund’s history, with multiple companies preparing for IPOs and different partners receiving positive feedback.
5Y Capital has shifted from emphasizing overweight positions and cautious deployment to allocating more densely. Source Code operates both a main fund and Source Code Rhythm; BAI previously invested in Mexico and Latin America, then sent a renewed signal of attention to mainstream Chinese founders by investing in hardware companies such as Looki. 刘旌’s summary of Monitor: “An extremely strong tidal style.”
14. Big-Tech Strategic Investors Can Write Large Checks, but Have Yet to Build New Ecosystem Loops
Ant has invested in projects including Zhipu, Kimi, PixVerse, Liblib, Stardust Intelligence and Future Intelligence. Baidu provided roughly $30M when Liblib was at its most difficult point, turning the situation around, and also invested in companies such as DeepWisdom and Wuwen Xinqiong.
Koji asked about the long-term value of these investments to the parent companies. 刘旌’s candid answer: “Right now, it is actually hard to see.” Baidu has yet to build a Tencent-style ecosystem feedback loop, while Alibaba’s early ecosystem ambitions around foundation-model investments ultimately gave way to building Qwen itself.
JD.com and Meituan have relatively clear business synergies from their embodied-intelligence investments. Meituan has invested in companies including Stardust Intelligence and XYZ Robotics. Outside those cases, most investments in hardware, agents and software currently look closer to financial investments.
Xiaohongshu is one of the few clear exceptions. The episode said its profit has exceeded $1B and that it has ample cash, while favoring early-stage products aligned with Xiaohongshu’s native sensibility, including flying companion robots. It has also become an important channel for AI startups to build brands and for investors to find deals.
15. Solo GPs and New Accelerators Are Filling Organizational Gaps
New GPs include 陈哲’s AlphaList, 温永腾’s new fund, Queen Stone and “Little Little Fund,” most of them operating as Solo GPs. AlphaList reportedly raised roughly $60M-$70M and received substantial backing from 刘芹, but reaching $100M for a first fund remains exceptionally rare today.
The obstacle is not only money, but a gap in track records. One former boss asked an investor preparing to go independent: “You have been investing for years without a single exit—what gives you the right to run a fund?” The path from investing through IPO to exit takes nearly 10 years, and recent years have not given young investors enough time to complete the cycle.
New sources of capital also include the family office that 赵长鹏 said would be sized at $10B, whose incubator reportedly provides $500,000 per project. 王慧文 has also assembled a small team to invest seriously, but the market currently lacks enough projects capable of absorbing capital at that scale.
Koji sees the new accelerators as more product-focused. Shenzhen Innovation Academy has incubated roughly 80 teams, with 30 receiving angel investment. Spark Club takes 10 teams per cohort and houses them together for 42 days in a roughly 600-square-meter, 5-story townhouse in Shanghai. MiraclePlus is reportedly approaching its first IPO, humanoid robotics company Agile Robots, but because it entered at Series B, the deal is not enough to validate its early-stage model. Hypershell’s latest financing should have reached roughly $70M.
16. The Scale of US AI Financing Reflects the Contraction of Chinese Early-Stage Capital
Koji casually reviewed a week of US financings: AI search company A by former Twitter CEO Parag Agrawal raised $100M in Series A at a $740M valuation; Unconventional, founded by former Databricks AI chief Naveen Rao, was valued at $5B and was targeting a $1B raise.
Any one of these cases would be “unimaginable” in China, while in the US it is merely one of many headlines. Young Chinese founders can choose among US-dollar funds, RMB funds, state and government capital, and Big Tech, but the total pool of market-based US-dollar early-stage capital—the money everyone most wants—has clearly shrunk.
Good news about several VCs’ new fundraises failed to generate broad discussion, which is itself another signal. Funds that once routinely raised more than $1B are now much smaller; the silence may reflect not a lack of highlights, but a shrinking population of primary-market professionals and readers.
17. Good VCs Start with Talent and Luck, Then Need Organizations That Can Survive Generational Change
刘旌’s view is direct: “VC requires some talent.” Like journalism, some abilities cannot be fully supplied by effort alone; luck early in one’s career matters too. Someone investing in consumer companies in 2020 may have received negative feedback simply because of the cycle; the online education companies he liked disappeared because of what happened later, while the industry ultimately “only asks about results.”
VC also follows a power law. It is not winner-takes-all, but the very top funds still take the largest share of returns. 刘旌 therefore asks: “Does this industry really need so many people?” The positive side of GPT’s emergence is that it reconnected the “severed string” left after the consumer downturn in 2022.
The rarest quality in a good GP is to “embrace change while staying true to yourself.” That means deciding whether to chase younger founders, build a brand aggressively or imitate the new-media teams of US VCs, while preserving one’s own overweight principles. The cycle is long, the noise is high, and organizations can easily lose themselves in the process of change.
Funds are highly personalized: “An organization is an extension of a person.” When a founder steps down, the name may remain, but taste, relationships and judgment may not be transmitted. Mornings has replaced almost an entire cohort of young investors between the Post-95s and Post-00s over the past 2 years, showing that funds with more than a decade of history are already facing calcification and generational pressure.
18. The AI Era Will Ultimately Reward Resilience, Visibility and Continued Action
After joining ZhenFund as a Venture Partner, 刘旌 revised his earlier view: judgment is not the only variable in early-stage investing, but one factor among organizational capability, culture, brand, founder relationships and post-investment support. Content, community and investing can generate synergy with one another.
After looking back across more than 100 companies, 刘源’s answer was not intelligence or romance, but that “resilience is still the most important thing.” AI may let the same $1M support 5 attempts, but it also lets products be disproved and drowned out faster. Whether a founder can stand up again after the first 4 failures matters more than ever.
Communication is not the opposite of resilience. ChatGPT and Sam Altman generate 3 or 4 headlines every week, and Sam Altman and Jony Ive have appeared together 4 times in recent months, even though the product will not launch for another 2 years. 刘旌’s baseline is: “The worst thing is for nobody to know what you are doing.”
Koji keeps optimism within bounds. Encouraging people to quit their jobs and start companies is like a “siren,” because entrepreneurship remains a 9-out-of-10 failure proposition. What he truly encourages is positive action: “Action and entrepreneurship are 2 very different things.” AI and reliable teams free up time, but the final task is still prioritization—and the simple self-check: “Once I start taking action, I feel better.”