Vol.233 Media, Communities and Investing: Talking with 张鹏 About Geek Park’s History and Its Unitree Investment

Summary

• 张鹏’s core read on the terrain of this AI startup cycle: mobile internet was “filling potholes,” while AI and hard tech are “climbing uphill.” The previous generation maxed out connectivity through smartphones: “Water flows down from the mountain; find a hole, dig a channel, and fill it—that is your value.” In this generation, the paradigm has not converged and the mountain is still growing. “The biggest risk is becoming the loading program”—the mental model can change every 6 months, so iteration speed must far exceed your imagination of the end state.
• AI should be compared with the personal-computing revolution of the PC era, not with mobile internet. “We can now mass-produce intelligence; this is a computing revolution.” The first question for model companies is whether they can stay in the game—there will not be 100 model companies globally; “I think there will be 10, perhaps even 20.” If others have “broken through the atmosphere and entered the AI-improving-AI phase” while you have not caught up, the gap could keep widening. Embodied intelligence has not even converged on a data paradigm; the bet is on “mass-producing real productivity.”
• Unitree validated Geek Park’s playbook of putting the community ahead of capital. 张鹏 spotted 王兴兴’s hand-built robot dog on a niche public account with only a few hundred views—2 years earlier, he had taken founders to MIT to see a robotic cheetah—then flew to meet him and talked over paper cups of water on broken sofas in a hallway. After the lead investor walked away, he still put money into Unitree first: “My action may have been right, but the logic may have been very foolish.” Unitree later stood on its own.
• Variable Capital began with 大众点评 founder 张涛’s 2017 wake-up call: “You invited Musk, charged RMB2,000 for admission, and that was it?” Geek Park’s review found that its early selection record was strong, so it turned recognition into action: “I had already invested in him before giving him money; giving him money converted the option into formal equity.” That is also why Geek Park does not make FA its core business: it seeks continuous convergence and investment responsibility, not a project-based conveyor belt of deals.
• The predicament of the AI 1.0 generation was that it had to carry commercial workloads while still in kindergarten. AI had begun to emerge after AlphaGo, and OpenAI was already exploring, but the prevailing value system demanded rapid monetization and project delivery, leaving little room for Frontier Lab-style exploration. Researchers were forced to shine in areas that were not their strengths. Today’s model companies—MiniMax, Kimi, DeepSeek and 智谱—are doing the same work in a “better era.”
• There will “always be bubbles, and there must be”: emergence needs the energy of waste, but how individuals use a bubble is the dividing line. The early-stage failure mode is “not necessarily starting too late, but starting too early.” Uphill projects require “cumulative certainty,” not valuations reverse-engineered from public-market comparables. A positive example is Horizon Robotics’ 余凯, who raised a large amount of money around 2020 and then anticipated a winter: he “turned other people’s misunderstanding into the correct course of action.”
• 张鹏’s advice to mobile-internet veterans: “Do not become a ‘deng’”—pursue evolution, not a switch. “What makes one person think he can dominate the field alone for 10 or 20 years? That is pure delusion.” Anthropic has plenty of “mid-career veterans”; AI-native is an adjective that can be learned, while entrepreneur is a noun that is hard to replicate. “The cost of starting something has collapsed, but the difficulty of doing it well has not fallen at all.” A warning from the other side: 张斯马克 once showed at a conference how one person could spend $15,000 in tokens over 1 week to rewrite an AI-native office, rapidly washing away pure-software assets.
• This AI cycle is also restructuring VC itself: investment is inference based on historical datasets, and the winning models of consumer internet may not fit AI, embodied intelligence or hard tech. “China’s investment circle undergoes mitosis every few years.” Strong investors leave to start their own funds, exposing an unresolved question about whether the original institution’s production relations allow the people creating the most value to share enough of it.

Deep dive

1. Opening: Geek Park’s iconic moments and 张鹏’s quoted self-doubt

• 李翔 recapped Geek Park’s history: before launching Xiaomi, 雷军 gave a talk about phones and conducted live research at Geek Park; in 2014, an unknown 张一鸣 shared a stage with Musk; in 2017, 张鹏 organized a gathering in Wuzhen where the still-early-stage 王兴兴 demonstrated Unitree’s robot dog to 雷军 and 王兴. Geek Park was also one of Unitree’s earliest investors.
• 张鹏’s modesty is not a courtesy ritual: “The real stars are all these entrepreneurs…we are mostly recorders and bystanders.” Many young people “may genuinely not know Geek Park”; those memories are already more than a decade old.
• He eventually stopped trying to convince people that “Geek Park is not media, but a community.” “Who you ultimately are will show up in what you do and the results you produce.” 李翔 compared it with a16z, which manages more capital than Sequoia but often calls itself a media company. 张鹏 agreed: “Say what you lack—that is the most truthful state of affairs.”

2. Entering the industry in 1998: From the “factory manager world” to China’s frontier VC era

• 张鹏 graduated in 1998 and joined IT Manager World, which had launched that same year. A classmate wrote the publication’s name in his address book as “Factory Manager World”—IT was still a newly emerging term associated with the advanced economy. Most of the editorial staff held master’s or doctoral degrees, and he was assigned a “smaller area”: the internet.
• China had no established concept of venture capital at the time, aside from a handful of firms such as IDG. Sequoia did not truly expand into China until after 2000. After roughly 2 years on the job, he compiled 77 Internet Business Models. “That book did not have much of the meaning of a book; it was more like a collection.”
• The best-known founders of the era included 王志东, 田溯宁 and 张朝阳. “It was hard to see so-called ordinary people starting companies; the initial energy was still very strong.” 李翔 added China Dawn, written by David Sheff, the journalist who had interviewed Steve Jobs for Playboy.
• 张鹏’s foundational observation was that the internet’s “connectivity” was a huge dividend. “Most of the startups that developed in China on the back of the internet during the first 10 years were riding that dividend.”

3. Early VC had no clear logic either, but cycles resonate

• 张鹏 sought out VCs because he assumed “people who put up money must at least understand something” and could compensate for the limitations of a reporter. Instead, “I did not hear any particularly clear logic from them either. Many things that ultimately succeeded did not succeed according to the logic they had described at the time.”
• The reason an investment wins may not be the reason the investor thinks it will win, and the reason a company eventually succeeds may already have shifted.
• That led him to understand that what a founder says today “may make sense at this stage, but once the environment changes, he has to adjust. That does not mean he was deceiving you or that he was wrong; he has to change.”
• His view of cycles: VC only began to look like an industry in China after 2005. Mobile internet gestated after 2010, first broke out in 2011–2012 and peaked around 2014, resonating with the boom in the fund industry and its first full investment cycle. “Entrepreneurship is also like farming; the seasons and the environment matter enormously.”

4. A technology gap forced capital back into consumption

• 李翔’s experience was that when he encountered GGV in 2007–2008, it had invested in companies such as Qunar as well as a batch of consumer businesses. One was Biostime, which some people called “the future Coca-Cola of China.” It resembled 2021, when AI had yet to take off, mobile internet was nearing its end and capital flooded into milk-tea, gold and beverage companies.
• 李翔 first formed a serious idea of venture capital around 2005, after Global Entrepreneur covered Sequoia Global managing partner Michael Moritz, Sequoia China’s 沈南鹏 and 张帆, and KPCB managing partner John Doerr.
• He later developed a more intuitive understanding through GGV. A colleague told him the firm had invested in Alibaba, and his reaction was: “That is impressive.”

5. 王超’s methodology: Study how something works, not who made money

• IT Manager World editor-in-chief 王超 gave 张鹏 a worldview: “You need to look at how something works, not just at the result.”
• 张鹏 once spent 3–4 days reporting at Baogong Logistics and recorded more than 10 tapes, asking concrete questions about how its IT system went live and whether frontline operators found it useful.
• The experience shifted his attention away from “who made money, who went public and whose valuation rose again” toward the details founders actually face: how something operates and who is doing what new thing.
• He distinguishes newspapers from magazines. Newspapers are designed for information circulation; magazines should, in principle, circulate understanding. Depth means going to “places no one else will write about”—details that cannot be seen without being there and that are directly connected to how the thing gets done.

6. He left not because of an opportunity, but because of despair

• Asked by 李翔 whether working in media made him happy, 张鹏 answered directly: “I was not happy.”
• IT Manager World generated annual revenue of several tens of millions of RMB and roughly RMB20M in profit. 牛文文 had once used annual revenue above RMB100M as the standard for a major publication. But around 2009, the media industry began to decline, peripheral media businesses elsewhere in the group were losing money, and the company began shrinking.
• 张鹏 once requested RMB1M to pay bonuses to his team. He was not trying to distribute money among a few people; he wanted to retain strong colleagues and keep improving the content. The company did not support him.
• The trigger was the loss of talent. 李娜, whom he had managed, later became a managing director at Yunfeng Fund; 周源 later founded Zhihu, and after his first startup failed, 张鹏 helped arrange for him to take a trip to Tibet—“when he came back, he was a different person”; 顾伟伟 later worked on Mobike.
• His conclusion was unvarnished: “I simply became desperate.” He had not discovered a particular opportunity or dream. Under the existing business model and governance environment, he could no longer see a way to keep doing the work well, so he left with 刘湘明 and others to start a new magazine.

7. Business Value and the starting point of Geek Park

• In 2010, 张鹏, 刘湘明 and others launched Business Value. 陈天桥 and 周鸿祎 had both supported him because he had written their first covers for IT Manager World. “All that was left after so many years was a little personal goodwill.” Each may have invested several million RMB, but they may not have viewed it as an investment—just a gesture of support.
• They put full articles on Weibo for free relatively early because “understanding is not sold by the kilogram, nor by the article.” Understanding first had to be transmitted, after which the community could be used to regroup the right people.
• Geek Park initially existed only as 张鹏’s personal interest. He ran a website with an intern, and it did not become a company until 2014.

8. Geek Park initially wanted to build a product; only after failing did it move offline

• The initial concept was close to a Twitter-style product: condense information already online, select what was worth attention, spark discussion and add a “bet on the future” mechanism for users to judge whether a company or plan would succeed.
• The view at the time was that centralized content production could no longer meet demand and that users needed to participate in a Web 2.0-style way. After the product operated for 2–3 months, the emergence of Weibo made 张鹏 realize the path was wrong.
• After the pivot, they moved the participation that had not materialized online into the physical world: bring a group together to discuss rather than have one person finish an article for everyone else to read.
• “There has to be an atmosphere, a vibe, for people to want to discuss together; and the topic has to be set up properly for people to want to participate.” That became the beginning of his understanding of community.

9. Sony Explore Science’s basement: The value of a closed-door setting

• Early events borrowed the basement of Sony Explore Science in Chaoyang Park. They held 2 sessions a month with 40–50 people each, and even after registration, participants were screened for fit with the topic.
• A defining scene was the closed-door exchange before 雷军 launched the Xiaomi phone. 雷军 spoke for nearly 2 hours, followed by a vote on “who had been persuaded by Lei.” Fewer than half raised their hands. “Even after talking face-to-face for 2 hours—and I thought he had made a very good case—he persuaded only half the room.” Xiaomi’s phone nevertheless became a major success.
• 张鹏 believes a closed-door setting is safe and candid. “Only half the people raised their hands, and he did not feel embarrassed because no one outside could see.” 王兴’s first interview also took place in that basement; many clips circulating online came from that session.
• In another scene, 周鸿祎 warned the founders sitting below not to publicize a product as soon as it was built, lest 马化腾 see it and turn them all into Tencent product managers. 张一鸣, sitting on the floor, silently took note: speak only after the product had reached a certain stage. 张一鸣 later told 张鹏 this story himself.

10. Starting from obscurity: A highly personal bet vindicated by the era

• Asked by 李翔 how he resisted the temptation to report only on successful people, 张鹏 attributed it to sample size. When he entered the industry, he saw an internet that did not look especially reliable; after seeing enough of it, he still believed there was value there.
• His 11 years without changing jobs gave him exposure to both winners and non-winners, gradually building a dataset about people.
• His target reader was always clear: in the information age, he wanted founders to keep IT Manager World on their desks; in the mobile-internet era, he wanted them to read Business Value. When the magazine launched, some colleagues questioned whether these people would ever buy ads and whether anyone would read articles about them. 张鹏’s answer was: “I am fairly self-directed; I thought this was what we should do.”
• 2 years later, as mobile internet accelerated, Business Value generated more than RMB20M in annual revenue and came close to breakeven.
• Choosing a cover subject was itself a judgment and an aesthetic choice. 王兴 appeared on the cover while building Meituan; 雷军 appeared while building a phone, around 2011. In 2014, 张鹏 invited Musk to China and chose to put 张一鸣 onstage with him, signaling that he believed 张一鸣 deserved to be the main character on that stage.
• 李翔 drew a parallel with a biography of Hassabis. The publisher had hesitated about putting Hassabis on the cover because “who knew who he was?” 张鹏 acknowledged that judgments can be wrong, but giving someone a cover or a stage before they are famous is itself a judgment.

11. Information → understanding → action: “I am not a media person”

• 张鹏 believes the core deliverable of business content is understanding; after delivering understanding, the next step is action. If you merely put content out into the world without acting yourself, you cannot capture the larger value.
• 李翔 cited Caixin, while noting that he did not know whether this remained the case: Caixin reporters were previously barred from buying shares in companies they covered because it could create a conflict of interest.
• 张鹏 respects and supports that principle, but precisely because he respects the social mission of traditional media, he does not consider himself that kind of media professional. “I do not carry that mission.”
• What he prefers is to meet strong people through content, gain useful understanding and turn that understanding into effective action. 李翔 called him a “low-spec McLuhan,” while 张鹏 emphasized that he was simply doing the things he wanted to do well.

12. Filling potholes versus climbing uphill: The fundamental difference between two eras

• 张鹏’s core framework is that mobile internet directly maxed out connectivity, like an operating system that was already ready. Founders only needed to develop software on top of it, find use cases and design business models.
• “The terrain was water flowing down from the mountain. Find a hole, dig a channel and fill it—that is your value,” because connectivity came with its own gravity.
• Today, AI and hard tech form a “double helix,” but AI is still growing and the technical paths in areas such as embodied intelligence have not converged. “This wave is not about filling potholes; it is about climbing mountains and slopes.” One mountain is still growing while the founder is trying to climb it.
• The difficulty for founders has increased “by more than a little,” but accumulated wealth and completed startup cycles mean the resources and conditions supporting founders today are also better than before.

13. AI is a computing revolution: Compare it with the PC, not mobile internet

• 张鹏 says AI should not be casually compared with mobile internet; it is closer to the birth of the PC, a personal-computing revolution. “We can now mass-produce intelligence, and that intelligence creates value through computation.”
• The PC-to-internet transition took years and solved a huge number of problems. AI today has technical consensus, capital and strong founders, but it is still in a complex uphill climb.
• 李翔 inferred that the people currently at the table may amount to little compared with the companies and entrepreneurs that will emerge. 张鹏 added from a biological perspective: “Being around for a long time is itself an original sin; being old is an original sin,” because an individual rarely completes an evolution within a single lifetime.
• The goal today should not be to insist on immediately building a company that will last 20 years. It should be to find a core point that grows stronger over time and can also expand horizontally.

14. The biggest risk: Becoming the loading program

• “In the previous era, everyone feared being too slow in staking out territory. Today, the biggest risk is becoming the loading program of that era.” In other words, you once had value, but as the era moves forward, you become the waiting process for the next era.
• AI’s frameworks are changing rapidly: from Prompt Engineering to Context Engineering and Harness Engineering, a new mental model can emerge roughly every 6 months and reshape engineering thinking and product form.
• Mobile internet was different. Its basic framework was relatively stable; recommendation algorithms and other changes appeared in the middle, but not a continuing series of shifts on the same scale.
• Founders therefore need an iteration speed far greater than their own imagination of the end state.

15. The community gene is implicit: One conference spent all the money

• Around 2013, Geek Park held its first large-scale conference inside the giant tank at the 798 Art District. About 1,500 people attended; network access, toilets and construction all had to be arranged temporarily, including renting mobile toilet trucks. The cost was roughly RMB3M–RMB4M.
• Geek Park’s first round of support funding may have totaled only RMB5M–RMB6M, so one conference consumed almost all its cash. 张鹏 insisted on putting founders who were not yet mainstream stars on a carefully built stage and making them feel like stars.
• The community instinct “does not necessarily show up in the business model, but it shows up in actions.” The investment in design and aesthetics was not meant to showcase 张鹏; it was meant to honor the founders.
• The conference almost never lost money, but it was never treated as a profit product. It was the community’s annual gathering. They nearly became a professional conference organizer, then realized that exhibition companies had better scale and efficiency for exhibitions, and that the revenue was not aligned with Geek Park’s goal of delivering understanding. They abandoned the path quickly.
• 李翔 noted that spending more than RMB3M on a conference resembled giving Unitree money first after the lead investor withdrew. 张鹏 acknowledged that decisions often look brilliant in hindsight, while at the time they rely on confidence and may rest on logic that is “very foolish.” He had originally thought he could help Unitree raise more money, but Unitree in fact stood on its own through 王兴兴.

16. 张涛’s question: “You invited Musk and charged RMB2,000 at the door?”

• In 2014, during a Geek Park community event at Tesla, 张鹏 arranged the first private meeting between 王兴 and 张涛, who were competing at the time. He told each that the other would be there and wanted a chance to talk, helping establish basic trust between them.
• In 2017, 张涛 asked him: “In 2014, you invited Musk, put 张一鸣 onstage with him and then charged RMB2,000 for admission, right?” 张鹏 felt the remark sounded like criticism.
• 张涛 advised him to review the people Geek Park had selected and the guests it had invited to its conferences. If its early judgment accuracy was high, it should turn that judgment into investment action rather than only write articles and hold conferences, since those were merely content delivery.
• After reviewing the record, 张鹏 concluded that Geek Park’s selection of people and topics from 2010 through 2014–2015 had indeed been relatively accurate. This was not because he was smarter than others, but because daily exposure to early-stage founders produced a more continuous signal.
• Several founders who shared Geek Park’s views on technology trends and entrepreneurs then supported the launch of its first fund. Variable Capital began forming in 2017.

17. Becoming a GP for the first time: Only founders’ money, no institutional LPs

• 张鹏 did not approach large institutions partly because he lacked confidence. “This was my first time; I did not know whether I could make money, so I could not deliver a certain value proposition to them.” He mainly approached founders he knew, explained his background, investment direction and uncertainties, and asked them to place their trust in him.
• Rejection was also a form of alignment. If someone cared intensely about the team and professional investment capabilities, the project was not ready. Those willing to support him were willing to try with an angel-investor mindset.
• 张鹏 emphasizes not raising too much money at the beginning. Fund size itself becomes a constraint: limited ammunition means a limited number of investments.
• Investing cannot be driven only by the thought, “I might invest in the next 张一鸣.” That would be “insane.” More important is asking what happens if all the money is invested incorrectly.
• His early general problem was that he liked almost any project with an innovative element. He later understood that investing also requires assessing risk and the cost one can bear. Investment is like crossing a mountain range into another industry; at minimum, you must lower yourself first and recognize that you may not see through it.

18. Why FA is not the core: A conveyor belt does not fit convergence

• Many people thought Geek Park was suited to FA, but after studying the business, 张鹏 did not make it the core. An FA firm’s task is to sell a startup efficiently. It must meet a large number of companies continuously, creating a conveyor belt of deals.
• Geek Park wants to push its judgment further upstream. It does not want only to perform a rough screen; it wants to keep understanding the founder and the business so that its view continuously converges. “You keep going deeper, and ultimately you see more and more clearly.”
• The responsibility structure is also different. “Media does not have to take responsibility—people remember what you got right and may forget what you got wrong. Investment must take responsibility. Your judgment must converge and be accurate before you are entitled to earn excess returns.”
• 张鹏 also distinguishes projects from products and systems. Every article and every ad in media can become a project; a project-based model constantly dilutes bandwidth and talent density. A fund is closer to a product and a system: management fees resemble a subscription fee, value is shared through results, and LP trust, fundraising capability and investment judgment can compound over time.

19. Geek Park’s first principle in investing: Giving money converts an option into equity

• 李翔 argued that investing is fundamentally about access: whether a strong founder is willing to accept your money is itself a threshold, and Geek Park may have obtained that ticket.
• 张鹏’s view is that Geek Park had already begun investing before it transferred money. Serious reporting, free community events and helping a founder at a critical moment were all investments.
• “Giving money merely converts the option you had already invested in into formal equity.” But if you perpetually provide content and personal goodwill without converting the option into equity, that also has no meaning.
• Therefore, Geek Park’s first principle in early-stage investing is not the universal first principle of the investment industry. It is to “do the content well, build the community well and genuinely care about founders.” Keep doing that, and at some point a person or opportunity will appear.

20. Fund I mainly bet on non-consensus; Brownian motion was actually a good vintage

• Unitree was probably the second or third investment in Fund I. The fund had no star founders; most of its investments were non-consensus.
• One category consisted of hard-tech projects that did not yet exist in the market but that 张鹏 believed would eventually be needed. The other involved the go-to-market ecosystem for overseas expansion, related services and early-stage intelligentization projects.
• They did not participate in the local group-buying or sharing-economy models of the time because “the lowest-resistance link in the electronic world had already been filled.” Those businesses were entering the high-friction physical world, involving ground operations, finance and complicated accounting beyond Geek Park’s familiarity.
• In a super-consensus wave, a small early-stage fund has little room to participate. The mixed, “Brownian motion” environment of 2017–2018 was instead well suited to Variable Capital’s formation.
• 张鹏 also mentioned 傅哲宽. After investing in Momo, 傅哲宽 felt there was little left to look at in mobile internet and began surveying projects broadly. He later went to meet 陈天石 after hearing about a financing, but did not invest; he went out of curiosity.

21. Finding 王兴兴: A niche account, MIT’s robotic cheetah and broken hallway sofas

• 张鹏 says, “I really have to thank 张小龙.” He saw 王兴兴’s robot dog on a niche public account with only a few hundred views.
• 2 years earlier, he had visited MIT to see a robotic cheetah. He then invited Professor 金相培 to a Geek Park conference, and in the third year saw 王兴兴 build something similar. It clearly did not look like the product of a heavily funded large company; it looked hand-built. 张鹏 found him and flew over to meet him.
• 王兴兴 had no decent office and nowhere proper to sit. The two talked on 2 broken sofas in a hallway, each holding a paper cup under the water dispenser.
• At MIT, 张鹏 had already been thinking about Boston Dynamics’ path. Hydraulics and electric drive represented a major divide; before 2015–2016, robot algorithms were essentially written line by line. Later, technologies such as force control made them increasingly amenable to mathematical solution and training, creating the possibility of generalization.
• He believes the community also created the opportunity. MIT might not have received him alone, but it was willing to receive a group of Chinese founders he brought.
• His first impression of 王兴兴 was that, only 1 year out of his master’s degree, he understood technical paths far beyond what his age suggested. He was “very bad at telling stories, but exceptionally good at getting things done.”

22. Passion is nuclear power: The emergence of the unknown outsiders

• 王兴兴’s knowledge did not come from a suddenly hot sector or an open-source project. He had built bipedal robots as a child. 张鹏 believes that “passion is everything; passion is nuclear power—persistent and powerful.”
• 张鹏 discussed early robotics paths, hydraulics versus electric drive, joint actuation and changes in force-control technology with 王兴兴 and 朱秋国 of Deep Robotics. He found the technical route exceptionally complex.
• 李翔 connected this with Honda founder Soichiro Honda. In the internal-combustion era, engines were considered nearly impossible, yet some people still hammered them together in their backyards.
• 张鹏 believes emergence occurs in multi-variable systems. 王兴兴 and 张一鸣 were not figures from established lineages whose futures could be identified by looking at the historical record. But such “people with no name and no school in the jianghu” are not vanishingly rare; many who truly make it belong to this category.

23. O2O and the end of mobile internet: The medium-resistance thesis

• To judge when mobile internet ended, the key is the medium. The first O in O2O was completed in the digital world; once businesses entered the physical world, resistance appeared. By around 2018, digital connectivity had been maxed out and no new source of leverage was generating incremental growth.
• What remained required entering the physical world, working through details, doing ground operations and converting activity into commercial results—outside Geek Park’s area of familiarity.
• O2O demanded founders who pursued truth and could make the economics balance mathematically. 翟光龙, founder of Songguo Mobility, initially worked on shared bikes but concluded the city-level economics did not work. He exited, moved to electric bikes in county markets and later built a strong business.
• Different environments test different capabilities. In the early robotics phase, 王兴兴 was not tested on storytelling but on seeing the problem clearly and walking through the darkness one layer at a time until dawn.
• 李翔 noted that Musk planted the flag for robotics at AI Day in 2021, and more people became convinced of the direction after 2022. 张鹏 believes Musk saw scaling laws in Silicon Valley and connected AI, chips and robotics, accelerating robotics. He also said Chinese founders repeatedly benefited from Musk’s demonstration effect in intelligent EVs and commercial spaceflight, while brain-computer interfaces have already been included in the 15th Five-Year Plan.

24. Companies you cannot understand: When the era catches up, pivot ability is the essence

• 张鹏 admits he once could not understand Xiaohongshu. In 2017–2018, it was difficult to judge what the company would become. It kept going, and in the end “the era caught up with the entrepreneur.”
• The lesson was that sometimes the founder does not catch up with the era; the era later moves in the direction the founder had originally understood.
• 李翔 added that Xiaohongshu went through repeated changes, including PDF and overseas shopping. 张鹏 concluded that a technical path does not need to be 100% correct from the outset. The key is whether, after getting it wrong, you still have a chance to choose correctly.
• That opportunity may come from sufficient funding, a sufficiently flexible team or simply not exhausting all resources on the wrong path. Over a long enough horizon, no company can avoid pivoting and adjusting entirely.
• 张小龙 is an exception in 张鹏’s view. “Long Ge is a god-level figure” and cannot be analyzed within the normal range, though he may still have completed multiple deeply considered adjustments internally.

25. Why/how/what: The limits of benchmark thinking

• 张鹏 observes that Silicon Valley founders typically introduce themselves or pitch a business plan by explaining why first, then how and finally what. Chinese founders often start with what, find a benchmark, examine how and may never focus much on why.
• This may reflect China’s latecomer position and its habit of using benchmarks and the “time-machine theory.” Xiaohongshu was repeatedly compared with Pinterest or Instagram.
• 李翔 believes why is becoming more important because it will be difficult to follow established paradigms indefinitely. 张鹏 also believes that from 2000 onward, why has gradually become a key source of power and vitality for founders; when why is unclear, risk rises.

26. The cost of AI 1.0: Carrying commercial workloads in kindergarten

• 张鹏’s biggest regret concerns China’s AI 1.0 and speech-recognition founders around 2016. They were China’s first true technology innovators and the first group to pay the price for it.
• AI had already begun to emerge after AlphaGo, and OpenAI had been established. The path built on Transformer and scaling laws, capable of continuous expansion, was already being explored, but AI was still in kindergarten or first grade.
• OpenAI had raised more than $1B without an immediately deliverable product. As a Frontier Lab, exploration itself was valuable. 张鹏 compared it with the moon landing: it might not produce commercial value immediately, but could first generate the emotional value of “we are incredibly powerful.”
• China’s value system at the time demanded immediate monetization and recurring revenue. Researchers were forced to deliver projects in areas where they were neither strongest nor necessarily most capable of shining.
• He recalled one AI founder who was a sunny young talent when they first met, but became haggard and hoarse several years later under the pressure of project delivery. 张鹏 believes demanding commercial output from AI while it was still in kindergarten diluted its growth potential.
• By contrast, today’s MiniMax, Kimi, DeepSeek and 智谱, along with 杨植麟, 闫俊杰 and 梁文锋, are doing this work in a better era. That advantage also contains the cost paid by the previous generation of Chinese technology founders.

27. The logic of staying in the game and the question: “What asset did you burn?”

• 李翔 noted that when Amazon and Meituan burned cash, progress could still be measured through revenue, users, transaction volume and engagement. For model and embodied-intelligence companies, it is much less clear how to measure the model capability and technical capability being burned into existence.
• 张鹏 believes model companies must first solve the problem of “staying in the game.” That remains a matter of life and death. If models can mass-produce intelligence and become an important form of productivity, the game has meaning. But there cannot be 100 model companies globally; he estimates there may be 10, perhaps even 20.
• The standards include the speed at which the capability ceiling rises, relative position and the size of the capability gap. If a competitor has broken through the atmosphere and entered the AI-improving-AI phase while you have not, the distance may keep widening until your existence no longer matters.
• Embodied intelligence is even harder because the data paradigm has not converged. There is not yet a kind of data that can directly lead to a brain capable of scaling. The ultimate bet is not just on mass-producing intelligence, but on mass-producing real productivity.
• 张鹏 emphasizes that money raised is essentially money borrowed against the future. It must eventually be repaid through value creation. Bubbles and failures will exist, but the value cycle still has to close.

28. From confusion to lightning: The vote in February 2023

• Geek Park’s shift from mobile internet to AI was not smooth. After 2019, especially during the pandemic, 张鹏 saw many phase-specific opportunities but could not find the next certain wave. He also studied Web3 seriously but was not persuaded.
• When ChatGPT appeared in 2022, he described the moment as “a bolt of lightning illuminating the night sky.” Because he had been searching, he immediately recognized it as the direction he was looking for.
• In February 2023, Geek Park’s Frontier Society held a closed-door discussion with 李志飞, 王小川 and others on the changes brought by ChatGPT. Around 20 people attended; 15 provided feedback, and 11 said it was an industrial-revolution-level event.
• 张鹏 therefore confirmed AI as the clear direction: “Even if we did not want to switch, we had to.” The subsequent speed and cadence of change, however, were almost entirely beyond his original expectations.

29. Web3 and the metaverse: Plenty of logic, few people who can execute it

• 张鹏’s view of Web3 is that the world contains many self-consistent logics, but few people can truly execute on them. He found many such people in AI, but not enough in Web3.
• Blockchain is a technology; Web3 is closer to a sector or industry. A philosophy must become a product, service and source of value that solves a real problem. If it creates a self-consistent value loop only within one circle but cannot reach the larger world, it becomes a problem.
• A metaverse may emerge, but the key question is how far away it is. Real headsets and a complete technology chain require 5–10 years and sustained investment of significant capital; even Meta alone would not be enough.
• The direction is not wrong; the timing has not arrived. AI may in fact help the metaverse arrive in a better form.

30. Starting early is more dangerous than starting late: 徐新’s framework and 余凯’s positive example

• 张鹏 believes that the early-stage failure mode is “not necessarily starting too late, but starting too early.” Yet startups must begin early. The key is whether they can survive until dawn through sufficient resources, flexibility or talent.
• 李翔 cited 徐新的 view: some opportunities are always available, such as great consumer brands; others, such as mobile internet and AI, are missed once missed. When uncertain, founders can invest on a small scale, learn along the way and increase their commitment after gaining confidence.
• 张鹏 warns that public markets currently tend to assign a benchmark first and then reverse-engineer the valuation of an early-stage company: “If you can achieve one-third of that, your future valuation should roughly be this.” That is dangerous for uphill projects, which require cumulative certainty rather than a single charge.
• He cited Horizon Robotics’ 余凯, who raised a large amount of money around 2020 while anticipating the winter ahead, then steadily converted the different visions carried by investors into his own method of use. “If you can master other people’s misunderstandings and turn them into the correct course of action, the result is correct and everything is correct.”

31. The way out for old-timers is evolution, not switching

• 李翔 said that many major figures from the previous mobile-internet era now seemed to be called “old deng.” 张鹏’s own reminder to himself is “do not become a deng.”
• The business world should not expect one person to dominate alone for 10, 20 or even more years. Being a leader in the previous era and a follower in this one does not mean having no value; there is no need to panic.
• For companies with existing businesses, the core task may not be switching but evolving: bring intelligent productivity into the existing business and allow “old trees to bloom again.”
• Pure-software companies may face a different shock. 张斯马克’s 景鲲 once showed at a Singapore conference how 1 person could spend 1 week and $15,000 in tokens rewriting an AI-native office. A product that once required hundreds of thousands of lines of code may now cost far less to build, washing away assets based on software and accumulated code engineering.
• Tencent represents another form of evolution. Products such as Manus and Trae previously put capability modules in the hands of high-autonomy Pro C users. Tencent packaged those capabilities into products ordinary users could also use, then added a smooth experience built around check-ins, points, tasks and milk tea. 张鹏 sees this as Tencent’s product gene.
• When 马化腾 was initially recommending “lobster” products everywhere, 张鹏 did not understand it. He later concluded that 马化腾 may have sensed that ordinary users also wanted to actually use AI.
• What the eventual switch will look like remains unclear. At minimum, one must keep up and move one’s own legs rather than be dragged along by the era.

32. Mid-career veterans need not panic: AI-native is an adjective; entrepreneur is a noun

• China spent the past few years enthusiastically investing in “genius” founders born after 2000, on the view that they were AI-native, free of historical inertia and facing a lower cost of failure. 张鹏 agrees that the logic holds, but Anthropic’s average age is not as low as imagined; it contains many “mid-career veterans.”
• “AI-native can be learned, evolved, downloaded and replicated, but entrepreneur—the noun—is difficult to replicate.” The capabilities forged through long experience and repeated failure cannot simply be skipped.
• The cost of starting something today is extremely low, but the cost of doing it well has not fallen. Entering a sector is easy; winning still requires market development, operations and fundraising, as well as strategy under constant change.
• 李翔 suggested that AI-native may simply be a label for the traits of leading organizations in different eras. 张鹏 agreed that each era produces words people aspire to, but the core transformation may not be earth-shattering; technological conditions merely allow organizations to break through a new ceiling.
• He added that a defining trait of an AI-native founder is thinking toward the ceiling. If AI is used only to cut costs and raise efficiency rather than challenge the upper bound, the founder has missed the highest-odds opportunity.

33. AI is also restructuring VC: “Mitosis” every few years

• 张鹏 views investment as the result of inference. Investors build models from historical datasets and use them to judge the next 5 or 10 years. The model of a successful consumer-internet investor may not fit AI, embodied intelligence or hard tech.
• China’s investment circle undergoes “mitosis” every few years: strong investors leave the previous generation of institutions, form new funds and bring their old datasets into a new era.
• He believes this exposes a structural question: do the production relations of existing investment institutions allow the people creating the most value to share enough of it? The industry may need to keep thinking about that.
• Another feature of AI-native entrepreneurship is that the cost of challenging the ceiling has fallen. In the past, challenging the ceiling might have required hiring 1,000 engineers; now it may require only more tokens. Founders who focus solely on cost reduction and efficiency gains are missing AI’s highest-odds opportunity.

34. No fear of missing out, only a review of incorrect judgment

• When a company’s valuation rapidly moves beyond the fund’s range, 张鹏’s attitude is to help where he can, applaud the company and become a partner and friend.
• The point to review is not “we failed to invest in such a good company,” but whether he misunderstood the person, misjudged the founder or misread the trend. “That is your basic livelihood.”
• A bad judgment may result from applying an old model to a new situation, or from not speaking deeply enough with the founder. Insufficient depth can produce excessive optimism or excessive pessimism.
• The community provides a continuous, relaxed observation environment. If an investment decision is made after a single meeting, the relationship may end after a pass. Content, community and other infrastructure allow Geek Park to keep learning about the founder.
• Geek Park insists on not building a very large fund. Once the fund grows too large, its focus and core capabilities may drift away from its axis. 张鹏 would rather ask why he did not encounter a project at the earliest and most correct moment than expand his investment range into batch investing after Series A.

35. Bubbles must exist; do this because founders are worth supporting

• 张鹏 believes bubbles “will always exist, and they must exist,” otherwise the world will not move forward. Innovation is an emergent act, and emergence requires a base. Without enough wasted energy, nothing new will emerge.
• How an individual responds to a bubble is a separate question. Super-consensus in the public markets requires caution, but under any broad consensus there are still many non-consensus opportunities. A destination that is certain 10 years out may still have no consensus about which point will sprout next year.
• The most important thing is not to be dragged along but to walk on your own legs. Even if the judgment is wrong, you should be able to acknowledge that it was your own judgment.
• 张鹏 ultimately returned to Geek Park’s motivation: Chinese founders still have a difficult path. China and the United States do not have equal startup environments, but China is improving and gradually developing its own character.
• He envies Silicon Valley’s ability to organize the wealth and experience accumulated by successive generations of founders into nourishment for the next generation. China is beginning to develop similar forces: 尹方明, one of 王兴兴’s earliest investors, was himself a founder.
• 张鹏 says that even when ordering takeout at home, one should thank 王兴. Creating content may not produce the same sense of achievement, but if Geek Park can move the environment slightly forward, that is meaningful.
• “There is a high probability that your contribution is part of their success; if it is not, there is a high probability you did not do your part well enough.” Geek Park wants to support more excellent people without defining itself by a required gross margin or scale. If more communities and organizations like this emerge in China’s startup ecosystem, the environment will gradually improve.