Summary

Anker is moving from a “Tier 5” value-for-money company to a “Tier 7” extreme-innovation company. After 11 years of smooth sailing—cash flow stayed positive throughout, its first financing in 2017 was solely to price employee equity, and it went public in 2020—Anker suffered a systemic failure in 2022: nearly 20 of 27 product lines “couldn’t beat the unicorns outside.” The answer was not process but a rewrite of the mission, vision, and values: “No matter how good the battle formation, without mission, vision, and values, you still cannot win the war in the end.”
The most technically ambitious move is an in-house compute-in-memory chip. The project started in August 2023 and shipped in an earbud in May 2026; the logic chain is complete: once end-to-end models replaced divide-and-conquer, the von Neumann “memory-compute separation” architecture required all parameters to be repeatedly moved for hundreds of on-device inferences per second, sending power consumption through the roof. “Could we stop moving them out?” The chip has 4M parameters in total, with a maximum of about 2M in a single model; previous earbud models had fewer than 2M parameters. It separates voices from noise in noisy environments and enables “quiet-voice calling,” and is for internal use only—not for sale (a closed-loop model akin to Apple’s A/M and Huawei’s Kirin).
His core thesis on AI hardware is a hierarchy of on-device models. Trillion-parameter “superbrains” belong in the cloud; models with 1B to 10B parameters run on-device; models with several million to tens of millions of parameters sit in the “neurons” handling perception and control, analogous to the brain, cerebellum, and optic nerve. The long-term vision is to deploy a trillion-parameter superbrain in the home at a cost of a few thousand dollars. But he also throws cold water on the narrative: “Customers don’t care about AI.” AI-native hardware looks much like the IoT narrative of 10 years ago; after 10 years, “companies with real use cases win more.”
His category-selection framework maps directly onto valuation logic. Supercategories—smartphones at $500B, PCs at $200B, and TVs at $100B—are deep water with a $500M blind; shallow water comprises 100-200 categories below $50B. Anker stays in shallow water (No. 2 globally in security, with more than half of the premium segment for products priced above $500), does not make smart glasses (with 3 waves of phone makers, internet giants, and model companies blocking the route), and only conducts preliminary research on humanoid robots because “the tech stack does not converge at all”; if it works, it “might be spun off.”
The “third kind of company” is the end-state narrative. Type 1 companies do a handful of supercategories (Apple, Tesla); Type 2 do a handful of small and midsize categories; Type 3 systematically excels across dozens of small and midsize categories (P&G, Nike, Texas Instruments). The management model must shift from “king and knights” to “president and federation.” The allocation rule is explicit: “creators take 70%, shareholders take 30%”; average employee income rose 20% in each of 2 consecutive years, the number earning over RMB1M a year rose from roughly 500 in 2024 to 800 in 2025, with a goal of more than 1,000 this year.
The data behind AI-driven organizational change are unusually concrete. Since September last year, internal token consumption excluding consumer-facing services has exceeded 10T, current daily usage is about 150B, and this year’s cost is a low few hundred million yuan; there is no ROI accounting because “we see because we believe.” The path is a 3-part program: people embrace AI; capabilities are codified into pre-orchestrated agents and connected across domains; and the organization changes. He pushes back on Jack Dorsey: given today’s model context—explaining a company requires 10B to 1T tokens—middle managers will not disappear and will instead become more powerful.
His three-layer moat theory is worth preserving. Technology advantages are eventually flattened by human perceptual limits (Retina displays, card-sized 146W chargers); brand trust is the second layer ($10,000 residential storage with no physical store and more than 1,000 units sold in the first month; a recall of several million power banks last year with losses in the hundreds of millions, decided without endless debate); but the deepest layer is “mission, vision, values, and a group of people who truly believe.” His self-rating: technology 0.2-0.3, brand 0.6-0.7.

Deep dive

1. A good student’s consistent choice: pursue only visible, real value

阳萌 sees himself as a “good student who never stepped out of line”: Yali High School, computer science at Peking University, and a PhD at UT Austin under Reuben Mooney, an ML pioneer and multiple-time ICML chair. Dropping out was not the product of careful deliberation but “a calling from within”: during his doctorate, he was picking apart 1 or 2 differences in a small dataset of more than 20,000 Reuters articles—“you couldn’t even feel the practical significance.” In 2005 he interned at Google: “I suddenly jumped to building algorithms across 41 web pages … changing even 0.1% of search is a big number,” and stayed. On the academic side, the push was that “the SVM mine had been pretty much dug out wherever you could see.”
Google’s legacy for him was culture rather than code: egalitarianism and transparency (he still has no office, and all meeting rooms have glass walls), plus the worldview that comes from having submitted a change list to Jeff Dean—“that legend was there; you felt the whole world had been created by him, and we were making improvements inside the world he created.” He admits he lacks a strong urge to “create a world” himself; he is a tech-lead-style engineer drawn to abstract architectural thinking.
The startup’s origin was intensely practical: his wife struggled to find work in the US, so they started selling goods on Amazon. He quickly realized that “selling is not the hard part; finding the best products is.” In 2011 he brought in Google China’s head of ad sales, 赵东平, to found Anker, originally called “Haiyi E-commerce”; he kept the e-commerce team in Changsha and trained campus hires, then moved to Shenzhen himself in early 2012. His mother, a sales head at a state-owned enterprise, provided RMB3M as angel capital: “For something reliable, we should use our own money.” He later returned the money and took no equity.
The first in-house product was an ultra-thin power bank that attached to the back of a phone and included a built-in cable—“it took 10 years before you saw the whole market move in that direction”—but weak marketing and “even good wine fears a deep alley” kept it from breaking out. The first breakout hit was the 2013 “lipstick power bank,” which sold several million units over its life. 2012 revenue was “certainly over RMB100M, as I remember it,” and cash flow was positive from day one.

2. The 1-3-5-7 ladder and shallow vs. deep water: the first decade was a Tier 5 shallow-water company

His customer ladder: Tier 1 buys the cheapest, Tier 3 buys value for money, Tier 5 pays a premium for quality, and Tier 7 wants the most expensive and best. In bubble tea, the ladder runs from Mixue at RMB6, Yidiandian at RMB10-12, Shanghai Auntie at RMB15-16, to Heytea at RMB20-plus. The car version is more cutting: with the same joint-venture engine, “put it in a BYD and sell it for RMB100,000; put it in a Volkswagen and sell it for RMB160,000-170,000—surely both have buyers?” The market’s “overseas Xiaomi” (Tier 3) label is a misread: Anker prices 20-30% above the market average, backed by quality of 4.5 stars or higher, and “has never manipulated rankings or faked orders.” Not faking orders while still requiring 4.5 stars forces product managers to push quality exceptionally hard.
The shallow/deep-water distinction is often misunderstood: “shallow does not mean simple; it means small in scale.” Consumer electronics has only 4 supercategories: smartphones at $500B, PCs at $200B, TVs at $100B, and tablets, the No. 4 category, now down to $60B. There are 100-200 shallow-water categories below $50B: power banks at $5B, charging cables at $5B, earbuds at $20B-$30B, and consumer drones below $5B. The Texas Hold’em analogy: in shallow water the blind is a few million dollars, with several hundred million dollars at stake; in deep water, “one blind is $500M—you bring $5B to the table, and after 3-5 hands it’s gone.”
The candid postmortem: category expansion in the first 10 years was driven mainly by intuition—“looking back, this is also where I think I failed as CEO.” It was only after he conducted a systematic strategy review for the first time in 2020 that he realized Anker was a sprawling company with “more than 10 categories, mostly Tier 5, one Tier 7 (North American security), and several Tier 3” businesses. When Epson’s CEO visited this February and praised him—“Yang-san, you’re amazing; you do so many categories”—his real reaction was: “I’m definitely very ashamed.” The ideal path, he says, would have been to derive 1, 2, or 3 categories from a single underlying technology.
One archival anecdote: at Hupan in 2016, when asked whether the category could become a $100B category, he answered, “Power banks will probably die in a few years.” Consumer electronics is born and dies fast: the time from buying to no longer buying MP3 players, tape players, and CD players was only 10 years. That fear briefly focused him on e-cigarettes, a “category that would not disappear” (“nicotine is probably a demand that will not disappear for humanity”). He ultimately concluded that heavily regulated e-cigarettes were not especially well suited to a startup. In hindsight, he admits he was “a bit too worried” about power banks dying, but “we believed this category would definitely disappear.”

3. The systemic failure of 2022: formations cannot save an army without conviction

After introducing a product-line reorganization and layered delegation in 2020, Anker hit a systemic failure in 2022: “The product lines we built could not compete with the unicorns in their industries in real commercial competition.” For a time, the whole company was asking, “Why should one product line be able to beat an outside unicorn?”—creating “a stifling, even despairing mood.” For 11 years, everything had gone his way: “We didn’t lack money, the products were going well, and we went public.” The world had been particularly fair to him; recent years turned painful.
The diagnosis went beyond organizational mechanics: the issue was not the number of product lines. At the lower layer, it was “whether there is genuinely leading core technology”; at the upper layer, “whether your organization has a particularly clear mission and whether these people’s behavior can achieve it.” “No matter how good your processes and methods are, you still cannot win the war in the end.”
The solution was to break “how to create extreme innovation” into a 3-layer behavioral framework: first principles (how to think), pursuit of excellence (how to act), and growing together (how to conduct oneself). Growing together further breaks into long-termism, self-awareness (“first see your own part in the problem”), and self-evolution (“solve problems by changing yourself”). The framework’s provenance is very Anker: document the behavior of Jobs, Musk, and successful creators inside the company, then abstract and distill it—“you find that it boils down to these 3.”
He repeatedly emphasizes the mental difficulty of moving from Tier 5 to Tier 7. The first 2 rungs—surpassing yesterday’s self and surpassing the best peer—are “walking on a road of certainty”; the third, “surpassing consumers’ most fundamental needs,” requires leaving the existing track: “There may be no one on this road; even people who took it may have died on it.” The charging category is relatively easy: the Prime line’s average price is already above $100 and it has firmly reached Tier 7. In earbuds, where AirPods already dominates, “truly reaching Tier 7 tests your first principles and your pursuit of excellence.”

4. The compute-in-memory chip: an architectural revolution derived from a call-quality pain point

The starting point was product insight, not tech showmanship: in 2021 he built an acoustic-algorithms team that used small models and divide-and-conquer for beamforming, noise reduction, echo cancellation, and voice enhancement. After seeing ChatGPT in 2023, he concluded that “the first 80 years of computing were an era of solving problems with divide and conquer … the future is an end-to-end era of learning from data plus reinforcement learning.” But end-to-end models hit a wall in earbuds: the best chips at the time saw power consumption explode even running models with a few hundred K parameters. The underlying issue was that with hundreds of time slices per second, every inference had to move all parameters from memory into the NPU; “the movement itself consumed the most power.”
He went back to first principles: “Can I stop moving them out?” His architectural case was fully worked out: von Neumann (memory-compute separation), divide-and-conquer, and the program/data model formed a stable triangle for 80 years; the end-to-end era should break it. His left/right-brain analogy: “In the past, moving only a small part of knowledge into the right brain each time you solved a problem made sense; today, if you have to move the entire left brain into the right brain to compute, is that starting to seem less reasonable?” The human brain is compute-in-memory—neurons both store and compute; “not moving any parameters during computation is inherently more efficient.”
The market search turned up no usable compute-in-memory chip (“there were companies doing compute-in-memory, but the products they had defined at the time were completely different from what we needed”), so Anker co-developed a custom chip with the field’s strongest company. The project started in August 2023 and a product carrying it launched in May 2026. The chip has 4M parameters in total, with a single-model ceiling of about 2M; previous earbud models had fewer than 2M parameters—“that is already huge for a model that can run in an earbud.” The result: accurate voice/noise separation in any noisy environment, plus “quiet-voice calling”—speak very softly and the other party can still hear you clearly. When the host asked whether building such a large hammer for such a small optimization was worth it, he answered: “It is the first hammer in the hammer series.”
The commercial model is clear: the chip is for internal use, not sale. It is fully customized to Anker’s own models—“giving the chip to someone else means giving them the model too.” The reference points are Apple’s A/M series and Huawei’s Kirin: “putting it in our own Tier 7 products to let them command a higher price creates far more commercial value than selling the chip.” He is measured about compute-in-memory’s limits: training is still out of reach; on inference, models with tens of billions and hundreds of billions of parameters are “visible within the next few years,” while the trillion scale is “still hard to reach.”

5. On-device model hierarchy and “real intelligence”: AI-native is not first-principles thinking

His model-stack framework is biological: trillion-parameter superbrains in the cloud solve the hardest problems; 1B to 10B models “will most likely run on-device”; models with several million to tens of millions of parameters run in the organs of perception and control—“Does your eye send every parameter it sees directly to the brain for processing? No. The optic nerve processes the light signal and sends only a small amount of signal to the brain.” He acknowledges another camp, in which raw sensor data goes directly into a large model, but asks: “Is that really the best and most correct approach over the long term?” The most aggressive vision is that, at the limit of compute-in-memory, “there is a chance to deploy a trillion-scale superbrain in the home for a few thousand dollars.”
The smart-home demystification was the episode’s best product passage: the more than 10 buttons on a toilet are “just features, or what you might call adjustability.” Products have 3 stages: non-adjustable; adjustable, as with an ergonomic chair that “can be adjusted in more than 10 places; first, you don’t really know how, and second, you forget to adjust it”; and true intelligence—self-adjustment, with the chair sensing that you are gaming and reclining the backrest, or sensing that you are in a meeting and moving the seat forward. Why is the home still not smart in 2026? “There are simply no models in these products.” True intelligence needs perception, planning, and control—the 3 capabilities that, after maturing in autonomous driving, “are now ready to spill over.”
His cooling of the AI-native hardware narrative is worth retaining for investors: it is a direct replay of the IoT narrative of 10 years ago. “After 10 years, when companies with solid use cases and the ability to evolve compete with companies that have AI technology and insist on entering the category, the companies with use cases still win more.” The conclusion is “customers don’t care about AI; customers care about how to make the experience better”—just as consumers never cared about your IoT. Take the microwave: it has more than 10 buttons but you use only 1 or 2; the minute dial is the control you use most. It should sense what you put inside, make suggestions, and interact with you by voice.

6. Security and the watchdog: 3 stages of embodied intelligence and 2 deep seas

A little-known piece of the core business: Anker’s security system is No. 2 globally, with more than half of the premium segment for products priced above $500; its differentiation is “no cloud”—a base station at home, long-range Wi-Fi backhaul, local storage, and a model of under 2B parameters running locally. “You can ask it directly: Has my daughter come home? Where are our cat and dog?” By contrast, Amazon-acquired Ring relies on the cloud and is priced at Tier 3. The AI engineering team has grown from a handful of people when security began in 2017 to nearly 300 today.
Robots come in 3 forms: 2D/2.5D planar robots (robot vacuums, lawn mowers, pool robots—already an existing business); 3D mobile interaction (robot dogs, in product development since last year); and manipulation robots/humanoids, the “crown jewel,” but “the tech stack does not converge at all … every company’s solution for what the waist joint should look like is completely different,” so Anker is doing only preliminary research. The robot dog’s user-value loop is unusually concrete: dogs first entered the home for the functional value of guarding it. “When the security system detects an intruder, this dog can drive them away—not just detect an intruder, but make your home genuinely safe.” User research shows that “a substantial share” of Tier 7 users are willing to spend “a substantial sum” to buy it; “when we started, we were already sure it could sell in very large volumes.”
Two deep-sea categories are certain to emerge in the future: humanoid robots and smart glasses. Smart glasses are explicitly off the table—3 waves of giants (phone makers, internet giants, and model companies) are all fighting for the interaction gateway, and “any company entering this category has to answer how it can beat these 3 types of giants.” For humanoids, Anker will wait for the technology to converge: “most likely it will not be done by a part of the company’s existing organization; it may require a more independent team,” with a possible spin-off acknowledged—like Huawei’s handset business growing out of its carrier business.

7. The third kind of company: president and federation, 200 roles in the value chain, creators take 70%

The landscape of the 3 company types: Type 1 does a very small number of supercategories (Apple, Tesla); Type 2 does a small number of small and midsize categories (most companies); Type 3 systematically does dozens of small and midsize categories well. Historical examples include Sony, Philips, P&G, Nike, and Texas Instruments: “every industry ultimately produces several successful Type 3 companies.” Management divides into “king and knights” versus “president and federation”; every fully formed Type 3 company has moved to the latter. His role is to “decide the shape of the company—strategy, capabilities, organization, talent, reward framework, and which broad categories to pursue,” while product details are delegated to product lines, leaving “many opportunities to take full ownership” across the company’s 3-tier structure. Current layout: 3 major areas—charging and energy storage, audio/video, and home automation—plus a health incubation effort, covering roughly 20-odd categories, with a long-term target of “40 to 60.”
His platform pitch to hardware entrepreneurs rests on value-chain economics: Anker has 6,000 people and 200 different roles linked into a value chain. “As the value chain grows, organizational complexity rises quadratically or even faster.” That is why the smart-hardware companies that were funded and “have survived and are doing well today can certainly be counted.” His offer is a mature value chain, people who execute exceptionally well, and agents distilled from accumulated know-how—giving a company “a higher probability of surviving and thriving in the category,” while “giving most of the value created to the creators themselves.” A good company, by his definition, “continually—at least 2, 3, 4, or 5 times—gives employees opportunities with room to act, grow, earn, and find meaning.”
The first-principles derivation of the split starts with “total residual value”: the portion of revenue left after mandatory costs that can be divided between employees and shareholders. First-rate companies usually have more than 30%; roughly 10 percentage points for shareholders is “the basic threshold of a good company.” Take 30 points and subtract 10, and “it comes out to roughly 70/30.” The data bear this out: after the move to Tier 7, gross margin rose by 1.1-1.5 percentage points a year while the shareholder allocation barely changed; “all the extra money went to employees.” For employees who were on staff in 2023, total income rose 20% on average in 2024 and another 20% in 2025. The number earning more than RMB1M a year was about 500 of 5,000 people in 2024 and 800 of 6,000 in 2025; the target this year is over 1,000, with the ideal being “over RMB1M per capita.”
In the AI era he intends to keep the ratio unchanged: “I keep telling everyone that the extra profit created by AI productivity gains will not automatically stay with me … In the short term, it may seem that the extra value you create can be allocated to yourself, but to keep the value cycle running over the long term, you need to allocate value to the people who create it.” The macro observation behind it is that in every productivity revolution, “the additional value created is often first allocated to the people who allocate value—the person cutting the watermelon takes the first slice.” This is also how he understands Sam Altman’s universal basic income (UBI) thesis: AI creates value for companies → governments collect taxes → distribute it to everyone.

8. AI-driven organizational change: 10T tokens, pre-orchestrated agents, middle managers survive

Start with the data: since September last year, internal usage excluding consumer-facing services has exceeded 10T tokens; current daily use is about 150B, which at a 6,000-person company means 25M tokens per person per day. Roughly 1/3 use the top-tier model, more than 40% the mid-tier, and 20% entry-level models; this year’s cost is “a low few hundred million yuan.” The management philosophy is “you see because you believe”: “Up to this point, we have not rigorously measured what the tokens we spent specifically produced … We simply believe the organization of the future will be an organization strengthened by AI, so it is best to start early.” The ceiling? “If we can make a lot of money from customers, there should be no ceiling.”
The dividing line between enterprise AI and personal AI: Open Call, as spoken, represents “instant orchestration” for task handling—assemble the workflow on the fly, but it is ad hoc, unstable, and unaware of internal enterprise data. Enterprise applications should be “pre-orchestrated agents”: integrate proven methods, processes, and existing datasets to “deterministically produce the value we expect,” then host them on an AI orchestration layer and route them to different models. The trigger for change was Sakit’s workflow agent: “Things the model could not complete effectively ran very smoothly once we added this layer of agent description and constraints.” That led to the 3-part program: people embrace AI; capabilities are codified into agents and connected across domains (otherwise “you speak German and I speak English”); and organizational change, targeted for completion by year-end.
He is precise about what changes and what does not in organizational form: customer value does not change, the value chain does not change, and the individual Scrums along the chain will not disappear. But a software Scrum may shrink from 8-10 roles to a few people as roles merge; a hardware Scrum “deals with the physical world, so its size may not shrink.” His rebuttal to Jack Dorsey’s vision of world models running companies and eliminating middle managers is context math: explain a project, a Scrum, and a department, and the context compounds layer by layer; “the context needed to explain a company must exceed 1B, 10B, and eventually 100B to 1T tokens.” Models today may lose information once context exceeds 1M; “what to keep and what to discard in compression is an extremely important value choice, and models clearly still do it poorly.” Add Conway’s law: organizational form determines output, and organizations are “cultivated through the painstaking work” of layer upon layer of managers. Conclusion: middle managers will not disappear; “they may become more powerful with AI.”
His counterexample to “entry-level jobs disappear = the pipeline to senior jobs dries up” is architectural history: ancient architectural designers emerged from craftsmen who laid bricks and plastered walls; modern architectural design is now a standalone discipline. The company already has “campus hires with bachelor’s degrees who had never written a program, yet used AI to deliver a complete system in a clearly defined business domain.” For the talent structure he coined “n-time efficiency worker” (NEW), divided into “heavy hitters” (people who build agents and codify data workflows) and “efficient operators” (people who use agents effectively): “a small number of heavy hitters plus more efficient operators, with rewards allocated according to the n-fold value created.”

9. Three-layer moat: technology is flattened by perceptual limits; values run deepest

The first-layer technology moat has a shelf life; the evidence is the “limit of human perception.” Apple’s Retina definition is pixels packed densely enough that normal people cannot see the difference—do you still think an iPhone’s screen is better than others’ today? Chargers have reached card size (146W): “make it half as small again and you may still see a point; smaller than that has no meaning.” “An excess technology advantage will inevitably be flattened over a sufficiently long period—this is basically a first principle.”
The second layer is brand trust, with 2 concrete data points. In February this year Anker launched residential energy storage with a $10,000 average order value; there was no physical store to see it in, customers ordered directly from the website, and it acquired more than 1,000 customers in its first month. Last year it recalled several million power banks globally, with economic losses in the hundreds of millions, but the decision “went through no repeated workshops or extensive debate—once you start from user value and know what to do, you do it.” Trust is asymmetric: “Consumers need many repetitions to build trust in you; make 1 or 2 mistakes and that trust disappears forever, and may never return in their lifetime.”
The third layer is the ultimate answer: “Over the longest time horizon, neither technology nor brand is the deepest moat. The deepest is mission, vision, values, and a group of people who truly believe.” With Nokia, Sony, and Philips, “you still trust them, but the probability that you choose them is no longer that high.” His self-rating is unsparing: technology 0.2-0.3 (with the compute-in-memory chip, “for a period you don’t see anyone else with anything similar,” but it remains far from ideal); brand 0.6-0.7 (the longest of the 3 planks); and people density, which “has gone from quantitative change to qualitative change, accelerating over the past 1-2 years.” His reason for appearing on the show is equally direct: to attract creators. “In another 2 years, I hope I won’t need to come on the show—the products will speak for themselves, and the creators can come out and tell their own stories. The chance to stand on stage and tell them proudly should not be mine.”
His attitude toward competition draws on Finite and Infinite Games. In 2004, a Microsoft executive was explaining where his MP3 player was better than Apple’s, while an Apple executive was talking about what users were still dissatisfied with. “When you fixate on competition, you always end up with a faster horse, not a real car.” A passing “the bear is coming” joke: “You don’t need to run the fastest; you need to run a little faster than your peers.”

10. Game difficulty, a 3/5 risk appetite, and raising a child: an INTP’s way of being

He and president 东平 are “the North and South Poles of the earth” (INTP vs. ESFJ). After 15 years, their personalities have not converged but become more polar: “If you have someone who can safely watch your back, you become bolder about moving toward the side you are good at.” The company saying is: “Without me, this company might not be as complex and fun as it is today; without 东平, it would probably already be dead.”
Game difficulty is the through-line of his self-positioning: Anker went from easy to medium and then deliberately chose hard, with some categories at nightmare. “DJI is a company that chose hell mode from day one”—hard first, easy later; Anker did easy first, hard later, “so it has been making up the coursework in recent years.” He sees Insta360 as more like DJI. On a 5-point risk scale (5 = add leverage, 4 = go all in, 3 = invest 60-70%, 1 = do not invest), he consistently chooses 3: “It looks middle-of-the-road, but over a sufficiently long series of games, as understanding, capabilities, and creator density rise … it leads to a successful Type 3 company.”
On going public, there are 2 perspectives: employees “almost always want an IPO”; founders need to see the cost—many 80-year-old chairmen of listed companies in Taiwan “enjoyed the pleasure of ringing the bell once at 40, then still had to keep giving to the company when much older.” The benefit of staying private is “more exit routes”: it can be sold to a larger Type 3 company (P&G and Texas Instruments both grew through M&A). But “China’s current environment is not yet well suited to M&A; everyone thinks an IPO is the better choice.”
The ultimate metaphor is raising pigs, horses, or children: “东平 and I really are a bit like raising a child—we hope, if possible, that this company will live longer than we do.” How would Anker die? “A successful Type 3 company that keeps winning across multiple categories and keeps having great people will not die. It will ultimately die because everyone forgets the mission and vision and stops holding to the values.” The closing personal fragments: he deliberately limits his time at the company to no more than 8 hours a day (“I push a lot of things forward at the company all day long; it probably will not grow into a Type 3 company”); he cannot remember films he has seen—“watch one again 6 months later and it feels brand new”—but “my memory of abstracted structural things is steady and firm”; he recommends Ram Charan’s The Leadership Pipeline, whose experience and insight in corporate leadership development span decades. His off-the-cuff riff on the studio name “Language and World” was unexpectedly technical: language is naturally aligned with “Tao” (TAO = Task-Action-Observation); “no data is like language in being able to observe simultaneously the two-way changes from O to A (VOA) and from A to O (the world model).” He sees this as the foundation of future embodied robots. It closes with Interstellar: “You must always continue to believe in human nature—the good side of human nature will ultimately get us safely to the next era.”

Verification Notes

The claim that the author was responsible for leadership development at GE (the spoken reference was “wing,” likely GE) could not be verified against the RAW; the GE attribution has been removed.

Outbound Investment Rules, the Dollar Tide and the AI Capital Frenzy Why Secondary Markets Are Eating the IPO | All-In Liquidity Secondary Markets Panel