Thoughts Inspired by "Xiaomi's Entrepreneurial Thinking"

Before starting my business, my wife bought me “Xiaomi’s Entrepreneurial Thinking,” but I never read it. Recently, I had some time to go through it and found it very rewarding. I used to dislike such books, thinking these experiences were processed and beautified, and some advice might not be applicable. However, after having personal entrepreneurial experience, reading books by industry leaders makes a lot of sense.

The essence of “Xiaomi’s Entrepreneurial Thinking” is in Chapter Six, “The Seven-Word Formula for the Internet,” which is Focus, Extreme, Reputation, Speed.

The development approach of MIUI fully embodies the “Focus, Extreme, Reputation, Speed” seven-word formula for the internet:

• Focus: Initially, only four functions were developed (phone, SMS, contacts, and desktop), with extreme restraint.
• Extreme: With customizable lock screens and themes, it could simulate any phone, pursuing an extreme experience.
• Reputation: The entire company communicated with users on forums, making friends with them. It was very popular on the XDA forum and became a hit abroad, with its earliest internationalization starting from MIUI.
• Speed: Weekly iterations, adopting an internet development model.

Focus

Focus is the most important of the seven-word formula for the internet and applies to all companies and products.

Companies Need Focus

Lei Jun shared his first entrepreneurial failure experience. Lei Jun was technically strong, completing four years of credits by his sophomore year. In his junior year, he wrote the antivirus software “Immunity 90,” which sold for a million yuan—a significant amount in the 1990s. So, in his senior year, he founded the Tricolor Company with two tech experts, Li Ruxiong and Wang Quanguo (both of whom are very successful now), but this venture quickly ended in failure.

Lei Jun said that he and his partners had good technical skills and some connections in Wuhan, seemingly having some “customer resources.” But the biggest problem with Tricolor Company was that from day one, they didn’t know what they wanted to do. They seemed capable of doing anything but actually knew nothing about what to do. They tried assembling computers, software development, trading electronic components, and even printing, but quickly fell into trouble. Additionally, the equal distribution of shares among the four co-founders led to chaotic management. At the most difficult times, they could only rely on partners playing cards with the cafeteria chef to win meal tickets to survive.

Lei Jun also encountered many such FOMO (Fear of Missing Out) teams with a high failure rate when he was an angel investor. Lei Jun said that one of his statements was widely misunderstood, “Even a pig can fly if it stands at the right wind.” But his intention was to go with the flow, not to encourage a speculative mindset.

Firstly, before the wind comes, most pigs that can fly are well-prepared. Many successful founders, although young, have mostly experienced business training and have deep accumulation in related industries. However, many people misunderstood that students without business experience could also succeed in entrepreneurship, so Lei Jun does not encourage college students to start businesses. Lei Jun said that he honed many basic skills during his 16 years at Kingsoft and accumulated a lot of intangible things (such as observations and thoughts on entrepreneurship) during his years as an angel investor. Only when he was ready did he start Xiaomi.

Secondly, if a pig doesn’t have enough ability when it’s at the wind, it will quickly fall. For example, during the two waves of mobile internet in 2010 and 2014, many internet giants and startups made phones, but today, only Xiaomi is doing relatively well among these two waves of companies. Instead, many Xiaomi ecosystem companies that focused on niche products like power strips, air purifiers, and wristbands survived.

Therefore, a company must have a clear mission and vision. It may not need to be explicitly documented in the early stages, but it must have a simple dream that is not just following the crowd (“non-consensus”). For example, Xiaomi’s initial dream was to change the manufacturing industry with internet thinking and methods, promoting an efficiency revolution. In 2014, Xiaomi’s documented mission was “to let everyone enjoy the fun of technology,” and its vision was “to be the coolest company in the hearts of users.”

I believe that many AI entrepreneurs today also have a FOMO mentality. They think that those working on foundational models should benchmark against OpenAI, those working on applications should benchmark against TikTok or Xiaohongshu, and those working on smart hardware should benchmark against Apple. These big tracks that everyone can see, like mobile phones, will certainly be very competitive, and those who survive in the end will be those who are well-prepared and capable. If you think you don’t have the skills of these big names, don’t force yourself into these big tracks.

When I was doing my Ph.D., my advisor said that submitting papers to top conferences is not like buying a lottery ticket. It seems like there is only a ten percent acceptance rate, but if I submit without being prepared, it will almost certainly be rejected. Therefore, the acceptance rate of papers is not proportional to the number of submissions. Additionally, don’t casually venture into fields you don’t understand and where no one around you understands. People shouldn’t always follow trends; it’s more important to think about what you can do and how to combine your abilities with hot issues.

Products Need Focus

Lei Jun believes that the most important thing for a product is focus. Don’t try to solve too many problems with one product; meeting one urgent need to the maximum is a success. This does not contradict the company’s imagination space. Xiaomi has a saying, “Break through at a single point, then gradually expand.”

Lei Jun summarized four points:

1. A clear and urgently needed product is easier to find a clear user group. This way, the product is less likely to deviate after development.
2. The chosen user needs should have some universality, which determines the future market prospects of the product.
3. Fewer problems to solve, faster development speed, easier to control initial R&D costs and risks.
4. Products that solve clear problems are easy to explain to users, making promotion relatively simple.

The book gives the example of the Xiaomi Air Purifier. At that time, smog and air quality were pain points for the Chinese people, and high-quality air purifiers on the market cost thousands or even tens of thousands of yuan. Xiaomi quickly incubated an ecosystem company that only focused on air purification, without fancy functions like aromatherapy. A product priced at 899 yuan could achieve the effect of imported brands costing over 5000 yuan, focusing on high cost-performance, and quickly became the number one in the domestic air purifier market, maintaining this position to this day.

Another example is the robot vacuum cleaner. At that time, robot vacuum cleaners couldn’t do the mopping function well, often dragging dirt everywhere. So Xiaomi abandoned the mopping function and focused only on “sweeping fast and clean,” thoroughly addressing the most urgent user need. If they had included the mopping function and not done it well, users would complain, damaging Xiaomi’s reputation for products that can be “bought with eyes closed.”

My wife said I like to aim too high and lack focus. She said that the organizational social network I worked on during my undergraduate years had many good ideas, proposing the concept of Web3.0 in 2011, and even talked about using a program to regulate organizational decision-making and profit-sharing between content producers and consumers, which is essentially today’s smart contracts. We wrote tens of thousands of lines of PHP code, published a white paper titled “Embracing Web3.0—The Conception and Implementation of the Gewu Zhizhi Network,” and set up stalls, distributed flyers, and held outdoor movies on campus to attract user registrations. But in the end, this organizational social network didn’t have many users because it didn’t solve the users’ pain points. Students didn’t have a need to make money by writing content, and how much money could students have anyway? This little money being transferred around was just child’s play. In contrast, Ethereum’s smart contracts, despite their inefficiency, crucially solved the pain point of anonymous transactions.

Compared to the organizational social network, the USTC Course Review Community is a more successful product. The Course Review Community originated in 2015 from a simple idea by jenny42. She wanted to know more about courses when selecting them, so she teamed up with me and my roommate Chang Zhen to create a website for students to share course experiences. Almost all USTC students have used the Course Review Community, and even many students from other universities come here to download course materials. This is also my most well-known project among USTC students. Over the past nine years, many students from other universities have asked me if I could help them set up one. I said the Course Review Community code is open source, and they can set it up themselves. As far as I know, several schools’ students have either rewritten or developed their course review communities based on our open-source code. Some students who went to other schools for graduate studies said that without the Course Review Community, they wouldn’t know where to find information about each course.

The key to the success of the Course Review Community is focusing on solving a clear and urgently needed product. When the Course Review Community was first released, I wanted to add a forum for free discussion, but jenny42 disagreed, believing it was unrelated to course evaluation. My wife said that if jenny42 hadn’t stopped me back then and let me add all the features of the organizational social network to the Course Review Community, it wouldn’t have succeeded.

During my Ph.D., my advisor repeatedly reminded me to focus! I always had many new ideas, and my advisor said to finish the current work first. At this stage, work should be done one after another, not in parallel. However, in the last two years of my Ph.D., I was relatively free and pursued several ideas simultaneously, resulting in fewer publications than in the first three years.

In the early days of our startup, we wanted to create an AI operating system. The investors told us, Sam Altman can say he wants to create an AI operating system, but we can’t. We don’t have that many resources and can’t start by creating a big system. In “Xiaomi’s Entrepreneurial Thinking,” Lei Jun said that when their small team started working on MIUI, they knew the difficulty of operating systems, so they focused on four functions: phone, SMS, contacts, and desktop.

I found that many AI startups also have a lack of focus, always thinking about creating AI Native general applications and unwilling to delve into industries. I know several companies working on AI voice calls, most of which are thinking about mimicking Character AI to create a hit application or a general voice assistant with great imagination like Her. Their technology is strong, with lower latency and cost than OpenAI, but they can only get some enterprise orders for intelligent customer service, and their to-C dreams are still far away. I think the successful case of voice calls is Duolingo, which is a foreign language learning scenario. A mature foreign language learning application combined with AI solves the pain point of finding a foreign teacher to practice speaking. Duolingo uses the expensive and slow OpenAI, but this doesn’t hinder its commercial success.

People Also Need Focus

It is well known that Lei Jun has very strong personal technical abilities. He was promoted to CEO at Kingsoft in just six years, but even after becoming CEO, he still managed during the day and coded at night, and his favorite thing was still coding. Until one day, a new employee formatted his computer, and he couldn’t code anymore (I suspect this was the intention of Kingsoft’s management), he then devoted all his energy to management, truly becoming a professional CEO.

Extreme

When Unknown, One Can Focus on Practicing

Lei Jun found that entrepreneurs who have succeeded once often have a lower success rate in their second venture. The main reason is that they always want to leverage existing resources, neglecting the pursuit of product excellence. In the early stages of product development, the harm of leveraging existing resources may not even be apparent in the short term because, with influence and paid promotion, products can always be sold, and the team has no motivation to pursue excellence in product, experience, and innovation.

When Lei Jun started Xiaomi, he required all new colleagues to sign confidentiality agreements, not letting people know it was Lei Jun’s company, all for one principle: everything relies on the product to speak.

MIUI was released as an interest group, initially with the team posting on forums to recruit volunteers to flash their system, relying on word-of-mouth communication without commercial promotion. In the first week, only 100 people dared to flash MIUI, and Xiaomi even made their IDs the first version of the boot screen. A year later, with anonymity and without leveraging any external resources, MIUI’s user base exceeded 300,000.

Lei Jun says in the book, “Today, many startups like to boast, as if having less than a million users is embarrassing.” However, these million users do not help in refining the product. The most crucial thing in the early stages of a startup is to refine the product with core users.

Even the Most Unassuming Products Have Optimal Solutions

Lei Jun believes that an ultimate product has two conditions: first, the product must be stunning, with impressive design and cost; second, it must exceed user expectations and truly make users scream.

To meet these two points, one must pursue the optimal solution for each stage of industry development. The importance of pursuing excellence in product design lies in the fact that pioneers have many advantages, and if one has the pursuit of excellence, they can often come up with the only answer. For example, before fingerprint unlocking technology matured, Xiaomi’s R&D team designed 652 unlocking solutions and found that Apple’s slide-to-unlock was the optimal solution.

The optimal solution often does not lie in technical indicators because, according to Moore’s Law, technical indicators will always improve. The optimal solution is a more simplified or integrated implementation based on user needs.

In many fields where product forms have not changed for decades, the optimal solution may not have been found. For example, calculators are generally considered to have no room for innovation, but the MIUI calculator app integrates functions like currency exchange, unit conversion, personal tax, mortgage calculations, and even “relative title calculations,” such as “my husband’s brother’s daughter.”

Similarly, power strips are considered a mature industry, but Xiaomi’s power strip has made many improvements in aesthetics and was the first to propose a structure with three strong and three weak electrical sockets, solving the technical challenge of being compact yet safe. Even the boss of Bull said to Lei Jun that Xiaomi’s power strip has advanced the industry significantly.

Lei Jun says, “If you haven’t found a recognized optimal solution in your focused field, then congratulations, you still have a great opportunity to keep approaching it, find it, and establish a strong competitive advantage.”

The Contradiction Between “Touching People’s Hearts” and “Honest Pricing”

Xiaomi’s excellence is not only in design but more importantly in cost-effectiveness. Lei Jun says, cost-effectiveness is not a simple pricing strategy but a systematic capability. For example, traditional phones have high gross margins because of high sales channel fees and marketing costs. Xiaomi uses internet tools, primarily new media marketing, with products that naturally attract traffic, resulting in low market promotion and advertising costs; it uses e-commerce sales, eliminating many intermediaries, and thus sales and channel costs are also low.

I believe that currently, AI is also in a period where the inference cost of large models is high, somewhat like Xiaomi phones, where the BOM cost accounts for a large proportion of the selling price. So, whoever can adopt a more cost-effective model combination and optimize the system well can launch more cost-effective AI products. If the cost drops to the same level as internet products, it might even break out of the logic of AI products needing to guide paid subscriptions and promote them in the free mode of the internet.

When looking at AI products, I generally use the “cost per thousand minutes” metric, which is how much OpEx (variable cost) is generated on average for every 1,000 minutes of product use, including model API fees, GPU and CPU server rental fees, public network bandwidth fees, storage fees, etc.

Similarweb statistics of several websites' visits in September 2024

The above image shows the visits of several websites in September 2024, as counted by Similarweb, including the USTC Course Evaluation Community (a course evaluation website for USTC students) that I participated in developing, an AI tool GenSpark for generating wikis and research reports, the most authoritative large model evaluation website Chatbot Arena, one of the largest AI applications on the desktop in China, Kimi, and the largest professional content community in China, Zhihu. Due to the data source, Similarweb severely underestimates the visits of domestic websites and only counts web visits, not the app visits, which obviously account for a larger proportion. For example, the actual page visits seen from the server logs of the Course Evaluation Community in September were 7.22 million times (the Course Evaluation Community does not have an app), but Similarweb gave 1.86 million times. Zhihu’s actual monthly page visits exceed 30 billion times, most of which are mobile visits, but Similarweb gave 1 billion times. However, we do not have authoritative data that can fairly compare any website, so we use Similarweb’s estimates.

According to Similarweb’s statistics, the USTC Course Evaluation Community had 650,000 visits in September 2024 (each browser is considered one visit within 30 minutes), with an average stay of 5 minutes per visit, so the total user stay time in September was 3 million minutes. The cloud server rental and bandwidth cost of the Course Evaluation Community is about $300 per month (thanks to taoky for optimizing the search function to reduce cloud server rental costs, and thanks to Smart-Hypercube for suggesting download caching to reduce bandwidth costs). So the cost for users to use the Course Evaluation Community for 1,000 minutes is $0.1.

For medium-sized video conferencing applications, the cost per thousand minutes is about $0.3, mainly due to traffic costs. Short video applications have a cost of about $0.2 per thousand minutes, also mainly due to traffic costs, but since it is CDN traffic, it is much cheaper. For applications like Zhihu, which mix text content, the cost per thousand minutes is generally below $0.1.

A big problem with AI applications is the high cost. For example, the Realtime API cost of GPT-4o voice calls is as high as $300 per thousand minutes, which is obviously unaffordable for most companies. If you use OpenAI’s voice recognition, text model, and voice synthesis capabilities to form a voice call, the cost is about $15 per thousand minutes, which is still too high, and the end-to-end delay will be as high as 2 seconds, making users feel that AI is slow to respond. But if you use the latest open-source model combination and appropriate system optimization, the cost of AI voice calls can be reduced to below $1 per thousand minutes, which is on par with internet applications, and the end-to-end delay is below 1 second, similar to human reaction time. Therefore, voice calls, which were generally paid services, will soon move towards the free mode of the internet.

Similarly, for tool applications like ChatGPT, using GPT-4o mini, where users ask one question per minute on average, with an average input of 2K tokens and output of 500 tokens per question, the cost per thousand minutes is $0.6. Although such AI applications still have higher costs than internet applications, they have at least dropped to a similar level, making it no longer an issue to let users use them for free.

Looking at a star startup like GenSpark, generating research reports requires 10 GPT-4o mini-level models to generate in parallel, and the cost per thousand minutes for users is only $6, with the total monthly large model cost of the entire website controllable at around $10,000, which individual developers can basically afford. For a general application like Kimi, a GPT-4o mini-level model is basically enough, and if Kimi’s 25 million desktop monthly active users use GPT-4o mini, the monthly large model cost is only about $1.5 million. If an AI startup has so many monthly active users, it can definitely raise money to burn a million dollars a month. According to third-party estimates, Kimi’s customer acquisition cost through advertising is as high as $1.5, far exceeding the cost of large model calls. Therefore, as long as cost-effective models are used, AI companies do not have to worry about large model costs becoming an issue.

As for search applications like Perplexity, much of the traffic is browsing content generated by other users in the Explore section, which is a kill-time scenario; even in tool scenarios where users ask questions themselves, many questions are similar and can be cached. Since Perplexity has improved the read-write ratio, costs are further reduced, so it may only need to rely on advertising to cover operating costs.

Previously, when talking with some game companies, they said that the reason they haven’t implemented AI NPCs yet is mainly due to high costs. If AI uses the GPT-4o Vision model to understand game screen content in real-time and perform actions, the cost per thousand minutes will be as high as hundreds of dollars, which is completely unacceptable. They said that reducing costs by 100 times might not be enough; it needs to be reduced by 1,000 times so that AI costs are only a small part of the entire game cost, and AI NPCs can be widely popularized. Now, the latest open-source model can be done with a single 4090 GPU, which means the cost is $0.2 per hour, $3 per thousand minutes, but this is only close to the cost of the game itself, not yet a small part.

The key competitive advantage of Xiaomi products is cost-effectiveness and design. Is it possible for a company to win in the AI field with cost-effectiveness and design? This question is left to the readers.

Even the Unseen Parts Must Be Excellent

Lei Jun gave the example of Steve Jobs, who demanded that circuit boards also look good, as excellent carpenters would not use inferior wood for the back of cabinets, even if no one could see it.

Take Xiaomi phone packaging boxes as an example. Xiaomi’s packaging boxes not only have a minimalist design style but also use sturdy kraft paper as raw material, which can withstand the weight of two people weighing over 350 combined without breaking.

For chargers, previous laptop chargers were large and unattractive, and none were as beautiful as Apple’s chargers. There is actually a material science problem behind this. Xiaomi only made GaN (gallium nitride) chargers with upstream suppliers in 2020, which are not only fast charging but also compact and exquisite. In just two years, GaN technology has rapidly spread in high-end product lines across the industry.

For software, the unseen part is code quality.

I feel that I am somewhat lacking in the pursuit of excellence and need to learn from Lei Jun.

During my undergraduate years in the Linux User Group (LUG), although my code had version control, there was no distinction between test and stable versions, and the code was deployed directly to the production server without testing, often causing service outages. LUG’s mentor, jameszhang, said that the software I made lacked the concept of versions. It was with the help of other LUG members that I got Docker deployment, CI, and CD working, which superficially reduced development efficiency but improved service stability and made it easier for other students to maintain.

Later, after working, I found that the quality of code style is not much related to technical level but rather to personality. Those who write clean and tidy code, with complete tests and few bugs upon submission, are often perfectionists who pursue excellence and are often J (Judging) types in MBTI, more suited for engineering. Those whose code lacks aligned indentation, barely passes code style checks, and has obvious bugs upon testing often have a laid-back work attitude and are often P (Prospecting) types in MBTI, more suited for research. The reason many AI researchers have poor code style when doing engineering is not necessarily a technical level issue but a personality trait. Of course, researchers with strong engineering skills and enough self-control can also write high-quality code.

When working on the Course Evaluation Community, jenny42 said I worked quickly, and after she proposed a requirement, I quickly implemented it, but it was rather rough. Once, when crawling teacher data, there were some abnormal data with format errors, and I handled some of them, but there were still some uncleaned among thousands of data, which cuihao couldn’t use. The open-source code of the Course Evaluation Community lacked a database initialization script, and many students who wanted to participate in the development of the Course Evaluation Community encountered difficulties when deploying the development environment locally. Students from other schools once said that they wanted to use the open-source code of the Course Evaluation Community, but found that our database table structure was highly coupled with USTC’s course setup and not universal, so they had to redevelop it themselves.

When writing a dissertation, sometimes data anomalies occur, and retesting is done to resolve the issue. My advisor mentioned that in other people’s papers, a script can directly measure all the data, and even the charts in the paper can be updated automatically. At that time, papers in the networking field did not require submitting artifacts, but now reproducibility is a basic requirement.

Extreme is not self-indulgence

Lei Jun said, extreme is not self-indulgence and self-moved. For example, Xiaomi once milled the Xiaomi logo on every screw, which was five times more expensive than other suppliers. Lei Jun found it amusing because it was just a show of skill and sentiment.

The effort to pursue the extreme needs to provide value to users. For example, making the circuit board neat and writing the software code neatly, although users can’t see it, actually ensures product quality and is valuable to users.

Reputation

Xiaomi products mainly win by word of mouth. I consider myself a Xiaomi fan. Before switching to Google Pixel in 2017, I had been using Xiaomi phones and bought three generations of them. I was also an early user of several popular Xiaomi products like the Xiaomi Band, Xiaomi power strip, Xiaomi air purifier, and Xiaomi power bank. For a long time, my family used a complete set of Xiaomi products, including Xiaomi TV, smart speakers, surveillance cameras, headphones, scales, routers, desk lamps, and remote controls. The Xiaomi power strip in my home has been used since 2015 and hasn’t broken yet.

The key to reputation is exceeding expectations

Lei Jun compared Haidilao and the Burj Al Arab Hotel. The Burj Al Arab Hotel claims to be the best hotel in the world, but Lei Jun felt it was “so tacky” because his expectations were too high. In contrast, some small details in Haidilao’s service moved him because Haidilao’s location and pricing are very friendly.

The success or failure of Xiaomi products is also related to user expectations. A successful example is the Xiaomi 1 in 2011, which had flagship-level configurations and only used suppliers other than Apple, selling for 1999 yuan with a BOM price of 2000 yuan, meaning no profit from hardware. When the price was announced at the launch event, the applause lasted for a full minute, and the number of orders far exceeded expectations. A failed example is an air conditioner priced at 4000 yuan in the Xiaomi ecosystem, which, despite having many features, was priced significantly higher than similar products, failing to meet user expectations for a 4000 yuan air conditioner, resulting in widespread criticism.

Lei Jun said he pondered for a long time before understanding this principle: A good product does not necessarily bring reputation, a cheap product does not necessarily bring reputation, and a good and cheap product does not necessarily bring reputation. Only products that exceed expectations can bring reputation.

Lei Jun believes that every consumer has a psychological estimate of the average performance of social products/services when purchasing products/services. Satisfaction comes when the service exceeds user expectations. Therefore, no matter what products/services we provide, we must strive to exceed the industry’s or society’s general expectations for comparable levels. In summary, the source of reputation is to always maintain a significant comparative advantage over peers.

The book cites successful products like the Xiaomi Band, Xiaomi power bank, Xiaomi air purifier, and Xiaomi power strip. Traditionally, these small appliances are not only unattractive but also expensive if of good quality. Xiaomi’s ecosystem products greatly exceeded user expectations in terms of aesthetics and cost-effectiveness.

In the early days of Xiaomi’s startup, Lei Jun chose to remain anonymous. I think it was not only to pursue excellence in products but also to manage user expectations. MIUI was released as an interest group, and a big shot on the XDA forum was surprised, saying, “How is this ROM so well made?” If people knew from the start that Lei Jun was behind MIUI, which was still imperfect, would anyone say, “How can such a big shot like Lei Jun only make a custom ROM with phone, SMS, contacts, and desktop?” In fact, after Lei Jun released the first Xiaomi phone in 2011, many people were skeptical about him making phones when he attended the Geek Park Entrepreneur Salon.

I found that many AI startups also encounter this issue of managing user expectations. Early products of startups cannot be perfect. After many startups release large models or products, people say, “How can a company that raised so much money and has such a strong team only produce this?” Entrepreneurs know that without the resources of big companies, doing anything is not easy, but ordinary people only think that the successful products made by big company executives before are so successful, so why are the things made after starting a business so mediocre?

Some AI companies take the open-source route, which I feel aligns with the idea of ‘exceeding expectations.’ If a product or model is open-sourced and can indeed solve users’ pain points, even if there are some flaws, users will not say much when comparing it to the quality of most open-source projects. But if a product is heavily promoted before release, raising users’ expectations, and then priced beyond expectations, the result will definitely be criticism.

Being the coolest company in users’ hearts

Xiaomi’s vision is “to be the coolest company in users’ hearts.” Lei Jun hopes that every Xiaomi product has high cost-effectiveness and can be bought with eyes closed. In the early days of Xiaomi, indeed, all the small appliances in my home were Xiaomi’s, and I never had a bad experience. Lei Jun also admitted that the threshold for user reputation is getting higher and higher, and as Xiaomi’s ecosystem grows larger, it is difficult to maintain reputation consistently.

In Lei Jun’s words, the first element of a blockbuster product is whether the product has a “tomorrow attribute.” The tomorrow attribute is to provide users with a new experience representing advanced trends, which they yearn for, and once they use it, they don’t want to let go. Because everyone yearns for it, these blockbuster products have strong self-propagation power. When Lei Jun was a child, whoever bought a TV would have the most lively house every night, indicating that TVs were destined to enter thousands of households, stronger than any advertisement.

Why be a cool company? Because users are willing to spontaneously spread cool products and contribute to cool companies. For example, early MIUI had 100,000 users, equivalent to a 100,000-person development team. Users would find bugs and suggest feature improvements and requirements. Especially beta users are often seasoned enthusiasts, providing Xiaomi with suggestions even more professional than internal employees.

I think it’s the same in the AI field. Successful products almost always make users feel cool to use and willing to promote them to friends. The companies behind these products also strive to create a tech-driven, cool feeling, rather than being filled with the stench of commerce.

For example, I often promote Cursor to friends around me. Actually, I haven’t received a penny from the Cursor team, but my development efficiency has doubled using Cursor, and many excellent programmers around me are using Cursor, so I feel using Cursor is a cool thing.

Other successful AI products, including ChatGPT, Perplexity, Claude Artifacts, NotebookLM, etc., all offered a brand new experience when they first came out, and once used, you don’t want to let go.

Making friends with users

Lei Jun said many companies claim to treat customers as gods, but actually try every means to extract more money from users. For example, in many industries, prices are not transparent, and only by finding connections can you get internal prices, causing many sales costs.

When founding Xiaomi, Lei Jun said, from the user’s perspective, being a “god” is too vague. Can you be my friend? Just don’t cheat me. Therefore, Xiaomi proposed “making friends with users.” This is completely different from fan economy, which is applicable to the creative industry, but a brand facing mass consumption cannot rely on personal worship of the founder. The core of “making friends with users” is respecting users, listening to their opinions, and taking action.

Making friends with users is not an easy task. Lei Jun admitted that many people, including internal colleagues and external friends, advised him to have employees keep quiet on new media because people often say the wrong things, and it’s impossible to pre-screen, constantly causing PR crises, and the brand’s loss is immeasurable. Many colleagues also said that live streaming is for sales anchors, and as an entrepreneur, you need to maintain an image. To appear high-end, you need a sense of distance.

Lei Jun said that compared to other titles like entrepreneur and angel investor, being a well-known digital blogger actually better summarizes his daily public performance. When he appeared on the National Day 70th Anniversary float in 2019, digital KOLs in the industry felt it was a face-saving moment for the “digital blogger” group. Using new media well is a profound skill, requiring an understanding of new media product forms and communication skills with users and fans. Moreover, as a public company, speaking on new media requires a more comprehensive awareness.

The biggest benefit new media brings to Xiaomi is efficient communication with users, listening to their opinions, and revising product issues 7x24 hours, avoiding directional errors caused by closed-door development. At the same time, new media is the most effective communication platform of this era.

The Mi Fan community is not only for users to discover problems and help improve products but also sincerely hopes to create emotional connections outside of work. Therefore, besides online communication, Xiaomi also established offline clubs, and many Mi Fans even found partners in the Mi Fan Club.

Lei Jun said in the book that he is relatively introverted, but when facing users, being thick-skinned is okay. For example, “R U OK,” colleagues disagreed with Lei Jun responding to the Bilibili ghost video, but Lei Jun insisted on making a Bilibili video to greet young users, solidifying his “singer” identity, and since then, “R U OK” has become a long-lasting Xiaomi meme.

Small and beautiful or big company

Why is Xiaomi not as cool today? The “New Retail” chapter in the book also reveals Xiaomi’s difficulties.

In 2015, Xiaomi’s shipment growth slowed, and a series of problems accumulated during the high-speed development period erupted at the end of the year; in 2016, Xiaomi’s shipments declined. In the history of mobile phones, companies whose shipments declined and then rebounded are almost nonexistent, so Xiaomi’s model faced severe challenges.

Xiaomi faced a major choice: whether to insist on only doing e-commerce and be a small and beautiful company or to become a big company.

Lei Jun said they almost had no choice. If Xiaomi were doing some non-standard categories, then being small and beautiful would be a suitable choice. But mobile phones are a highly standardized electronic technology industry. Without sufficient scale, there is no sufficient R&D capability, nor sufficient supply chain support, so being small and beautiful has long been proven to be a false proposition. Therefore, facing the “fundamentalist” belief issue of whether to do offline retail, they had to bow to reality and break the “purity of the Xiaomi model” iron rule.

OpenAI’s changes over the past year are also a good example. If OpenAI were just a company doing industry applications or domain-specific models (such as speech synthesis models), then being small and beautiful would be a suitable choice. But a general foundational model is something that must have sufficient scale to do, so it must get enough money, and investors are obviously not doing charity. As an engineer, I can understand the feelings of technical experts like Ilya Suskever; as an entrepreneur, I can also understand Sam Altman’s difficulties.

In the past year of entrepreneurship, I have also met some entrepreneurs and investors with net assets of tens of millions or even hundreds of millions of dollars, most of whom are in a state of having money but no time to spend. They are much more hardworking than 996 big company workers, almost 007, and have much more pressure than workers. Because entrepreneurs take investors’ money, investors will require the company to grow, go public, and be responsible for the stock price after going public. And investors’ money is often not all their own, and with such a high failure rate of entrepreneurship, they also have great pressure.

Entrepreneurship is a lifestyle. In fact, to accomplish something, besides entrepreneurship, I believe there are many ways:

• A typical startup company needs to solve clear needs, have a business model, and have a large imagination space. It needs to introduce external investors to provide more resources, help the company expand rapidly, and the ultimate goal is generally to go public or be acquired.
• The “small and beautiful” company that Lei Jun mentioned is more suitable for projects that can solve clear needs and have a business model but are not easy to grow.
  • Many companies that remain “cool” for a long time are small and beautiful companies. Because small and beautiful companies focus on a specific niche, if competition is not fierce, they can live comfortably. Domestic internet companies that can achieve 965 (work-life balance) are either foreign companies or such small and beautiful companies.
  • If the business model is clear, small and beautiful companies can even do without investors. But Lei Jun believes that investor supervision can help standardize company operations, avoiding turning into a small workshop like Lei Jun’s first entrepreneurial attempt. Small workshops not only have a high failure rate but also easily bring financial and legal risks to the founder.
• Community, more suitable for projects with clear needs but lacking a business model, often serving a public good.
  • Open-source community: It’s hard to imagine Linus walking into an angel investor’s office with the initial version of Linux, saying he wants to make an open operating system similar to UNIX. Open-source communities can also commercialize, but most open-source communities’ commercialization progress is not smooth because it’s hard to balance the team’s commercial interests and the community’s public interests.
  • Non-profit organization: It’s also hard to imagine Jimmy Wales walking into an angel investor’s office, saying he wants to make an encyclopedia shared by all humanity. Non-profit organizations rely on love to power, and maintaining a balance of income and expenditure is not easy.
  • Web3 community: Similarly, it’s hard to imagine the founder of Bitcoin walking into an angel investor’s office, saying he wants to make a decentralized, unregulated digital currency. Web3 is a more mature business model for community projects, but the market is filled with speculative financial projects, and idealistic fundamentalists are relatively marginalized.
• Academic research, more suitable for projects that sound interesting but lack clear needs.
• Big company, more suitable for things that can solve clear needs and have a business model but require specific big company resources (such as technology, platform, users, customers) to succeed. Many of these are optimization tasks, such as optimizing network transmission protocols for a 10,000-card cluster, which alone is not enough to support a startup, and open-source communities and academia do not have 10,000-card clusters, so it’s more suitable to do within a big company.

I found that once most people start a business, it’s hard for them to return to a big company as an ordinary cog in the machine. The reason is that entrepreneurs often have many of their own ideas, thinking about what model to use and what resources are needed to realize these ideas. Even if they go to a big company, it’s like internal entrepreneurship, using the company’s resources to get things done. Most big company employees don’t know what they should do and can only wait for instructions from their leaders.

Fast

The importance of “fast” in the internet industry is self-evident. For example, an MVP should be developed in two to three months, no more than half a year at most, and then quickly iterate based on user feedback. Products with poor market feedback should stop losses early.

Lei Jun gave the example of a 98-inch ultra-large screen TV. Ding Lei wanted a 100-inch TV, and Lei Jun thought it was a valuable demand, so he made a 98-inch TV priced at 19,999 yuan in five months. From project initiation to market release, it took only five months, a speed unimaginable in traditional manufacturing. Before Xiaomi entered the market, the average monthly sales of 98-inch and above TVs in the Chinese market were only 78 units. After Xiaomi entered, its average monthly sales reached 5,000 units in half a year. Another half year later, 98-inch TVs became a hot spot for domestic TV brands to compete.

Fast is a unique global capability of the internet development model

Is the pursuit of perfection contradictory to “fast”? Lei Jun said, “Fast” sounds simple, but it is essentially a unique global capability, a huge difference between traditional software engineering development models and internet development models.

1. There needs to be a product line distinction between beta, development, and stable versions;
2. A community gathering enthusiasts needs to be built, supported by operational products through four-quadrant feedback reports, user opinion voting, etc.;
3. There needs to be a group of enthusiasts who see the team as friends and sincerely participate in co-creation design and development;
4. There needs to be a passionate, highly self-driven, and technically outstanding team that refines products with the enthusiasm of “optimizing before eating.”

In 2011, Microsoft executives came to Xiaomi to exchange ideas on how Xiaomi managed weekly updates, while Microsoft feared breaking users’ systems with each upgrade. Lei Jun said Xiaomi created two partitions for the new and old systems, automatically rolling back to the previous version if the update failed. Microsoft executives were shocked.

Building a passionate, highly self-driven, and technically outstanding team

Lei Jun said, to build a passionate, highly self-driven, and technically outstanding team, one must be particularly serious and strict when recruiting. Even someone as influential as Lei Jun said that building a team is an extremely painful process.

In the first half year of Xiaomi’s establishment, Lei Jun spent 80% of his time recruiting partners. To do the “Ironman Triathlon” of software, hardware, and the internet, he recruited the seven most important top talents from Google, Microsoft, and Motorola, forming a “combination of local and foreign” company. He found that the problem with domestic products was not just technical but more about design and user experience. Therefore, among the eight founders, there were six engineers and two designers. These partners could each manage a part and quickly form a whole, enabling Xiaomi to use this innovative model.

Lei Jun said that when he interviewed each top talent, they were also interviewing him. His experience was to be thick-skinned; while others might visit three times, he could visit thirty times. For an intern, Xiaomi conducted nine interviews, including two co-founders as interviewers. Besides investing time, one must also do well with existing products and businesses to showcase the company’s future development space and opportunities, achieving the effect of attracting talents. Besides salary and options, the most attractive thing for employees is whether they can work with top talents and grow personally in work and career.

Lei Jun said, “What we want to do is very difficult. Being capable is not enough; we must find people with a shared vision, ability, and responsibility, who can self-drive and work without a manager watching over them. Because such a complex model as the Ironman Triathlon cannot be managed out, it can only rely on enormous initiative. At the same time, trust is an extremely powerful force. We are willing to trust every colleague, so we must be serious and cautious when recruiting.”

Lei Jun gave two counterexamples. First, if someone joins with a try-it-out mentality or has other things going on simultaneously, Xiaomi absolutely does not welcome such people. Second, a candidate had a perfect resume and excellent sales performance but claimed to have the ability to turn straw into gold, which is completely contrary to Xiaomi’s values. Xiaomi’s value is to be friends with consumers and make good products that touch people’s hearts and are reasonably priced.

Lei Jun said that the goals pursued by startups are always higher than their current capabilities. Many positions are not quite suitable, but they can’t afford to hire people at high prices, so they have to present a “small horse pulling a big cart” state. At this time, one must use a magnifying glass to find strengths, gradually build confidence, and not let the company emit too much pessimism.

Local slowness is global fast

Lei Jun’s statement means not to cover up strategic laziness with tactical diligence. Xiaomi’s early ultra-high-speed growth covered up many problems, which exploded at the end of 2015.

A few days ago, a big shot from MSRA had dinner with me, saying that the joint training students I recommended to MSRA for internships were very smart but often didn’t come to the company, with some giving reports outside, and their mentors didn’t know. In the history of MSRA’s joint training, there were very few who quit PhD, but among the students I recommended, there were several. I said they are smarter than me, and there’s no need to worry about their career development. They found a truly interesting blue ocean industry and chose to do what they are interested in and economically rewarding. A few years ago, some of them often complained to me about not achieving research results and not publishing papers. But after finding the blue ocean industry, they discovered that some very simple things were not done by anyone, making it easy to realize their value.

Lei Jun’s investment company is called Shunwei Capital, meaning to follow the trend. Lei Jun’s famous saying, “Standing on the wind, even pigs can fly,” was summarized from his insight at a dinner at the age of 40 and was misunderstood by many as opportunism. At that time, Lei Jun’s original words were, “I realized that people cannot push stones up the mountain; it will be very tiring and will be hit by stones rolling down from the mountain at any time. What needs to be done is to climb to the top of the mountain first and kick a stone down at will.”

The days of pushing stones up the mountain that Lei Jun referred to were at Kingsoft. “The colleagues at Kingsoft were very diligent and hardworking, gathering a group of the smartest engineers. But this high-tech company, founded for 16 years, took a full eight years to complete its listing.” He repeatedly expressed regret in his book for not entering the internet industry earlier. He was exposed to the internet very early, discussing with Ma Huateng, Ding Lei, and others online every day, but due to being refused a visa to the United States, he couldn’t form a deep understanding during the booming internet entrepreneurship in the 1990s. Later, the excellent network he worked on at Kingsoft had to be sold to Amazon because the company couldn’t raise enough funds to support it.

Summary

The book “Xiaomi Entrepreneurship Thoughts” summarizes the core concepts behind Xiaomi’s success through Lei Jun’s entrepreneurial experience, especially the “Seven Words of the Internet” proposed in Chapter 6: Focus, Extreme, Reputation, Fast.

The book emphasizes the importance of focus in entrepreneurship. Lei Jun reviewed the lessons from his first entrepreneurial failure, believing that both the enterprise and products must focus. Lei Jun shared Xiaomi’s process of creating explosive products by focusing on the most urgent needs of users, solving core problems, and quickly iterating on products like MIUI and air purifiers.

He pointed out that the key to extreme products is to exceed user expectations, including cost performance and design, and this extreme pursuit often brings a strong reputation effect.

Lei Jun also emphasized that entrepreneurship is not only a process of solving business needs but also making friends with users and listening to user feedback. Successful companies not only pursue perfection in products but also manage user expectations, creating “cool” products with “tomorrow attributes” that users are willing to promote spontaneously.