Wang Yajun on Jensen Huang’s Organizational System: Distributed Decision-Making for Innovation

The episode opens with a line from Jensen Huang: “Strategy is not words; strategy is action.” It then introduces NVIDIA’s counterintuitive practice of paying its executives equally, signaling that the conversation is about how an organization turns principles into daily work, not about one-off management tricks.

Zhang Xiaojun introduces Wang Yajun, former CSO of Longfor Group and a researcher of strategy and organization. Many entrepreneurs return from visits to NVIDIA surprised by Huang’s no-layoff policy and 60 direct reports. The episode aims to place these practices within a complete organizational system.

Zhang notes that NVIDIA’s management is often called counterintuitive: Huang does not hold one-on-ones, has 60 executives reporting directly to him, and rarely lays people off. She asks Wang for an overarching framework before judging each practice individually.

Wang says he began by studying strategy, then realized that understanding how an organization works is essential to putting strategy into practice. NVIDIA’s approach is rare in most industries, and any single practice can seem strange in isolation. He prefers to connect management principles to strategic choices through the idea of a highly entrepreneurial, agile organization.

Zhang turns to the first principle: 60 executives report directly to Huang. Traditional management says a leader can only handle a dozen or so direct reports, so she asks whether he can really manage that many people.

Wang first unpacks the word “management.” If it means giving orders and monitoring execution, then 60 people would indeed be hard to control. But Huang says each executive can independently run an area of the business; he does not need to direct them one by one, only create opportunities for them. These are two different logics: control and enablement.

Enablement often fails to take hold because an organization keeps its old decision-making habits. Wang recalls that about 50 people once reported directly to Longfor’s CEO. The team worried that removing middle layers would hurt development and cross-functional problem-solving, but simply adding another executive in charge would not automatically improve decisions.

When departments cannot resolve conflicts, companies often add a management layer to mediate. But the real questions remain: who makes the decision, and on what basis? If the new leader simply decides based on personal judgment, without giving the organization a way to solve problems, the extra layer only lengthens the information loop.

Having 60 executives report directly to Huang does not mean every layer of the company has a span of 60. Huang says the broad span is mainly at the executive level; there are more managers and coordination further down. It works because the direct reports can each own their area independently.

Wang speculates that NVIDIA’s executives and the managers below them may be well matched in ability, responsibility, and organizational experience. Huang’s absence from operational update meetings suggests that the relevant owners handle those details. At his level, he focuses on direction and opportunities instead of pulling every operational detail back to himself.

Huang has also said that the 60 executives are paid equally, down to the dollar or even the cent. Wang sees the common standard as an expression of equality: if everyone carries independent responsibility, differences in rank should not automatically translate into differences in whose voice counts.

Traditional companies use titles such as general manager, vice president, and senior vice president to distinguish seniority. In meetings, that can lead to the assumption that the highest-ranking person wins. NVIDIA’s equal pay for direct reports at least reduces the interference of formal rank and keeps discussion focused on the problem itself.

Huang does not hold one-on-ones, which runs counter to much management advice. Wang does not think one-on-ones are useless; they can help with feedback or personal matters. But Huang prefers to discuss work-related disagreements in the room, where the relevant teams can hear the exchange and learn from it.

Wang connects the relationship between organization and product to Conway’s law: the communication structure inside an organization affects the architecture of its products. The software industry’s shift from monolithic waterfall development to agile methods also shows how a change in communication can reshape collaboration and product design.

When someone makes an unreasonable point in a meeting, Huang explains on the spot what is wrong and frames it as learning rather than criticism. Others can learn from a colleague’s mistake; feedback comes faster than in a private follow-up, and the information is no longer confined to two people.

Zhang asks why traditional companies still use long reporting chains. Wang says they commonly add layers of titles and executives, with decisions handed down from the top. When problems arise, information has to climb back up layer by layer and may be mistaken for noise until the issue grows too large to ignore.

Wang says Huang often uses the phrase “Let’s reason together.” Rather than simply announcing an answer, he and the team break down the problem together. By repeatedly taking part in this reasoning, employees gradually learn how to make judgments and solve problems.

This shared reasoning also helps explain why NVIDIA can rely less on direct control over time. As the organization grows, working methods and judgment spread among its members, so managers do not have to intervene in everything. Wang says some practices only make sense as part of a whole system built over time.

The same chain of command can trap middle managers in relaying messages, compiling spreadsheets, and forwarding information, gradually draining their sense of purpose. Wang says that in the AI era, roles devoted only to moving information and orders may be easier to automate.

Organizational hierarchy can seem natural simply because people see other companies using it. Wang cautions that copying the form does not mean understanding the underlying logic. NVIDIA’s approach is worth studying because it offers a way to improve decision quality, not because it looks novel.

Zhang quotes a listener’s description of NVIDIA’s organization as “like a GPU.” Wang says the analogy fits. He has studied chip architectures and found that AMD’s modular chiplet design and the GPU’s many-core parallelism both help explain how an organization can coordinate multiple units.

Wang gives the example of Baizhu, founder of Youzan, visiting NVIDIA. Baizhu saw Huang hold a company-wide reading meeting each month and concluded that a CEO should personally pay attention to employees two levels down. Wang distinguishes control from context: inspecting how people work is control, while giving everyone the information needed to understand the business creates context for decisions.

A GPU’s individual core may not be more powerful than a CPU, but it has many specialized cores and high-speed connections that can handle numerous tasks at once. Wang uses this to show how an organization can have many capable people work in parallel, then rely on high-bandwidth communication to build collective capacity instead of routing every problem through a central hub.

Huang has said that managers should not treat exclusive information as a source of power. The better informed a team is, the more likely it is to make sound decisions. Deliberately restricting information may seem to consolidate a manager’s authority in the short term, but it harms the organization’s judgment over time.

Extending the chip analogy, Wang says organizational structure is closely connected to the business and the shape of its products. There is little value in asking which organization is best in general; what matters is whether its communication fits the product and market. That fit is not designed once and for all, but evolves through practice.

Zhang asks about pay gaps among executives at ordinary companies. Wang says traditional organizations often have three to five levels of titles and salary bands, and rank can affect who gets heard in a meeting. These differences are not just about pay; they also change how teams collaborate and resolve conflicts.

Companies often match a new hire’s former title to attract outside talent, or the candidate may not join. Wang says this can gradually add more layers to an organization and erode equality. In discussions, people may start caring more about who has the higher rank than whose argument makes sense.

Huang’s meetings may include both executives and interns, and he evaluates an idea by whether it makes sense, not by the speaker’s rank. A company can still preserve equality if culture and shared ways of working absorb title differences. But if people join only for titles and prestige, the organizational system may shift in another direction.

Zhang observes that many companies hold all-hands meetings, but sometimes use them only to announce results and targets. Wang says the tool itself is neutral. An all-hands meeting can provide information for decisions, or it can simply distribute tasks from the top down.

If an all-hands meeting is meant to help more people develop distributed decision-making skills, its content should help them understand the business and think about the future. If it only tells everyone to hit their targets, the meeting remains a platform for issuing orders and cannot create NVIDIA-style organizational capabilities.

Wang says team-building activities such as singing and dancing cannot replace collaboration in real work. Organizations need to build connections while making decisions and solving problems. If companies only copy the outward forms of all-hands or reading meetings, everyone may get tired while the business problems remain unsolved.

On the essence of management, Wang first offers a definition: management means solving problems together, making decisions, and driving action. Organizations exist because individual transactions are costly; working together creates value, and other systems and tools should serve that process.

He adds another important task: continually creating useful new knowledge. Organizations cannot rely only on past experience to make decisions. Through real business practice, they need to discover new methods, a better understanding of markets, and new operating models, then adjust their actions accordingly.

Wang describes innovation as the creation of new knowledge. For example, if a company discovers that AI can substantially improve supply-chain management, it has developed a new understanding of its business. Any discovery that improves organizational capability can become new knowledge: some can be learned from others, while some must be created through the company’s own work.

He offers a resource analogy: oil is not a resource if no one knows how to use it, and the same is true of data. Value appears only when someone finds a use for it. An organization that repeats old work for too long may already be declining, even if its results look fine; it should ask how long it has gone without a new product, method, or understanding of the market.

Zhang asks whether distributed decision-making is right for every company. Wang says top-down decisions and execution may be enough in stable industries with slowly changing demand. But in technology and innovation, by the time information has traveled up the chain and a decision has come back down, the market may have changed.

Distributed decision-making is not about being ideologically progressive; it is about speed. People on the front line encounter change earlier and should be able to process information, make judgments, and act quickly, then feed what they learn back into the organization. In uncertain businesses, this can prevent management from noticing a problem only after the losses have grown.

Wang points out that many Chinese industries once enjoyed sustained growth, with plenty of opportunities to produce and expand, so top-down management worked well. Today, technology- and knowledge-intensive companies need to keep creating new things. Management training designed for the industrial era may not suit businesses changing this quickly.

Zhang mentions that Huang encourages employees to email him their five most important priorities. Wang suspects the purpose is not to collect every opinion, but to help the CEO sense what is happening across the company and get the reality needed for decisions.

For employees to write down their five most important priorities, they first need to know what information matters and feel willing to speak honestly. Leaders can piece together the company’s actual state from these messages, then judge whether its business direction aligns with everyday work. That can be more useful than a long strategic proposal written by one frontline employee.

Wang was also surprised by Huang’s no-layoff policy. He says it begins with a belief that people can grow. Breakthrough work often involves a long, painful period of experimentation; if a team is disbanded before results appear, the learning it has built up is lost.

The line “I will torture you to greatness” is not an endorsement of tormenting employees. It means that for work with long-term value, leaders will go through difficulties with the team and give people time to grow. The process itself must be sound and the team must keep learning; pain is not proof that management is effective.

Wang says a no-layoff approach also requires more careful hiring, with an emphasis on learning and growth. Professional skills matter, but a person unwilling to learn may only be useful to the company for a while. The ability to grow is more likely to help someone adapt as the business changes.

Avoiding layoffs over the long term also means resisting the pursuit of quick money. If a company expands capacity and hires rapidly to meet a sudden surge in demand, it may have no choice but to lay people off when demand falls. Some real-estate companies paid the price for exactly this kind of short-term expansion.

Huang has also said that NVIDIA focuses on doing things others cannot, rather than chasing market share. Wang connects this to innovation: if a company only copies others and pursues an established market, competitors and excess capacity soon follow, leaving the company at the mercy of short-term cycles.

Zhang asks about Huang’s practice of not making plans years in advance. Wang is not surprised: when he studied strategy at Longfor, he found that China’s real-estate market was strongly affected by market conditions, regulation, and financial policy, making it hard to predict what things would look like three years later.

In that highly uncertain environment, Longfor did not rely on five- or three-year strategies. It kept a relatively stable direction while using annual operating plans and ongoing adjustments to respond to change. Wang emphasizes that this is not giving up on planning; it is recognizing the limits of prediction and focusing on responding quickly.

Zhang turns to one of NVIDIA’s stated values: love and care. Wang says Huang’s organizational beliefs are consistent with his long-standing approach to strategy and management. For innovation that requires long-term exploration, control and orders alone are not enough.

Wang says innovation means facing the unknown and trying something new, which often involves many failures before success. If people are punished for a single failure, they will hide risks and avoid trying. To make innovation a core capability, a company must learn to manage failure and keep its costs under control.

Wang sees “Strategy is not words; strategy is action” as a call to unite thought and practice. Writing an ambitious strategy means little if the organization does not act on it. A good strategy shapes which opportunities a company chooses, how it allocates resources, and how it adjusts its operations.

He connects this to lean startup thinking: teams should treat failure as a chance to learn and revise their assumptions based on the results, rather than as proof that someone did not try hard enough. Huang’s public feedback also lets more colleagues learn together instead of keeping every issue inside a one-on-one conversation.

Wang compares it to a child learning to hold a cup: the child may break a few cups, and the adults have to clean up. Without love, it is hard to tolerate the cost of that experimentation patiently. Care at work is similar: when leaders make room for learning, people dare to try new approaches.

Zhang asks Wang what he thinks of love as a corporate value. Wang says he knows little about NVIDIA, but compares it with Netflix’s “context, not control.” Both are exploring how employees can work autonomously, though the cultural expressions feel different to him.

Wang guesses that NVIDIA may face longer and more uncertain innovation cycles, while Netflix’s creative projects can also fail but tend to have shorter cycles. He stresses that this is only a hypothesis based on interviews, not a definitive conclusion about either company.

Taking Huawei as an example, Wang says he has heard that management methods such as the business-leader model may suit relatively stable periods better than innovative businesses. Applying tools designed for a mature company directly to a startup in training can overlook the conditions under which those methods work.

Zhang mentions Huang’s story of rising from a dishwasher to CEO. Wang says that, at least in the way Huang tells such stories, he seems committed to doing things exceptionally well and willing to understand problems through the most basic work.

Wang also discusses Netflix’s small game or film studios: small teams create content, while their relationship with the larger platform may resemble an investment relationship. How an organization connects product teams, platforms, and markets has to be judged in light of the specific business, not inferred from a single management template.

Zhang asks him to compare the organizational and product structures of Netflix, Tesla, and Huawei. Wang says he has not studied these companies systematically, but Conway’s law may apply beyond software: the way a hardware company communicates could also affect its product architecture.

Wang says that doing something at a level above 90 out of 100 requires a different way of thinking. You cannot simply find the best-performing peer and copy its actions; you have to get to the heart of the problem and reason through how to solve it. Company visits often reveal only the surface, and truly understanding a method takes sustained involvement.

He links Huang’s first-principles thinking and reasoning to the need for managers to work out what might happen in the future, rather than merely observing what eight out of ten companies do or projecting from historical data. If you ignore time and change, collecting more examples can actually pull you further from the essence of the problem.

Zhang adds that NVIDIA has about 28,000 employees and a valuation in the trillions of dollars. Wang says the number of people creating value does not necessarily scale with the size of the organization. NVIDIA also began using AI in its own work early, freeing employees to shift from organizing knowledge toward creating more of it.

Wang says AI is good at organizing existing knowledge, but people remain essential to the ongoing creation of useful new knowledge. AI may gradually propose hypotheses, design solutions, and test them; until that process matures, organizations should let employees spend their time exploring and creating instead of merely moving information around.

Zhang shares the story of Huang waiting for his dog to wake up before getting out of bed so he would not disturb it. Wang sees empathy in the story: Huang notices how a being more sensitive than he is might experience sound and vibration. Wang suspects that this habit of thinking may carry over into working with people.

Zhang asks whether Huang is prone to angry outbursts. Wang says public accounts do not suggest that he manages through fear. If a CEO relies on threats to get people to comply, it is hard to sustain collaboration or explain NVIDIA’s continued creativity.

Wang stresses the importance of reasoning things through in an organization: discussions should focus on process and evidence, not assume that seniority or a high title makes someone right. Leaders and teams can build judgment about early signs of future success by reasoning together; demanding results without understanding the process cannot force innovation into existence.

Zhang asks how important trust is in an organization. Wang mentions his book *Enterprise-Level Algorithms* and sums up several values of agile organizations: candor, transparency, equality, and trust. Huang’s openness with information and the space he gives teams to discuss things as equals provide a foundation for trust.

Wang uses a trust formula to explain that competence, empathy, logic, and authenticity all affect whether people trust you. Logic lets colleagues anticipate how you will act, while authenticity means your words and actions align. If a leader is carried away by emotion, the team cannot feel secure when collaboration matters most.

Trust cannot be built through team outings or slogans alone. Wang believes leaders have to take the first step, put themselves in a position to speak honestly and accept trust. They may be let down along the way, but that is part of leading the effort to build a relationship; trust has to be tested and deepened gradually.

Zhang asks what happens when an organization lacks trust over time. Wang says organizations exist to lower the cost of individual transactions, so internal trust should be stronger than trust outside the organization. If executives only answer upward and undermine one another, cross-functional collaboration weakens and the team struggles to complete complex work together.

Trust evolves gradually: people test one another and follow through on small things, then extend more trust based on the response. Trust carries risk, so a major decision should not rest on one person. But it is unrealistic to demand complete trust from everyone at the outset.

Wang recalls a question people outside Longfor used to ask: how could the chairman and CEO avoid getting pulled into daily business decisions while the company still ran well? He says the key was lateral collaboration: teams could solve problems at the same level, leaving senior leaders free to focus on how the system worked instead of personally deciding every detail.

When lateral collaboration is weak, an organization becomes layers of orders and supervision, sometimes using digital systems to monitor employees. Wang supports using digital tools to gather information and provide feedback, but if the aim is to monitor people because they are not trusted, digital transformation will not create an agile organization.

Wang says a company first has to recognize, on rational grounds, that trust has business value. Over a horizon of five or ten years, a trust-based organization is more likely to keep innovating. But trust must become part of everyday collaboration and business decisions; it cannot be confined to HR activities.

Leaders should set the example, kindle the first spark of trust, and let the team respond over time. An organization also has to make a choice about people who are highly capable but continually damage trust and weaken others; depending on an individual’s skill in the short term can undermine collaboration over the long term.

Wang compares an organization to a bamboo grove: above ground, you see separate stalks, but beneath the surface their roots are connected. An organization’s real resilience comes from invisible lateral ties and shared ways of working, not from boxing departments in like boards and relying on the CEO to serve as the bottom of the bucket.

For NVIDIA’s broad reporting structure to work, teams need the capacity to collaborate organically. Wang says this does not emerge from holding a few more meetings; it takes organizational practices with methods and training that can keep growing over time. The conversation closes by returning to the idea of an organization like a GPU.

Zhang closes the show by inviting listeners to subscribe and share in the comments which guests and topics they would like to hear about. The interview began with NVIDIA’s management details and ended with a broader question: how can an organization create the conditions for sustained innovation through candid information, shared reasoning, and lateral collaboration?