NVIDIA CEO adopts flat structure

Nvidia is no longer just a chip company. It is trying to coordinate GPUs, CPUs, networking, software, power, cooling, racks, and entire AI systems as one stack. That kind of work usually creates more management layers. Jensen Huang says he went the other way. In a March 23, 2026 conversation with Lex Fridman, the Nvidia CEO described running the company with more than 60 direct reports and without routine one-on-one meetings. ### Why is this suddenly interesting? Because the management style matches the product problem. Huang spent a lot of that interview talking about “extreme co-design” — basically, Nvidia now has to make many interdependent parts work together at once, not hand work cleanly from one silo to another. If the company is building systems where every tradeoff touches five other teams, a flatter org starts to look less like a quirk and more like infrastructure. (lexfridman.com) ### What exactly is the structure? Huang has described two versions of it in public. In a Stanford talk from March 2024, he said 55 people reported directly to him and called the setup “designed for agility” and fast information flow. In the March 2026 podcast, he said he had more than 60 direct reports. So the exact count moved, but the principle did not — keep layers thin and keep information moving directly. (lexfridman.com) ### Why skip one-on-ones? His argument is simple. Private recurring meetings create bottlenecks and uneven access to information. Huang has said he does not schedule one-on-ones unless someone needs him, and then he will “drop everything.” The default is group problem-solving — put the issue on the table and have everyone attack it together. That reduces the classic telephone-game problem where each layer edits the message before the next team hears it. (cnbc.com) ### Doesn’t that sound chaotic? A little — but that is also the point. Huang seems to prefer productive mess over tidy hierarchy. The tradeoff is obvious. You get speed, direct context, and fewer political choke points. But you also put a lot of load on one person, and the model depends heavily on having unusually strong senior leaders who do not need constant supervision. Even some coverage of Nvidia’s org chart framed this as both a speed advantage and a governance risk. (cnbc.com) ### Where does “character” fit in? This is the part people miss if they focus only on the org chart. Huang has repeatedly argued that a company’s character matters more than slogans on a wall. In his telling, resilience, adaptability, and the ability to handle adversity live in people, not in formal process. A flat structure works only if the people inside it share judgment, trust, and mission. Otherwise you just get noise. (sovereignmagazine.com) ### Why does that matter for engineering teams? Because Nvidia’s hardest work now crosses boundaries constantly. A VLSI team cannot optimize in isolation if packaging, memory, networking, thermals, and software all shape the outcome. In that environment, middle-management handoffs can slow decisions or blur tradeoffs. Huang’s setup tries to keep the engineers closest to the problem closer to the decision too. That is not universally portable, but it fits Nvidia’s current shape unusually well. (startuparchive.org) ### Is this a template for everyone? Probably not. Most companies would break if they copied the number and ignored the culture. The real lesson is narrower — if your product requires constant cross-functional tradeoffs, you may need fewer layers, more shared context, and leaders chosen as much for judgment and temperament as for raw technical brilliance. Nvidia’s org chart is extreme. But the logic behind it is pretty clear. (lexfridman.com) ### Bottom line Huang is treating org design like system design. Keep latency low. Keep information loss low. Push decisions closer to the people who understand the constraints. For a company trying to build the full AI stack at once, that is not management theater — it is part of the product strategy. (lexfridman.com) (startuparchive.org)