Post

Cognitive Framework | Power Law: Why I Still Believe NVIDIA Is the Biggest Winner in AI? A few days ago, I recommended Peter Thiel's "Zero to One." What impressed me most was not "competition is for losers," but a thinking model that can analyze almost all outstanding companies—the Power Law. Cognitive Framework | Power Law Many people researching a company first look at PE, revenue, profit, and cash flow. These are certainly important, but in my view, a more important question is: Does it have the chance to become the one that takes most of the industry's value? I usually use the four questions of the Power Law to judge. Winner Takes All: Is this a "winner takes all" industry? Not all industries give rise to super-giants. In dining, clothing, real estate, most of the time many companies make money together. But the AI chip industry is different; it’s more like search engines, social platforms, operating systems—once you lead, you get further ahead. I believe NVIDIA fits the Power Law mainly due to three advantages: First is network effect—the more people use it, the more valuable it becomes, like WeChat, where everyone uses it so newcomers continue to use it. NVIDIA not only sells GPUs but also built a development platform called CUDA, which you can think of as the "Windows" of the AI world. The more developers there are, the richer the software; richer software attracts more developers, eventually forming an increasingly strong ecosystem. Second is scale effect—the bigger the company, the more obvious the advantage. NVIDIA invests tens of billions of dollars annually in R&D, securing the most advanced wafer capacity in advance and attracting the world’s top AI talent, advantages that small companies find hard to replicate. Third is brand effect—today, many enterprises building AI data centers first think of NVIDIA; the brand itself has become a competitive edge. Image Secret Bet: Has it bet on a future others don’t believe in? Peter Thiel’s favorite question is: "What important truth do you believe that most people disagree with?" Great companies almost always have their own "contrarian consensus." NVIDIA’s contrarian consensus was insisting on investing in CUDA. Twenty years ago, most thought GPUs were just for gaming, but NVIDIA believed it would become a general computing platform. After ChatGPT appeared, everyone realized that AI’s biggest need was GPUs. Many saw the AI boom; I saw a 20-year strategic commitment finally paying off. Image Escaping Competition: Has it made it hard for customers to leave? A true moat isn’t just a better product but making it very hard for customers to switch even if they want to. If an AI company has invested tens of millions of dollars and built all models on CUDA, switching platforms means engineers must relearn, code must be redeveloped, models retrained, and the company bears huge time and financial costs. So many companies don’t want to switch; they simply can’t afford to. This is switching cost and one of the strongest moats. Image Rule Maker: Has it started setting industry rules? I think this is NVIDIA’s most terrifying aspect. Many think it just sells GPUs, but it actually sells the entire AI infrastructure. CUDA is the software platform for AI development, equivalent to the operating system in the AI world; DGX is a supercomputer specialized for AI training—enterprises can quickly train large models without building from scratch; NVLink can be understood as the "highway" between GPUs—training models like ChatGPT requires hundreds or thousands of GPUs working simultaneously, and NVLink enables high-speed communication among them, like many computers becoming one supercomputer; InfiniBand is like the "high-speed rail network" of the entire AI data center, connecting servers and GPUs to enable rapid data transfer, otherwise even the most powerful GPUs waste performance waiting for data; AI Factory is a new concept emphasized by Jensen Huang in recent years—whereas enterprises used to buy servers, in the future they buy an "intelligent factory" that continuously produces AI models and services. From GPUs, servers, network devices to development software and model deployment, NVIDIA offers an almost complete solution. Enterprises no longer buy just a chip but an entire AI ecosystem. As more enterprises build AI data centers according to your standards, your competition is no longer peers but defining the entire industry’s future development. Image My understanding is: The Power Law is never about predicting stock price rises or falls; what it truly teaches me is how to identify whether a company has the ability to become the winner that "takes most of the industry’s value." Financial reports determine a company’s present; business model, ecosystem, and competitive advantages determine its future. In short: GPUs are just the products NVIDIA sells, while CUDA, DGX, NVLink, InfiniBand, and AI Factory are the ecosystem that truly builds its moat.