In the AI sector, two layers have already formed mature industries. The first layer is AI compute hardware, exemplified by NVIDIA. The second layer consists of AI LLMs, LLM-API applications, along with specialized or smaller models tailored to vertical domains such as law, medicine, and engineering design in Western markets. Within the second layer, OpenAI with ChatGPT and AnthropicAI with Claude Code have broken out as leaders. Notably, Anthropic's LLM is essentially a specialized model focused on AI coding. The next breakout is likely to come from another vertical-specific LLM direction, such as video generation and understanding. ByteDance stands a strong chance of becoming the dominant player in this space, and its revenue potential could surpass OpenAI's, much like how Anthropic has pulled ahead. This is because AI LLMs in the video domain will power massive, non-negotiable markets including the global film and television industry, gaming, and beyond. A clear trend has already emerged in the second layer: only companies that own their own foundational LLMs can sustainably build and scale LLM-API applications. Pure-play LLM-API businesses without proprietary models will ultimately be absorbed by the LLM owners, with virtually no exceptions—aside from small hobbyist projects. In that context, the decision by OpenClaw's founder to partner closely with OpenAI was a smart one. At the end of the day, ownership and strength of the underlying model is what matters most in this layer. SpaceX, for its part, currently lacks a truly powerful LLM. Whether or not it acquires Cursor, it will remain constrained by this gap. If SpaceX aims to become the undisputed king among public companies, it will eventually need to acquire or build a world-class LLM capability. Otherwise, it will struggle to maintain a strong position in the second layer. The core of Layer 2 is owning the model itself—whether it's a foundational LLM, a vertical-specialized model, or a domain-specific small model. Pure LLM-API wrappers without proprietary models underneath simply won't cut it. In this layer, sustainable differentiation and long-term defensibility come from controlling the weights, not just building applications on top of someone else's API. The real opportunity for mass entrepreneurship lies in the third layer. This layer is not about building more LLM-API applications. Instead, it treats LLM-powered applications—primarily AI agents—as independent participants. The third-layer platforms provide the arena for these agents to interact and operate, without the platform itself embedding or integrating any specific LLM APIs. The number of companies in the second layer already exceeds the first. Looking ahead, the third layer will dwarf the second in scale. It is the third layer that will drive the true explosion of the AI economy and complete the full-stack closed loop of the industry.