The Next Step for Large Models: Moving Toward an Open Computing System Like Bitcoin The development of large models is currently facing two fundamental bottlenecks: first, the lack of a closed-loop commercial structure, and second, the centralization of training paradigms. Today’s AI systems are essentially centralized production systems dominated by platforms—data is provided by users, but the value is captured by the platforms, leading to a severe mismatch between data producers and value beneficiaries. Meanwhile, model training relies on closed datasets and offline processes, artificially locking the path of intelligent evolution and making it difficult to achieve truly open growth. One potential breakthrough is to reconstruct large models into an open computing system similar to Bitcoin. The core idea is to introduce a "PoW-like (Proof of Work) mechanism," transforming data and the training process into verifiable and incentivized computational contributions. In this structure, actions like data provision and model optimization can be seen as "mining," where participants earn rewards by contributing high-quality data or computational power, thereby achieving a value loop where "users are producers, consumers, and beneficiaries." A deeper transformation lies in the reconstruction of the training mechanism itself. Traditional models rely on static data and periodic training, whereas PoW-enabled large models evolve into continuously running dynamic systems: data flows in real-time, models are updated continuously, and the entire network approaches optimal states through competition and collaboration. This process is akin to the transaction and block generation mechanisms in blockchain, giving the evolution of model parameters a sense of temporality and irreversibility. When "data" itself is PoW-enabled, AI systems will no longer be simple function approximators but will become complex, self-organizing evolutionary systems. The growth of intelligence will no longer be entirely predictable but will emerge through nonlinear processes, constantly generating new structures and capabilities. This marks a shift for AI from "closed Turing machine computation" to "open evolutionary computation networks." From this perspective, the future of artificial intelligence will no longer just be a trained model but more like a continuously running computational ecosystem that generates new information—a truly living intelligent network.