In the next decade, the most valuable data will not be on the Internet. In the industry, there are not many articles that truly influence an investment era. All I can think of is: Fat Public Chain, Thin App Public chain is a country Now, another article should be added to this list: Data at the Edge - Data at the Edge This is the latest report released by the top venture capital firm USV in the United States. It only talks about one thing, but it could reshape the investment logic for the next decade: The most important data in the Internet era comes from software and network; In the next stage, the most important data comes from the real world. USV was founded in 2003, with representative investments including Twitter, Tumblr, Etsy, Coinbase, Uniswap, and others. These projects may seem unrelated, but there is a similar structure behind them - network effects. 1/What is network effect? Simply put, it is a positive cycle. ▌ More users → More content/data → More valuable products → Attract more users → Stronger network effect This is the common moat of great companies in the Internet era. Facebook, X, Airbnb... are all like this. In a new report, USV pointed out that this data flywheel is about to undergo a major migration. 2/Internet vs the real world: two data flywheels The data flywheel in the Internet era: More users → More content or data → Products and networks become more valuable → Attract more users The data flywheel of the real world: Deploy more devices and robots → Generate more real-world data → Data improvement models → Models make devices more efficient and affordable → Deploy more devices and robots In the Internet era, enterprises accumulate data through user behavior; In the future, the most valuable data will come from sensors, robots, and every corner of the real world. 3/Robots: Dual Identity Under this new framework, robots play two roles: Firstly, it is an execution tool. Transport goods, inspect equipment, and clean the environment. Secondly, it is a mobile data collector. Every time a task is completed, a large amount of data is generated, which in turn feeds the model. Robots are not just labor, they are themselves data machines. 4/Dilemma: Did the chicken or the egg come first? It sounds great, but the robotics industry is entering a vicious cycle. To create a sufficiently good embodied robot, a large amount of real data is needed To obtain a large amount of real data, it is necessary to deploy sufficiently good robots to customers The industry needs disruptors. 5/Caspius: Taking a Different Path The @ caspius_ai I am following is trying a different path. It does not directly build robots, but rather builds data infrastructure. Specifically, it is to collect data from the first perspective of humans to establish a learning foundation for embodied AI. The path is as follows: Ordinary users wear devices → Record first person operation videos → Accumulate human behavior and interaction data → Train embodied AI models → Help robots learn visual and operational abilities → Deploy in real scenarios in the future → Generate real robot execution data Caspius solves the "data cold start" problem in the robotics industry. It first teaches the model how humans observe and operate the real world, and then helps the robot acquire initial abilities. This type of data used to be extremely difficult to obtain on a large scale. Now, with the decrease in hardware costs and technological advancements, there is finally an opportunity to collect data on a large scale through community collaboration. 6/Write at the end The Internet has developed all easily accessible data, and the next generation of opportunities may come from real world data that could not be observed, understood or automated in the past. Robots are at the center of this change. And the meaning of Caspius should also be understood in this context. Sometimes, the most difficult thing is not the flywheel itself, but the first force that makes the flywheel turn.