The article explores the erosion of traditional moats in the AI industry, particularly as software development becomes cheaper and AI models more accessible. Companies like OpenAI and Anthropic once seemed to be securing their dominance, but the emergence of competitors like DeepSeek, which created a comparable AI model for under $6 million, challenges the notion of a durable competitive advantage. The piece highlights how the complexity and investment once required to build software are diminishing, making it easier for new players to enter the field. Projects like uv and Ruff demonstrate significant performance improvements through modern programming languages, aided by AI coding tools. Economic value is shifting away from proprietary software and data. The article cites Ilya Sutskever, who claims that the era of unique training data may be over, with synthetic data proving almost as effective and easier to generate. In this context, moats that remain relevant include physical infrastructure like compute resources, human relationships, and capital reserves. Nvidia's dominance, bolstered by surging demand for AI chips, exemplifies how compute serves as a near-term moat. The article notes the increasing scarcity of resources like water for cooling data centers, which adds another layer of geographic advantage for companies that secured access early. On-device AI is emerging as a counter-narrative to cloud-based models, with companies like Apple leading the charge. Their focus on privacy and efficiency through on-device processing shows that local models can handle many tasks previously reliant on cloud computing. Despite this, the article suggests that pure software companies, particularly SaaS and AI model creators, may struggle due to their weaker moats. As open-source models catch up rapidly, first-mover advantages shrink. Ultimately, the strongest moats will likely belong to companies that leverage energy and logistics effectively, along with those that build strong relationships and networks that are hard to replicate.