最新报道:Foundation models are increasingly viewed as interchangeable commodities, with AI startups focusing on fine-tuning and interface design rather than pre-training. The diminishing returns from hyperscaled models have shifted attention to post-training and reinforcement learning. This trend undermines the dominance of major AI labs like OpenAI and Anthropic, as open-source alternatives and application-layer competition reduce their pricing power. While foundation model companies retain advantages in brand and infrastructure, the lack of a durable technological moat suggests a future where they may become low-margin back-end suppliers rather than industry leaders.