“The 2023 DeepSeek model announcement was significant because it suggested that training a frontier LLM could be much cheaper than previously thought, potentially weakening the scale-based moat of major AI labs.”