Recently, DeepSeek stated that its models match or surpass its main American rivals at a fraction of the cost, including Meta’s own Llama models, challenging the prevailing belief that scaling AI requires vast computing power and investment.