DeepSeek didn’t really train its flagship model for $294,000

DeepSeek's AI model training costs were misreported as $294,000, but the actual figure is closer to $5.87 million. This discrepancy arose because the lower figure only represented the reinforcement learning phase, not the entire training process. The base DeepSeek V3 model required 2,048 H800 GPUs running for two months, totaling 2.79 million GPU hours. When factoring in hardware purchase costs and R&D expenses, the true investment likely exceeds $51 million, undermining DeepSeek's efficiency claims compared to Western models.