Mixture of Experts(MoE)란 무엇인가 — DeepSeek이 왜 1.6조 파라미터인데 싸게 돌아가는지

The shift toward Mixture of Experts (MoE) signifies that the AI industry is prioritizing cost-efficiency over brute model size, allowing massive models like DeepSeek to deliver frontier-level performance at a fraction of the cost. While it significantly lowers inference overhead through sparse activation, the resulting high memory demand highlights a strategic trade-off between computational speed and hardware requirements in the era of multi-trillion parameter models.