為何更便宜的 AI 模型可能會讓前沿實驗室破產
企業 ai 正轉向更小、更便宜的模型，可能會將 openai 和 anthropic 80% 的工作負載削減。
the ai industry faces a potential economic reset as enterprises abandon the assumption that bigger models are strictly better. brian armstrong predicts 80% of workloads will shift to 99% cheaper models within 12-18 months, retaining frontier models only for high-iq tasks. legal ai startup harvey validated this by cutting costs 3x using fireworks ai's glm 5.1 alongside claude opus. if broadly adopted, this optimization could severely reduce inference revenue for big labs, challenging the financial viability of training massive frontier models.