US firms shift to open-weight AI models as budget strain mounts

The shift matters less as an ideological win for open source than as a cost-allocation decision: firms are splitting workloads, keeping frontier models for hard tasks and moving routine inference to cheaper open weights. That pressures the economics of closed-model providers whose revenue assumes volume across all tiers, while benefiting operators able to arbitrage between them. Whether the migration extends beyond routine tasks, or stalls once cost pressure eases, is the open question.