科技巨头联合捍卫开放权重，Anthropic 为何站在联盟之外

The open-weight letter matters because companies with different business models coordinated around a shared policy boundary, not because they suddenly agreed on one technical philosophy.

Open-weight models are framed as infrastructure that makes advanced AI more accessible, adaptable, inspectable, and competitive across organizations.

NVIDIA benefits when more capable models can be run by more organizations, because an open ecosystem broadens demand for accelerators and infrastructure.

Meta can use open models to distribute innovation broadly while reducing the strategic power of laboratories that control access through closed APIs.

Anthropic's position is better understood as support for targeted controls on frontier risks and illicit distillation rather than an explicit blanket ban on all open-weight models.

A policy against model distillation requires evidence standards that can distinguish copying from ordinary training, fine-tuning, and mixtures of outputs from multiple systems.

Restricting Chinese access to chips may slow frontier training, but it also accelerates domestic substitution and can reduce long-term dependence on the American stack.

NVIDIA's investment in Safe Superintelligence couples strategic capital with access to next-generation compute, potentially multiplying the laboratory's resources without an immediate product business.

Circular financing can obscure risk, but the central question is whether underlying compute demand and deployable AI revenue justify the long-term commitments.

The Apple-Micron dispute shows how supply-chain lobbying can be presented as national security even when firms are also protecting cost, capacity, and competitive advantage.