SAN FRANCISCO — Anthropic CEO Dario Amodei said the company does not advocate for a ban on open-weight artificial intelligence models, a clarification that separates Anthropic from critics who accused it of favoring restrictions on model access.
The remarks follow criticism directed at Anthropic for its absence from an industry letter urging broader access to model weights. The letter was signed by several AI leaders and researchers.
Open-weight models give developers full access to a model's underlying architecture and parameters, allowing them to inspect, modify and deploy the technology freely. Closed or proprietary models offer only API access, keeping the core technology private.
The economics explain Anthropic's strategic position more plainly than any policy statement. The company has burned through approximately $800 million in compute costs over the past 12 months training its Claude model family, according to people familiar with the financials. That kind of capital outlay demands a commercialization strategy built around proprietary access, not open distribution.
Anthropoc closed a $2 billion funding round led by Google, valuing the company at $18.4 billion and bringing its total funding to $7.3 billion. OpenAI raised $6.6 billion in October. Mistral, the leading European competitor, has raised $1.1 billion.
Training a frontier AI model now costs between $100 million and $500 million in compute alone. That barrier concentrates the race among a handful of companies globally and makes decisions about model openness directly tied to competitive moats and return on capital.
Proponents of open-weight models argue they accelerate research and broaden access to advanced AI capabilities, citing rapid development cycles in open-source software as a precedent. Companies building proprietary models, including Anthropic, argue that closed systems allow more careful oversight of potential misuse.
For cloud providers, the debate has direct revenue implications. Alphabet, whose stock rose 2.1 percent to $326.56, benefits from the intensive compute demands of proprietary model training and inference. Nvidia, the dominant GPU supplier, saw its stock drop 5.0 percent to $196.51; the dynamics of model openness influence how compute demand is distributed and monetized across the AI ecosystem.
