[Editor's Pick] Who Bears the Cost and Risk of AI?

The financial and safety landscape of artificial intelligence is evolving rapidly. Wall Street is pooling $60 billion to back Anthropic's access to GPU chips from Google, while a U.S. startup called Reflection AI has launched an open-weight AI model named Beam. This shift indicates that risks and costs once borne solely by AI developers are increasingly being shared among financial institutions, enterprise users, and public authorities.

Bank of America, Citigroup, and Morgan Stanley have joined forces to offer the $60 billion financing package, which includes $42 billion in senior debt and $18 billion in junior debt, with Blackstone committing $9 billion of the junior portion. This move reflects the capital-intensive nature of AI infrastructure, requiring bank loans, guarantees, leasing structures, and private capital.

Anthropic's co-founder and CEO, Dario Amodei, has cautioned that misjudging future AI demand could lead to "painful adjustments," meaning losses may extend beyond the model developers to include lenders, investors, and infrastructure providers. This introduces a new layer of risk for financial institutions and investors within the AI boom.

Meanwhile, Reflection AI unveiled Beam, its first open-weight model, on October 5. Beam utilizes a mixture-of-experts architecture with 501 billion total parameters, focusing on coding, reasoning, and agentic tasks. Beam's creators aim to bolster U.S. leadership in open intelligence but must undergo rigorous safety evaluations before gaining widespread acceptance.

As open-weight models become more prevalent, companies and developers gain greater control over deployment and operation. This contrasts with closed models, where developers hold full control. The competition is not only about which model performs best on benchmarks but also about which models become embedded in companies' systems.

New York City Council recently held a hearing on AI safety, with representatives from major AI companies and former AI researchers testifying about safety measures and accountability for potential failures. Experts like Jacob Coxon, a former researcher at OpenAI and Anthropic, have raised concerns about the potential for AI systems to surpass human control. Policymakers are now scrutinizing corporate safety practices and considering measures such as independent evaluations and whistleblower protections.

This development signifies a shift in who bears the costs and risks of AI. Financial institutions are financing AI compute infrastructure, companies assume greater responsibility for operating open-weight models, and public authorities are beginning to assess safety standards. As the industry develops, the benefits and liabilities of AI are distributed across more participants, including lenders, investors, enterprises, and policymakers.

The key metrics now extend beyond parameter counts and benchmark scores to include the distribution of returns, losses, and accountability in the AI ecosystem.