The National Security Agency told Congress it will spend billions on AI testing this year, according to two sources familiar with classified estimates. Computing power is the primary cost driver, with staffing expenses rising as the NSA competes with AI labs for talent. Lawmakers expect full-scale AI oversight could cost tens of billions annually, a figure far higher than earlier estimates for a bipartisan AI risks centre.

AI providers are capitalising on the demand from large customers, with usage-based pricing models becoming the norm. OpenAI and Anthropic now sell most of their revenue through such contracts, and some enterprise customers can pay only for completed tasks. Frontier models remain pricey, and providers are increasingly selling capabilities through closed cybersecurity programs. OpenAI plans to spend about $856 billion on computing power through 2030, and Anthropic is planning an IPO for November.

In healthcare, the cost of AI is driven not by the technology itself but by how it is used for billing. A report from the Blue Cross Blue Shield Association accuses hospitals of using AI-assisted coding to run up nearly $1 billion in extra costs over two years. Hospitals coded cases as more severe and added secondary diagnoses, collecting an average of nearly $12,000 more per case without any apparent change in treatment. The hospital system McLaren Health Care brings in an extra $1 million a month using software from SmarterDx, according to CFO Dave Mazurkiewicz.

Both hospitals and insurers are using AI to fight over claims, with disputes now going several rounds because it is cheap. Former US health official David Brailer expects AI to raise overall health care spending, just as electronic health records did before it. There is already debate over who should foot the bill for AI oversight, with proposals for a levy on developers to fund independent testing. Anthropic and OpenAI have signalled they might be willing to pay for more government oversight.