NSA AI Testing Runs Into the Billions as Health Insurers See $1B Cost Bump

Government and industry are discovering that deploying advanced AI carries price tags far beyond the models themselves, from NSA compute budgets to insurer billing systems.
The National Security Agency is already spending billions of dollars to test advanced AI models, with the bulk of that money going to computing power rather than software or staff, according to reporting from The Washington Sun. The figure puts the intelligence community's experimentation well above earlier public projections and signals that the cost of simply evaluating frontier systems has become a budget item in its own right. Lawmakers now expect full-scale AI oversight to run into the tens of billions of dollars annually, a jump from the Congressional Budget Office's earlier estimate of roughly $20 million.
That gap between $20 million and tens of billions is the story. The CBO number reflected a narrower conception of what oversight would require, while the current estimates fold in the hardware, electricity, and personnel needed to run and audit large models at government scale. Compute, not headcount, is where the money goes, and that changes how agencies plan their technology budgets.
The cost pressure is not confined to Washington. In US health care, AI-assisted billing codes pushed up costs by nearly $1 billion over two years, according to the same reporting. Insurers and providers adopted the tools to speed up claims processing, but the resulting coding changes translated into higher payouts, a reminder that automation can raise spending even when it reduces labor.
Taken together, the two data points describe a pattern that is becoming hard to ignore across the AI economy. Testing and running advanced models demands massive compute capacity, and the organizations adopting them are finding that the operational bill arrives faster than the efficiency gains. Hospitals, insurers, and spy agencies are all learning the same lesson from different directions.
What happens next depends on whether these costs are treated as one-time buildout expenses or permanent operating lines. If oversight of advanced AI really does cost tens of billions a year, Congress will have to decide whether to fund it, scale it back, or push the expense onto the agencies themselves. The NSA's spending suggests the answer is already being written in procurement documents rather than policy speeches.