Glonce

NSA to spend billions testing AI as health-care billing costs surge

businessThe Glonce editorsThe Decoder🔥73Glonce rating73Impact70Novelty68Relevance95Surprise50Players75Credibility72
NSA to spend billions testing AI as health-care billing costs surge

The National Security Agency told Congress it will spend billions of dollars this year testing advanced AI models, according to The Washington Sun, which cited two sources familiar with the classified estimates. One source said computing power is the biggest single expense, with staffing costs added on top because the NSA has to compete with the pay packages AI labs offer. Because of this, lawmakers reportedly expect that full-scale AI oversight could eventually cost tens of billions of dollars a year, far above earlier estimates: the Congressional Budget Office had put the cost of a bipartisan bill to create a center for AI risks at about $20 million a year. Nat Purser of the AI Verification and Evaluation Research Institute has proposed a levy that would have AI developers help pay for independent testing, and according to The Washington Sun, Anthropic and OpenAI have signaled they might be willing to pay for more government oversight.

The pricing behind these costs helps explain why large customers pay so much for compute. According to SemiAnalysis, consumer subscriptions are heavily subsidized: a maxed-out $200 ChatGPT Pro subscription is worth up to $14,000 at API list prices, while a Claude Max subscription at the same price is worth up to roughly $8,000. Businesses increasingly pay by usage instead: OpenAI bills new enterprise contracts in tokens, and at Anthropic, 75 to 85 percent of revenue now comes from usage-based contracts, per SemiAnalysis. Agent workflows can burn through up to a thousand times more tokens than a regular chat. Providers are adjusting: The Information reports that OpenAI now lets some large customers pay only for completed tasks, and has introduced GPT-6 Sol and Luna, models priced at half the usual API rate, though frontier models remain expensive and are increasingly sold through closed cybersecurity programs. The business appears highly profitable: SemiAnalysis estimates gross margin on Anthropic's API business at more than 80 percent. The companies need the cash, though, the Financial Times reports OpenAI plans to spend about $856 billion on computing power through 2030, and Anthropic is planning its IPO for November at the latest.

In US health care, the cost problem runs differently: it isn't the price of AI itself but how hospitals and insurers use it to fight over billing. A Blue Cross Blue Shield Association report, cited by The New York Times, accuses hospitals of using AI-assisted coding to run up nearly $1 billion in extra costs over two years by coding cases as more severe and adding secondary diagnoses, collecting nearly $12,000 more per case on average with no apparent change in actual treatment. For hospitals it pays off: McLaren Health Care CFO Dave Mazurkiewicz says the system brings in an extra $1 million a month using coding software from SmarterDx, which takes a cut of the added revenue. Mazurkiewicz says insurers, in turn, use AI to comb through records for claims they can deny. Caroline Pearson of the Peterson Health Technology Institute says both sides now go several rounds per dispute simply because automated disputing is cheap. Writing in Health Affairs, former US health official David Brailer says he expects AI to raise overall health-care spending, just as electronic health records did before it.

Key facts

  • The NSA told Congress it will spend billions of dollars this year testing advanced AI models, mostly on computing power, according to The Washington Sun.
  • Lawmakers reportedly expect full-scale AI oversight could cost tens of billions of dollars a year, versus a Congressional Budget Office estimate of about $20 million a year for an earlier, narrower bill.
  • A $200 ChatGPT Pro subscription is worth up to $14,000 and a Claude Max subscription up to roughly $8,000 at API list prices; SemiAnalysis pegs Anthropic's API gross margin above 80 percent.
  • A Blue Cross Blue Shield Association report accuses hospitals of using AI-assisted coding to add nearly $1 billion in extra costs over two years, collecting almost $12,000 more per case on average.
  • McLaren Health Care brings in an extra $1 million a month using SmarterDx coding software, while insurers use AI to find claims to deny, per CFO Dave Mazurkiewicz.

Why it matters

The story lays out two separate ways AI is pushing costs up rather than down. Government oversight of frontier models is turning into a major line item, with estimates jumping from tens of millions to tens of billions of dollars a year, while in health care AI is not cutting costs but fueling a billing arms race between hospitals and insurers that raises the price of care itself.

Who it affects

US taxpayers footing the bill for NSA and broader AI oversight, Congress as it weighs how to fund regulation, AI providers like OpenAI and Anthropic navigating enterprise pricing and margins, and US patients and employers who ultimately absorb higher health-care costs driven by hospital and insurer billing disputes.

How to use it

Nat Purser's proposed levy would have AI developers help fund independent testing, and OpenAI and Anthropic have reportedly signaled openness to paying for more government oversight, an option Congress could weigh against the multi-billion-dollar oversight estimates cited here. On the enterprise side, the usage-based and per-task pricing shifts described (including OpenAI's cheaper GPT-6 Sol and Luna models) show large AI customers how providers are trying to bring compute costs down for high-volume workloads.

How solid is it

Key figures come from named sources: SemiAnalysis on pricing and margins, the Financial Times on OpenAI's spending plans and Anthropic's IPO timing, The Information on OpenAI's enterprise billing changes, and a Blue Cross Blue Shield Association report relayed by The New York Times on hospital coding costs. The NSA spending and oversight cost figures rest on unnamed sources described by The Washington Sun as familiar with classified estimates, so those numbers are secondhand and not independently confirmed here.

Risks and caveats

Several of the largest figures are estimates or ranges rather than confirmed totals: the NSA's spending is given only as 'billions of dollars,' and the tens-of-billions oversight projection is attributed to lawmakers' expectations rather than a finalized budget. The health-care cost figures come from a Blue Cross Blue Shield Association report, an industry group with its own financial interest in the billing dispute; similarly, SmarterDx's vendor cut of added hospital revenue is a conflict of interest worth noting when weighing McLaren's reported gains.

“because it's cheap”

— Caroline Pearson, Peterson Health Technology Institute