Editorial illustration for AI's High Cost: USD 856B in Compute, IPO Plans Strain Sectors
NSA Spending $856B on AI Compute Power Tests
AI's High Cost: USD 856B in Compute, IPO Plans Strain Sectors
The National Security Agency plans to spend billions of dollars this year testing advanced AI models, according to two sources familiar with classified budget estimates cited by The Washington Sun. Computing power eats most of that money. Staffing costs add to the total because the agency has to match salaries offered by private AI labs, a competition federal recruiters are largely losing.
Lawmakers now expect full oversight of AI systems across government could run tens of billions of dollars annually, a figure that dwarfs earlier projections. The Congressional Budget Office had priced a bipartisan bill creating an AI risk center at roughly $20 million a year, a number that looks almost quaint next to what agencies are actually budgeting.
The gap between sticker price and real cost shows up outside government too. Hospitals and insurers are folding AI into billing disputes, driving up the price of care itself rather than lowering it. And the way AI companies charge customers reveals just how expensive raw compute has become once the discounts disappear.
Consumer subscriptions look cheap because they're subsidized. Enterprise contracts, billed by usage, tell a different story.
The NSA is reportedly spending billions of dollars to test AI models, mostly on computing power. In US health care, AI drives up costs in a different way. It makes care itself more expensive because hospitals and insurers use it to fight over billing.
Why this matters
The $856 billion figure from OpenAI isn't a projection about better products, it's a bet that compute itself becomes the moat. Anthropic racing toward a November IPO in that same window tells us the industry needs public money to keep funding this arms race, not just venture capital. For founders and researchers, the NSA's billions in testing spend confirm what many suspected: government adoption of frontier models is already priced in at a scale that dwarfs most startup budgets.
But the health care example is the one worth watching closely. Hospitals and insurers aren't using AI to cut costs, they're using it to out-litigate each other on billing, and patients eat the difference. That's a preview of what happens when AI gets deployed inside systems built on adversarial incentives rather than efficiency.
The lesson for anyone building in this space: raw model capability doesn't guarantee lower costs downstream. Who controls the compute, and who the AI is actually working for, matters more than how advanced the model is.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source Box CEO Aaron Levie Says AI Will Drive "Biggest Cybersecurity Upgrade
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives BottleCap AI's Qwen3.8-27B Uses 37% Fewer Tokens With Minor Accuracy Trade-Off
From Archives Meta's AI Lead Cites OpenClaw as Muse Inspiration
Common Questions Answered
Why is the NSA spending billions of dollars on AI model testing?
The NSA is investing billions primarily in computing power to test advanced AI models, with additional costs driven by the need to match salaries offered by private AI labs to compete for talent. Federal recruiters are largely losing this competition for skilled AI researchers and engineers to private sector opportunities.
How does AI increase costs differently in the US healthcare sector compared to government spending?
While the NSA's AI costs are dominated by computing infrastructure, hospitals and insurers use AI to drive up care costs in a different way by leveraging it to fight over billing disputes. This represents a distinct mechanism where AI technology increases expenses through administrative and billing processes rather than computational infrastructure.
What does the $856 billion figure from OpenAI represent according to the article?
The $856 billion figure is not a projection about better AI products, but rather a strategic bet that computing power itself becomes the competitive advantage or 'moat' in the AI industry. This massive investment signals that the industry views raw computational capacity as the foundation for maintaining competitive dominance in frontier AI models.
Why is Anthropic's November IPO plan significant in the context of AI funding?
Anthropic's move toward a November IPO indicates that the AI industry requires public market funding to sustain the ongoing arms race in AI development, beyond what venture capital alone can provide. This reflects the enormous capital requirements needed to compete in frontier AI model development at the scale being pursued by major players.
What does the NSA's spending on AI model testing reveal about government adoption of frontier models?
The NSA's billions in testing spending confirms that government adoption of frontier AI models is already factored into industry pricing and investment strategies at a scale that significantly exceeds most startup budgets. This demonstrates that government demand for advanced AI capabilities is a major driver of the industry's massive capital requirements.
Further Reading
- OpenAI forecasts $278 billion cash burn by 2030 as AI spending soars - CoinCentral
- Anthropic wants AI to slow down. Its $517 billion spending plan says otherwise - Investing.com
- The trillion-dollar AI buildout is creating a cash crunch - CTech
- Goldman Sachs just answered the biggest question about AI - TheStreet
- Free isn't cheap: How open source AI drains compute budgets - TechTarget
