Anthropic’s revenue has surged from about $9 billion to roughly $30 billion in annualized terms by early 2026, with whispers of an IPO that could value the company up to $800 billion, highlighting the tension between market valuation and national security risk.

Taken together, the episode underscores a broader US AI policy dilemma: controlling access to sensitive providers while frontier AI capabilities remain concentrated in a few private firms, all amid hardware shortages that limit government leverage over supply chains.

Congress is expected to formally vote on the emergency package, while the government pursues specialized Nvidia Grace-based infrastructure with extensive cooling and power needs to support these systems.

Frontier AI models demand far more processing power than current classified networks were designed to handle, a hardware constraint that hinders testing and deployment.

A global memory reallocation driven by AI demand is constraining the government’s ability to build classified AI infrastructure, illustrating a broader tech supply chain crisis.

Anthropic’s position is paradoxical: its products are deemed a national security risk by the Pentagon, yet the NSA depends on them due to hardware bottlenecks and the lack of safe, viable alternatives.

The Pentagon has formally blacklisted Anthropic as a national security supply chain risk, even as the NSA continues to rely on Anthropic’s AI because of a critical chip shortage needed to run frontier models on classified networks.

In response, the White House approved a $9 billion emergency funding package to help spy agencies secure high-end semiconductors and build specialized data centers for running advanced AI on top-secret infrastructure, with $800 million earmarked immediately for capacity purchases.

Anthropic’s Project Glasswing has revealed thousands of critical vulnerabilities in Claude Mythos, underscoring the company’s dual role as a valuable security tool and a potential supply chain risk.