White House Quietly Bets $9 Billion on AI Chips as Spy Agencies Lean on Blacklisted Anthropic

The White House has greenlit a secret $9 billion request. Spy agencies get the advanced semiconductors they desperately need. Yet the move exposes a glaring weakness in American intelligence infrastructure. Shortages of top-tier chips have left the CIA and NSA unable to run the latest AI models on classified networks. Officials fear falling behind adversaries who race to weaponize similar technology.

This funding push, first reported by The New York Times, targets specialized data centers built around Nvidia's Grace Blackwell superchips. Those systems demand massive electrical capacity and liquid cooling. Construction won't happen overnight. In the meantime the government has turned to an unlikely partner. Susie Wiles, the White House chief of staff, personally authorized the National Security Agency to keep using an advanced model from Anthropic. The Pentagon had labeled the company a supply chain threat months earlier.

Anthropic and the government now finalize a classified contract. It lets the NSA maintain access to the firm's products. The model in question, known as Mythos, runs more efficiently on new chips. But it also operates on older generations. That flexibility makes it viable right now. "The compromise emerged alongside a secret $9 billion emergency funding request approved by the White House," noted The Next Web in its coverage Saturday.

Earlier this year the Defense Department demanded authority to use Anthropic's technology for any lawful purpose. That stance triggered a sharp dispute. The new contract drops that language entirely. Instead it includes a specific prohibition. The model cannot process Americans' data. White House officials hope this agreement becomes a template for deals with other AI providers. OpenAI's existing Pentagon contract does not yet cover the NSA. A separate arrangement will be needed there too.

The chip shortage runs deeper than one company. Frontier AI models consume far more processing power than anticipated. Classified systems simply cannot handle them at scale. Intelligence analysts rely on AI to sift enormous data volumes. They flag anomalies and potential threats. Without sufficient hardware those capabilities stay limited. Even larger sums will likely be required in coming years, experts told The New York Times. The administration has already reprogrammed $800 million from other accounts to buy computing capacity faster while Congress reviews the full request.

Anthropic itself has seen explosive growth. Its annualized revenue climbed from $9 billion at the end of 2025 to $30 billion by early April 2026. The company prepares for a potential initial public offering that could value it near $800 billion. Yet its relationship with the Pentagon remains tense. The blacklisting stemmed partly from corporate structure and foreign investment ties. It also reflected disagreements over how the technology could be deployed. Anthropic resisted broad "any lawful use" clauses that might open doors to mass surveillance or autonomous weapons.

Recent reporting adds context to the friction. In April Axios revealed the NSA had begun using Anthropic's Mythos Preview despite the Pentagon's objections. Cybersecurity demands apparently outweighed those concerns at the time. The Decoder reported Sunday that the finalized contract avoids the earlier sticking points. It positions the arrangement as a short-term bridge until new data centers come online.

But questions linger. How long can the government depend on a single provider flagged as a risk? What happens when the $9 billion investment finally delivers modern infrastructure? At that point the NSA could run models from multiple labs without hardware constraints. Anthropic's current edge, its ability to function on legacy chips, might erode. Rivals such as OpenAI could then compete more directly inside classified environments.

The situation highlights broader tensions in U.S. AI policy. Private firms hold the most capable models. Government agencies need them for national security. Hardware scarcity removes negotiating power. China looms as the strategic competitor. Officials worry Beijing could seize an advantage in intelligence applications if American compute capacity lags. Export controls on advanced chips to China have helped preserve a U.S. lead so far. Still the domestic supply remains tight.

Anthropic has not commented publicly on the latest contract. Its CEO, Dario Amodei, has spoken often about the strategic importance of maintaining American dominance in AI. The firm's models have demonstrated striking abilities in vulnerability discovery. One project reportedly identified more than 10,000 critical flaws in a single month. Such tools carry dual-use risks. They can strengthen defenses. They can also be turned against adversaries with devastating effect.

Congress must still approve the $9 billion package. Lawmakers have shown bipartisan support for bolstering AI capabilities in defense and intelligence. Yet budget pressures and competing priorities could slow things down. In the interim the reliance on Anthropic continues. The company's model buys time. It also creates an awkward dependency that neither side wants to acknowledge too loudly.

Recent days have brought fresh coverage. The Times of India detailed how the shortage has paradoxically benefited Anthropic despite the blacklist. Discussions on X reflect growing awareness. Users note the shift from consumer AI races to outright national security contests. One post observed that "AI is no longer just consumer tech, it's national security." Another highlighted Anthropic's revenue trajectory as evidence the bet on its technology keeps paying off.

This episode reveals the messy reality of technological competition today. Hardware constraints dictate software choices. Policy disagreements bend under operational necessity. Billions flow into classified projects while public debate stays limited. The $9 billion represents more than funding. It signals recognition that the United States cannot afford to let its intelligence edge slip. Whether the investment closes the gap fast enough remains an open question. For now the agencies improvise. They lean on a blacklisted vendor whose model happens to fit the available silicon. Pragmatism has won the day. Strategic clarity may take longer to arrive.