The White House has approved a classified $9 billion budget request to purchase advanced computing chips for U.S. intelligence agencies, according to information provided by current and former government officials to The Information news website and The New York Times.
3 View gallery
President Trump shows the order he signed at last year’s presentation of the administration’s AI plan
(Photo: Reuters)
The dramatic move stems from the U.S. administration’s belated recognition of the technological leap of the past two years. Advanced artificial intelligence models now require enormous computing power on a scale that even the most optimistic experts did not foresee just a year or two ago.
The shortage of advanced chips places key intelligence agencies, including the NSA and CIA, at a clear disadvantage, particularly as they examine the integration of AI-based espionage and decoding tools into classified networks.
Cold War 2.0
To bridge the immediate gap until the full budget is approved by Congress, the administration has already begun releasing about $800 million in emergency funding from other budgets for the rapid purchase of computing power. Most of the major funding is intended for the construction of dedicated physical infrastructure capable of supporting Nvidia’s new Grace Blackwell superchip.
These are computing monsters that cannot operate in ordinary server farms. They require unprecedented energy supplies and highly complex liquid cooling systems, preventing a quick upgrade of existing networks.
The American attempt to control the global chip market is now creating a boomerang effect that is hitting its own intelligence agencies. While the U.S. imposes severe export restrictions on chips to prevent China from accessing cutting-edge technology, China is bypassing those restrictions through sophisticated smuggling networks, model training in third-party countries in Southeast Asia and the accelerated development of domestic architectures.
Chinese technology giant Huawei is already reporting a breakthrough in the production of Ascend chips based on a “Logic Folding” architecture, which bypasses the need for advanced Western lithography machines manufactured exclusively by Dutch company ASML, and serves as the foundation for more than 40% of the country’s data centers.
At the same time, Europe is also trying to extricate itself from total dependence on American tech giants through technological sovereignty strategies such as “Apply AI” programs and dedicated funding for independent defense computing, based on the understanding that anyone who does not possess independent hardware loses political and security autonomy.
AI red lines and ethics
The severe hardware shortage forced White House chief of staff Susie Wiles to sign an exceptional waiver allowing the NSA to continue using an advanced model developed by Anthropic, despite the Pentagon’s formal designation of the company as a threat to the national supply chain. Officially, this was due to the company’s ties and ownership structure, but in practice it stemmed from the company’s refusal to allow uncontrolled use of its technology.
The company’s new model, known as Mythos, is exceptionally effective at locating and searching for complex cybersecurity vulnerabilities. Its major advantage over competing models such as those from OpenAI or Google is its ability to run efficiently even on the previous generation of chips.
A classified contract signed between the sides removed an earlier Defense Department demand for authority over any “lawful use” of the product, a clause Anthropic strongly opposed, and added strict restrictions barring use of the model on information belonging to American citizens.
While the NSA and CIA are forced to rely on isolated commercial cloud networks such as Amazon Web Services, which recently upgraded its government cloud services with a $50 billion investment, defense models elsewhere in the world show there is another way.
In Israel, for example, the IDF has chosen a strategy of focused internal development. The establishment of the Alumot cyber defense brigade, alongside the operational activity of the Matzpen unit, shows that intelligence superiority does not rely solely on the purchase of cutting-edge commercial hardware worth billions of dollars, but on the ability to analyze huge quantities of battlefield data in real time.
A prominent example can also be seen in the Lohem system, which managed the planning of recent strategic strikes without total dependence on external civilian cloud solutions that could expose classified information.
On the other hand, the IDF faces a danger similar to the Pentagon’s — namely, the refusal of civilian companies to provide it with technological solutions because of one political climate or another. That is exactly what happened recently after the IDF was removed from Microsoft’s Azure systems over use that allegedly did not comply with the company’s terms of service.