White House greenlights $9 billion for spy agencies to catch up on Artificial Intelligence
Artificial Intelligence is no longer a future project for American spy agencies — it sits at the heart of today’s operations.
But there is a serious problem holding everything back: the C.I.A. and the N.S.A. simply do not have enough chips to run the most advanced models available on the market.
It is not a lack of willpower or technology.
It is a lack of infrastructure — and that bottleneck is costing American national security dearly.
To fix it, the White House just approved a secret investment of $9 billion aimed at acquiring the cutting-edge chips the agencies need to finally put the real potential of AI into practice across classified systems. The information was revealed by current and former U.S. government officials in a report by The New York Times.
That is a lot of money, yes.
But when the technology race involves espionage, military intelligence, and cyber threats, the cost of falling behind could be even greater. 🔐
The problem nobody wanted to admit
For years, the official narrative was that American intelligence agencies were always one step ahead technologically. And in many ways, that is still true. But the rise of generative Artificial Intelligence and large language models — the now-famous LLMs — exposed a weakness that stayed hidden for too long: the agencies’ hardware infrastructure simply did not keep pace with innovation.
While companies like OpenAI, Google, and Anthropic build clusters with tens of thousands of high-performance chips, the C.I.A. and the N.S.A. operate in highly restricted environments where every piece of hardware must go through rigorous approval, auditing, and security certification processes before it ever reaches their facilities.
The newest AI models consume enormous amounts of computing power — more than many technology experts predicted even one or two years ago. That reality has fueled concerns at both the White House and Congress that the chip shortage is causing intelligence agencies to fall behind when it comes to testing and deploying these tools for top-secret espionage work.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The practical result is clear: C.I.A. and N.S.A. analysts do have access to AI tools, yes — but limited versions running on outdated infrastructure with capacity far below what the state of the art allows. In a field where every detail matters and where processing speed can mean the difference between anticipating or reacting to a threat, this technology gap is more than an inconvenience: it is a real risk to American national security.
As Vinh Nguyen, a former chief data scientist at the N.S.A. and a senior fellow on AI at the Council on Foreign Relations, put it well: Our intelligence community needs the frontier — the best AI chips, models, systems and talent — on a timeline that keeps pace with the threat.
What changes with $9 billion in chips
The investment approved by the White House is not a blank check for any technology vendor. The $9 billion request was designed in part to expand the availability of infrastructure capable of supporting NVIDIA’s Grace Blackwell superchip, which requires data centers that can supply enormous amounts of electrical power and specialized liquid cooling systems.
Congress still needs to approve the funding. However, the administration is already reallocating $800 million for a faster acquisition of computing capacity while the legislative process moves forward. And experts point out that even larger sums will likely be needed down the road.
In practical terms, what this infrastructure expansion is expected to achieve is that the C.I.A. and the N.S.A. will be able to run AI models in a far more robust way within their private, isolated networks — the so-called air-gapped environments, which are systems completely disconnected from the public internet for security reasons. In these environments, running an advanced language model or a pattern-recognition system on intelligence data requires that all computing capacity be available locally, with no option to tap into public clouds.
Currently, the agencies run their classified AI models primarily on Amazon Web Services cloud networks. Amazon announced last year a $50 billion effort to upgrade its cloud computing services for the government. Even so, even if the money were approved immediately, there would be a significant delay before the classified cloud networks operated by AWS and other providers could build data centers equipped with the Grace Blackwells. 🖥️
That is because the data centers hosting the classified cloud are physically separated from unclassified networks and follow much stricter security protocols. Companies cannot simply upgrade commercial or unclassified data centers to handle classified government work overnight.
Anthropic’s role and the contract with the N.S.A.
To work around the chip shortage while the long-term investments have yet to materialize, the White House found a pretty interesting interim solution. Susie Wiles, White House chief of staff, authorized the N.S.A. to continue using an advanced AI model created by Anthropic — even though the Pentagon had designated the company as a supply chain threat.
According to U.S. government officials, Anthropic and the government are finalizing a classified contract that will allow the N.S.A. to maintain access to the company’s products. Anthropic’s new model, known as Mythos, runs more efficiently on the latest chips but can also operate on a previous generation of chips — which is precisely what makes it viable for the agencies right now.
This arrangement did not come without controversy. Earlier this year, the Department of Defense demanded authority to employ Anthropic’s technology for any lawful purpose, which sparked a standoff between the two parties. The new contract does not include that language.
Additionally, the contract will include a specific clause to ensure the AI model is not used on data belonging to American citizens. White House officials want this contract to serve as a template for agreements with other technology companies.
Currently, OpenAI’s contract with the Pentagon does not include the N.S.A. So, beyond resolving the chip issue, OpenAI and the government need to reach a separate agreement so the spy agency can use technologies like ChatGPT. Intelligence officials expect the Anthropic contract to pave the way for a similar deal with OpenAI.
The technology race behind national security
It would be naive to think the United States is alone in this race. The national security of virtually every major world power now depends, to varying degrees, on the ability to process and interpret data with the help of Artificial Intelligence. China, Russia, Israel, and several European countries all have their own AI programs focused on intelligence and defense, and the level of investment in these programs has grown steadily over the past several years.
The central point here is that the competition for advanced chips is no longer just an economic or commercial issue. It has become a matter of geopolitical power. It is no coincidence that the United States has been implementing increasingly strict restrictions on the export of high-performance semiconductors to certain countries. NVIDIA’s Blackwell class of chips, launched last year, currently has its export banned to China. Controlling who has access to the most advanced chips is, in practice, controlling who can develop the most powerful AI applications.
The N.S.A., in particular, has a track record of operations that depend heavily on massive data processing — from analyzing communications metadata to monitoring networks and detecting cyber threats on a global scale. AI technology has already been helping military and spy agencies sift through enormous amounts of intelligence data, proving particularly valuable for tasks like finding intercepted communications that slipped past human analysts.
Large-scale language models are being integrated into systems like Maven, which helps the military choose targets on the battlefield — although the Pentagon has not explained precisely how Artificial Intelligence was used to find targets in Iran or other locations. 🛡️
The alternatives and their risks
Faced with the chip shortage across classified systems, the agencies have some alternatives, but none of them are ideal. One option is to work on certain problems using unclassified networks, which run on the same internet and the same commercial data centers that the general public uses. However, doing so could risk the exposure of classified information, according to former intelligence officials.
The scarcity of Anthropic’s new Mythos AI model, launched in April, added even more urgency to the picture. This model is considered so effective at finding and exploiting cybersecurity vulnerabilities that it was initially shared only with a limited number of government agencies, banks, and other companies in the United States and the United Kingdom. So far, it still has not been made available for broader use.
Meanwhile, Anthropic has also been dealing with problems on the flip side. The company has suffered service outages on its Claude chatbot and at times has been forced to throttle supply during peak hours — something like a utility company running rolling blackouts to manage electricity demand.
The White House stance and the political context
In a strongly worded statement, the White House declined to discuss the chip shortage or its efforts to address it. Steven Cheung, White House spokesperson, stated: Sensitive national security deliberations are conducted with the seriousness they require — not leaked to reporters and repackaged through selectively attributed and unverified claims designed to generate headlines rather than truth. The fact is, the United States is leading the world in technology and is well-prepared to handle a variety of issues that may arise.
Artificial Intelligence has also become an increasing focus of the Trump administration. Last Thursday, the White House abruptly canceled a signing ceremony for a new executive order on AI just hours before it was scheduled to begin. President Trump told reporters he did not like some aspects of the document.
The order sought to formalize a process for AI companies to share their models with the government — including national security agencies — before they are released to the public, in recognition of the potentially broad threat these models pose to cybersecurity.
Earlier this month, the Pentagon announced it had struck deals with some of the biggest companies in the tech industry to expand the military’s AI capabilities and increase the number of companies authorized to operate on classified networks. The companies agreed to allow the Pentagon to employ their technology for any lawful purpose — the standard that Anthropic initially resisted.
While Pentagon officials consider that standard important for giving the military maximum flexibility in using AI models, White House officials did not view it as necessary for intelligence agencies. The N.S.A. and the C.I.A. are prohibited from collecting intelligence within the United States and significantly restricted in how they collect information about Americans abroad. For that reason, the spy agencies felt comfortable accepting the limitations proposed by AI companies to prevent their technologies from being used for surveillance of American citizens.
What this move reveals about the future of government AI
When agencies like the C.I.A. and the N.S.A. secure nearly $9 billion in chips for Artificial Intelligence, the message goes far beyond the investment itself. This kind of decision signals a structural shift in how governments view AI — no longer as an auxiliary or experimental technology, but as critical infrastructure, just as essential as satellites, military bases, or encryption systems.
This shift in perspective is likely to spread quickly across other departments and agencies of the U.S. government, creating a cascading effect that could profoundly transform how the state operates in the coming years.
Another important point is what this move means for the semiconductor industry. A government contract of this magnitude, specifically targeting high-performance chips for AI, is exactly the kind of demand that sustains entire development and production cycles. NVIDIA typically launches a new chip every year with the goal of delivering stronger processing capacity for cutting-edge AI models. Companies like NVIDIA, Intel, and AMD — along with manufacturers like TSMC — have a direct interest in ensuring their technologies meet the standards demanded by this type of client.
And when the client is an intelligence agency with extremely specific security requirements, the level of customization and technical collaboration involved can result in innovations that eventually reach the commercial market as well.
Representatives from AWS, OpenAI, and Anthropic declined to comment due to the classified nature of the work. NVIDIA did not respond to requests for comment. The Office of the Director of National Intelligence, the C.I.A., and the N.S.A. also declined to comment.
At the end of the day, what we are seeing is the consolidation of a new paradigm: Artificial Intelligence as a pillar of national security. And for that pillar to hold, it needs robust, modern hardware available in sufficient quantity. The chips approved in this investment are not just pieces of silicon — they are the foundation on which the next generations of intelligence tools will be built. The C.I.A. and the N.S.A. know this. And now, with this billion-dollar infusion, they finally have the resources to act on what they have known was necessary for quite some time. 🚀
