Since the introduction of artificial intelligence (AI) in 2023, its functions have changed from a chatbot for writing to document analysis, to video, image generation, audio generation, programming and, most recently, the ability to find security vulnerabilities in existing systems. It has also opened the eyes of intelligence agencies to the potential of AI to be used for the purpose of collecting important secret information.
The New York Times reported that the White House has approved an investment of $9 billion for the Central Intelligence Agency (CIA) and the National Security Agency (NSA) to build AI data centers with their own state-of-the-art chips. This will allow the CIA and NSA to run the latest AI models for intelligence purposes. So far, they have not had sufficient hardware infrastructure to run modern generative AI models, causing intelligence operations to lag behind the computing capabilities of private sector AI such as OpenAI and Google. Although the investment has been approved by the White House, it still needs Congressional approval to proceed.
The tight security process has slowed the installation of new hardware, while AI models now require far more computing power than previously expected. The NSA, for example, uses Amazon Web Services for its current computing needs. However, AWS took a long time to upgrade to the new NVIDIA Grace Blackwell chips and data center isolation was necessary to meet the country's top secret needs.
With its own data centers, the CIA and NSA want to run AI models entirely in a closed network, without relying on commercial cloud systems. This $9 billion investment marks the fact that AI is now considered critical infrastructure, on par with satellites or national encryption systems. The government's demands will drive innovation in the semiconductor industry, including NVIDIA, Intel, AMD and TSMC, due to its very specific security requirements.
At the same time, the United States has imposed export restrictions on high-powered AI chips to countries considered risky such as Russia and China. As a result, China has invested heavily in developing local chips through Huawei, so NVIDIA's AI chip market share in those countries is now zero.
Google last week detected the use of AI to carry out cyber attacks. At the same time, Claude Mythos has found more than 10,000 security vulnerabilities in the systems currently in use. This is why Mythos is not given access to everyone. If AI with the same capabilities falls into the wrong hands, the impact on security is quite significant.
