United States: $9 Billion for Nvidia Superchips

The rapid rise of artificial intelligence (AI) has triggered a genuine technological arms race, where even U.S. intelligence agencies struggle to keep pace with the frantic advancements. To remain competitive against AI giants like Anthropic and OpenAI, the United States has approved a secret funding request of $9 billion to acquire Nvidia superchips, specifically intended for the CIA and NSA. However, this amount still requires Congressional approval.

Cutting-Edge Superchips

Current AI models require considerable computing power, as well as specialized energy management and cooling. Nvidia's Grace Blackwell superchips are designed to meet these demands. They integrate a 20-core Arm CPU, manufactured by MediaTek and named Grace, paired with a GPU based on the Blackwell architecture. With 128 GB of LPDDR5x memory and 4 TB of storage via an NVMe M.2 SSD, these chips achieve a performance of one petaflop in AI while consuming only 140 watts.

Colossal Computing Power

These superchips are capable of handling AI models with up to 70 billion parameters, requiring about 140 GB of storage space. The starting price of a GB10 system at Best Buy is around $5,000. However, when these systems are deployed at scale, energy consumption increases significantly. For example, the GB300 NVL72 model can house up to 72 GPUs and 36 CPUs in a single liquid-cooled unit. A rack of this type can cost between $1.8 million and $4 million, and a data center can contain up to 100,000 of them.

A National Security Concern

AI is seen as a next-generation tool but also as a potential threat to national security, evolving faster than legislation. A draft order for AI companies to voluntarily submit their models for government testing was abandoned under pressure from industry leaders. The U.S. government aims not only to harness AI but also to examine publicly used models, which requires powerful hardware. The lack of investment in computing hardware and the current chip shortage necessitate massive spending to remain competitive.

A Strategic Investment

While $9 billion may seem like a huge sum, it remains modest compared to the investments of other players. For instance, Amazon Web Services plans to invest $50 billion to modernize its cloud services, widely used by intelligence agencies. Additionally, the next generation of chips, the Vera Rubin platform, is under development. These chips promise up to 10 times more performance per watt compared to the Grace Blackwell.

AI has become a modern arms race, and governments must invest heavily to avoid falling behind.