AI Is an Arms Race, and the US Wants $9 Billion in Nvidia Superchips to Keep Up

Why It Matters

Securing high‑performance AI hardware is critical for U.S. intelligence to analyze advanced models and maintain national‑security parity with private sector AI leaders. The investment underscores AI’s emergence as a strategic technology rivaling traditional defense spending.

Key Takeaways

• US seeks $9 billion for Nvidia GB10 superchips to equip intelligence agencies
• GB10 chip delivers 1 petaflop FP4 performance at 140 watts power draw
• Rack systems cost $1.8‑$4 million, enabling 70‑billion‑parameter model fine‑tuning
• AWS plans $50 billion government cloud upgrade, dwarfing the $9 billion request

Pulse Analysis

The request for $9 billion reflects a broader shift in how governments view artificial intelligence—not merely as a research tool but as a cornerstone of national security. Intelligence agencies need to process massive language models in real time, a capability that only the latest generation of GPUs, like Nvidia's GB10, can provide. By integrating these chips into secure data centers, the CIA and NSA aim to run proprietary models such as Anthropic's Mythos, reducing reliance on commercial cloud providers and safeguarding sensitive data.

Beyond the immediate hardware purchase, the funding highlights a strategic gap that the United States is racing to close. Private firms such as OpenAI, Anthropic, and DeepSeek have poured billions into custom silicon, achieving performance per watt gains that outpace legacy systems. The GB10’s 1 petaflop at 140 watts is a stark contrast to consumer GPUs that consume ten times that power for far less AI throughput. This efficiency is crucial for deploying AI at scale in classified environments where power, cooling and physical footprint are tightly constrained.

The $9 billion figure, while sizable, is modest compared with corporate AI spend. Amazon Web Services alone is allocating $50 billion to modernize its government cloud, a move that will likely provide the underlying infrastructure for many federal AI workloads. As the next wave of chips—codenamed Vera Rubin—promises tenfold performance per watt, the current investment can be seen as a bridge to keep U.S. agencies operationally relevant while the next generation of silicon matures. The outcome will shape not only defense capabilities but also the regulatory landscape governing AI deployment in the public sector.