AI in Defense Weekly - 12 December 2025
GenAI.mil and China's H200s dominated the news (and the halls of the Pentagon and Congress).
Different Flags, Different Approaches to Bring AI to the Military
The Israel Defense Forces (IDF) has consolidated all its disparate AI units under a single command structure, called “Bina.” According to Cybernews, this is designed to streamline data flow, break down silos between units, and create a single authority for AI policy and deployment.
On the other hand, you have the UK, which (while centralizing strategy, governance, ethics, and data standards) has decentralized experimentation and delivery to the Services. For example, the Royal Navy just announced a new AI-powered subsurface sensing network to detect and track Russian submarines.
These different models offer different bets about where the value of AI lies, and the best ways for it to provide a military advantage. Do you centralize and standardize for better governance and economies of scale, at the risk of creating single points of failure and increasing organizational distance from users? Or, do you try and embed AI talent and capabilities across the force to promote broad-scale adoption at the risk of diluting capabilities?
What’s particularly compelling is whether the U.S. Department of War would contemplate something similar after launching GenAI.mil?
GenAI.mil debuts and succeeds NIPRGPT/CamoGPT
The U.S. Department of War has officially rolled out GenAI.mil, a new generative AI platform designed to bring commercial-grade artificial intelligence tools directly into the hands of the American military workforce — roughly three million service members, civilians, and contractors.
GenAI.mil provides secure access to frontier AI models, beginning with Google Cloud’s “Gemini for Government” — an enterprise-ready model certified to handle controlled unclassified information. It will initially support tasks such as deep research, document drafting, video/image analysis, and workflow automation.
Why it matters:
Scale and accessibility: For the first time, generative AI isn’t siloed in R&D labs but available across the entire DoW enterprise, from administrative work to support for analytical and operational decision-making.
Strategic push: The initiative reflects a top-level directive to build an “AI-first” workforce and maintain U.S. leadership in the global AI race. Pentagon leadership has emphasized that mastering these tools is critical to outpacing potential adversaries.
Future growth: Gemini is just the first capability — other commercial models and agentic tools are expected to join the platform over time, expanding both breadth and depth of use cases.
What to watch:
How GenAI.mil evolves beyond unclassified use into more sensitive workflows and analytical domains.
Adoption and training uptake across service branches as the DoD pushes toward embedding AI into everyday military “battle rhythm.”
What comes next?
This phase is the appetizer….a chat box isn’t a weapon system, and real combat power will come from “Digital NCOs” and orchestration layers that can turn the commander’s intent into action—and keep working even when the cloud goes dark.
China, Nvidia H200 Chips, and the AI Compute Competition
This week’s defining AI story isn’t about models — it’s about compute and control.
A Major Policy Shift on H200 Exports
The U.S. government, under President Trump, has authorized the export of Nvidia’s H200 AI chips to approved customers in China, reversing earlier export restrictions and imposing a 25 % fee on those sales. These chips are significantly more powerful than the previous China-compliant H20 variants, and while the most advanced Nvidia architectures (like Blackwell or Rubin) remain off-limits, the H200 still represents a substantial boost for AI workloads.
Strategic and Security Implications
Experts argue this export decision could narrow the AI hardware gap China faces and accelerate the development of its large-scale AI models and infrastructure. National security analysts that allowing H200 access undermines a core export-control strategy aimed at limiting China’s ability to train cutting-edge AI — a capability with clear military and dual-use implications.
In Washington, lawmakers from both parties have reacted:
Democratic leaders are calling for hearings with Nvidia’s CEO and administration officials to probe potential security risks, including how these chips might be used by China’s defense sector.
Republican critics argue the move weakens U.S. technological leverage even as geopolitical competition with China intensifies.
Real-World Activity in China
Long before the official export decision, H200 chips were already circulating in China via gray markets and opaque channels, and multiple Chinese universities and state-linked research institutes were using them to train models and build AI infrastructure. This underscores how difficult export controls are to enforce and how demand for top-tier AI compute remains robust.
At the same time, U.S. authorities have busted smuggling networks that tried to export large quantities of restricted Nvidia GPUs — including H100 and H200 units — to China and Hong Kong, highlighting ongoing challenges in controlling chip flows.
Bottom line
The H200 export decision is more than a trade tweak — it’s a geopolitical computing contest. As AI becomes central to economic and military power, who controls the silicon stack matters as much as who writes the algorithms.