AI Geopolitics News: China AI Intelligence on US Forces, DeepSeek on Huawei Chips, Pentagon AI — April 7, 2026

April 7th 2026

Chinese firms track U.S. carrier groups with AI, the intelligence community elevates AI to a top-tier threat, and DeepSeek prepares to break free from American silicon.

The line between AI competition and AI conflict blurred further this week. A Washington Post investigation revealed that private Chinese companies -- some with PLA ties -- are using AI to market real-time intelligence on U.S. military movements in Iran. Meanwhile, the ODNI's 2026 Annual Threat Assessment formally elevated AI from a background enabler to a cross-cutting national security threat on par with state adversaries. And in a move that could reshape the chip war, DeepSeek confirmed its upcoming V4 model will run on Huawei processors, not Nvidia -- a deliberate pivot toward hardware independence that U.S. export controls were designed to prevent.

Watch & Listen First

  • The AI Policy Podcast: Inside Project Maven and AI-Powered Warfare (CSIS, March 26) -- Gregory Allen interviews Katrina Manson on Maven's role from Ukraine to Iran.
  • How AI Could Upend Geopolitics -- Foreign Affairs Interview (Spotify) -- Ian Bremmer and Mustafa Suleyman on the power shift between states and AI companies.
  • Manifold Podcast: Geopolitics 2026 -- Crossover with Seeking Truth From Facts (Spotify) -- Deep dive on China AI, DeepSeek, tariffs, and U.S.-China-Europe triangulation.

    • Key Takeaways

  • AI is now an official top-tier U.S. national security threat. The 2026 Annual Threat Assessment treats AI not as a single capability but as a cross-cutting force amplifying every adversary -- from China's military modernization to cybercriminal operations.
  • China is watching the Iran war with AI-powered eyes. Private Chinese firms are openly marketing satellite-derived, AI-analyzed intelligence on U.S. carrier groups and base activity -- and some hold PLA supplier certifications.
  • DeepSeek's V4 model will run on Huawei chips, not Nvidia. Months of joint engineering with Huawei and Cambricon signal that U.S. chip controls may be accelerating Chinese self-sufficiency rather than preventing it.
  • The Pentagon is building a multi-vendor AI arsenal. xAI's Grok joins Google's Gemini on GenAI.mil, giving military and civilian staff access to competing frontier models at Impact Level 5.
  • Europe is making its sovereignty play concrete. The EURO-3C project -- 75 million euros, 70+ organizations, 13 countries -- aims to build federated cloud-AI infrastructure to reduce dependence on U.S. and Chinese providers.

    • The Big Story

    Chinese Firms Market AI-Powered Intelligence Exposing U.S. Forces in Iran | April 4 | The Washington Post

    → As U.S. military operations in Iran entered their fifth week, a Washington Post investigation found that private Chinese companies -- including MizarVision (which holds a National Military Standard Certification for PLA supply) and Jinghan Technology ("China's Palantir") -- are using AI to fuse satellite imagery, flight tracking, and shipping data into detailed assessments of U.S. deployments. The intelligence products catalogue base activity, track carrier movements, and identify specific aircraft and missile defense positions. Analysts warn this is not merely commercial opportunism: the data is being used to refine Chinese missile trajectories, radar capabilities, and electronic warfare strategies relevant to a future Taiwan scenario. This is the clearest example yet of how open-source AI intelligence has become a geopolitical weapon -- and a preview of how every future conflict will be watched, analyzed, and monetized in near real-time.

    Also This Week

    ODNI Elevates AI to Cross-Cutting National Security Threat | April 2 | The Defense Post → The 2026 Annual Threat Assessment gives AI far more prominence than previous editions, treating it as a force multiplier across every threat vector rather than a standalone risk. China is named the "most capable competitor" in AI-driven military and intelligence operations.

    War Department Adds xAI to GenAI.mil Platform | April 2 | U.S. Department of War → Grok-based models join Google Gemini on the Pentagon's AI platform, cleared for Controlled Unclassified Information at IL5. Users will also get real-time feeds from the X platform -- raising questions about information quality in military decision-making.

    DeepSeek V4 to Run on Huawei Chips, Breaking Nvidia Dependence | April 6 | Tech Startups → DeepSeek reportedly spent months with Huawei and Cambricon reimplementing core model architecture to run on non-Nvidia hardware. If V4 performs competitively, it undermines a central assumption of U.S. export control strategy.

    EU Council Moves to Streamline AI Act Ahead of August Deadline | March 13 | Council of the EU → With full AI Act enforcement set for August 2, the Council agreed on a position to delay certain high-risk obligations and relax some data-use restrictions for AI training -- balancing competitiveness against the regulation's original ambitions.

    Startup Debuts Agentic AI Assistant for Warfighters | April 3 | Defense One → Edgerunner AI's WarClaw represents a shift toward smaller, operator-trained military AI models rather than frontier LLMs -- built by veterans for actual combat workflows.

    Worth Reading

  • Eight Ways AI Will Shape Geopolitics in 2026 -- Atlantic Council's framework for the year ahead.
  • What the 2026 Intelligence Assessment Gets Right -- and Wrong -- FDD's critique of the ODNI report's AI framing.
  • Inside China's AI Machine: Models, Chips, and Strategy -- Comprehensive breakdown of China's full-stack AI independence push.

    • The old theory was that choking off chips would choke off competition. This week's evidence suggests the opposite: restrictions are forging a parallel AI ecosystem, and the intelligence it produces is already pointed back at us.