https://magnus919.com/categories/geopolitics/Recent content in Geopolitics on Notes from the Rabbit HoleHugoen© [Magnus Hedemark](https://github.com/magnus919)Wed, 05 Aug 2026 13:12:00 -0400https://magnus919.com/2026/08/washington-tried-to-starve-chinas-ai.-it-fed-it./Wed, 05 Aug 2026 13:12:00 -0400https://magnus919.com/2026/08/washington-tried-to-starve-chinas-ai.-it-fed-it./<p>On April 24, 2026, <a href="https://www.reuters.com/technology/chinas-deepseek-returns-with-new-model-year-after-viral-rise-2026-04-24/">DeepSeek launched the V4 preview</a>: the Pro model a 1.6-trillion-parameter mixture-of-experts with 49 billion active per pass, one million tokens of context. The world&rsquo;s most capable open model family went live on <a href="https://www.reuters.com/world/china/deepseek-v4-chinese-ai-model-adapted-huawei-chips-2026-04-24/">Huawei&rsquo;s Ascend chips and CANN software</a>, not CUDA. <a href="https://www.reuters.com/world/china/big-chinese-tech-firms-scramble-secure-huawei-ai-chips-after-deepseek-v4-launch-2026-04-29/">Alibaba Cloud and Tencent Cloud served it within hours</a>. Huawei&rsquo;s own chips had trained part of V4-Flash.</p> <p>Jensen Huang had warned on the Special Competitive Studies Project&rsquo;s <a href="https://scsp222.substack.com/p/episode-43-jensen-huang-on-generative"><em>Memos to the President</em> podcast</a>: &ldquo;The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.&rdquo;</p>