{"id":1678,"date":"2026-11-20T13:32:35","date_gmt":"2026-11-20T13:32:35","guid":{"rendered":"https:\/\/www.geo-strata.com\/cms\/huawei-advances-chinas-ai-training-capabilities-with-ascend-910c-chips-amid-us-export-controls\/"},"modified":"2026-11-20T13:33:39","modified_gmt":"2026-11-20T13:33:39","slug":"huawei-advances-chinas-ai-training-capabilities-with-ascend-910c-chips-amid-us-export-controls","status":"publish","type":"post","link":"https:\/\/www.geo-strata.com\/cms\/huawei-advances-chinas-ai-training-capabilities-with-ascend-910c-chips-amid-us-export-controls\/","title":{"rendered":"Huawei Advances China\u2019s AI Training Capabilities with Ascend 910C Chips Amid US Export Controls"},"content":{"rendered":"<p>\n<img decoding=\"async\" alt=\"Huawei Advances China\u2019s AI Training Capabilities with Ascend 910C Chips Amid US Export Controls\" src=\"https:\/\/www.geo-strata.com\/cms\/wp-content\/uploads\/2026\/11\/us-semiconductors-ai-huawei-advances-chinas-training-capabilities-7590.webp\" style=\"width:100%;height:auto;border-radius:8px;\"\/>\n<\/p>\n<h2><span class=\"gspe-dropcap\" data-gspe-initial-letter=\"1\" style=\"display:inline-block;position:relative;margin-right:0.08em;line-height:1;vertical-align:bottom;white-space:nowrap;\"><img decoding=\"async\" alt=\"\" aria-hidden=\"true\" src=\"https:\/\/www.geo-strata.com\/cms\/font\/webp\/elegant-flourish-letter-c.webp\" style=\"display:inline-block;height:90px;width:auto;vertical-align:bottom;object-fit:contain;margin:0;border:0;\"\/><span aria-hidden=\"true\" style=\"position:absolute;width:1px;height:1px;overflow:hidden;clip:rect(0,0,0,0);clip-path:inset(50%);white-space:nowrap;border:0;padding:0;margin:-1px;\">C<\/span><\/span>ontext of China\u2019s AI Semiconductor Challenges<\/h2>\n<p>China\u2019s semiconductor industry has made notable progress in supporting AI inference tasks, which involve running pre-trained models to generate outputs in response to user queries. However, the more computationally intensive process of AI training\u2014where models are built or refined\u2014has remained a significant hurdle. This gap is particularly consequential given the tightening US export restrictions on advanced chips, which have historically limited China\u2019s access to the high-performance hardware necessary for large-scale AI training.<\/p>\n<h2>Breakthrough with Huawei\u2019s Ascend 910C Chips<\/h2>\n<p>A research collaboration involving Huawei Technologies, the Shenzhen Loop Area Institute, Harbin Institute of Technology\u2019s Shenzhen campus, and the Shenzhen Research Institute of Big Data has demonstrated a major leap forward. The team successfully completed post-training of the DeepSeek-V4-Pro model, a large-scale AI architecture comprising 1.6 trillion parameters, using a cluster powered by at least 1,000 Huawei Ascend 910C chips. This marks the first time China\u2019s domestic semiconductor technology has achieved full-parameter post-training on such a complex model, a process that updates the entire model architecture without simplifications.<\/p>\n<h2>Technical Significance of Full-Parameter Post-Training<\/h2>\n<p>Post-training, distinct from initial pre-training, involves refining a model\u2019s capabilities by teaching it to follow human instructions, safety protocols, and task-specific guidelines. Achieving full-parameter post-training indicates that the model\u2019s entire parameter set was actively adjusted, a process that demands significantly higher computational and communication resources than inference alone. Previously, domestic efforts were largely limited to inference tasks, which are comparatively straightforward. This advancement introduces complex computational pathways akin to multi-directional data flows rather than the one-way input-output mechanism typical of inference, thereby multiplying resource demands.<\/p>\n<h2>Implications for China\u2019s AI Self-Reliance<\/h2>\n<p>The success of this project underscores a critical step toward reducing China\u2019s dependence on foreign semiconductor technology for AI development. Historically, training large AI models has relied on hardware from US companies such as Nvidia and AMD, whose advanced chips are subject to export controls. Although Nvidia\u2019s H200 chips have some export clearance, they remain unavailable for sale in China. The ability to conduct large-scale post-training domestically addresses a strategic vulnerability in China\u2019s AI supply chain and supports broader goals of technological independence.<\/p>\n<h2>Broader Industry Developments and Strategic Context<\/h2>\n<p>Other Chinese companies have also reported progress in AI training on native hardware. Baidu recently completed training a major version of its Ernie 5.1 model on its Kunlunxin chip cluster, while Meituan has introduced a trillion-parameter AI model reportedly trained entirely on domestic chips. Huawei, in addition to chip development, is advancing AI\u2019s agentic capabilities\u2014enabling AI to perform tasks beyond conversational responses. Its cloud division recently launched an &#8220;Agentic Infra&#8221; platform designed to optimize compute resource allocation for both inference and training, enhancing utilization efficiency by over 30%.<\/p>\n<h2>Conclusion: Strategic Leverage Through Domestic Innovation<\/h2>\n<p>Huawei\u2019s demonstration of stable, uninterrupted training of a trillion-parameter model on Ascend 910C hardware represents a pivotal advancement for China\u2019s AI and semiconductor sectors. This progress reflects a strategic response to US export restrictions, accelerating domestic innovation and infrastructure development. As China moves toward greater self-reliance in AI model training, it enhances its capacity to compete in the global AI landscape, potentially reshaping the technological balance amid ongoing geopolitical competition.<\/p>\n<h2>Related Analysis<\/h2>\n<ul>\n<li>\n            <a href=\"https:\/\/www.geo-strata.com\/cms\/us-semiconductor-export-controls-and-the-xi-trump-summit-a-strategic-pause-amid-rising-chinese-self-reliance\/\"><br \/>\n                US Semiconductor Export Controls and the Xi-Trump Summit: A Strategic Pause Amid Rising Chinese Self-Reliance<br \/>\n            <\/a>\n        <\/li>\n<\/ul>\n<p class=\"gspe-source\">\n<small><br \/>\n<strong>Source:<\/strong><br \/>\n<a\n    href=\"https:\/\/www.scmp.com\/tech\/article\/3356117\/huawei-chips-refine-deepseek-model-major-leap-chinas-ai-self-reliance\"\n    target=\"_blank\"\n    rel=\"noopener noreferrer\"\n>South China Morning Post<\/a><br \/>\n<\/small>\n<\/p>\n<p style=\"font-size:18px; font-weight:600; color:#1e3a8a; margin:25px 0;\">\r\n    Can you afford not to enter the Chinese market? Talk to us, we\u2019ll help you succeed in China.\r\n<\/p>\r\n\r\n<a href=\"https:\/\/www.geo-strata.com\/cms\/contact-us\/\" \r\n   style=\"background-color:#0073aa; color:white; padding:14px 28px; \r\n          text-decoration:none; border-radius:6px; font-size:17px; \r\n          font-weight:600; display:inline-block;\">\r\n    Talk to us \u2192\r\n<\/a>\n<!-- Calendly link widget begin -->\r\n<link href=\"https:\/\/assets.calendly.com\/assets\/external\/widget.css\" rel=\"stylesheet\">\r\n<script src=\"https:\/\/assets.calendly.com\/assets\/external\/widget.js\" type=\"text\/javascript\" async><\/script>\r\n<p style=\"font-size:18px; font-weight:600; color:#1e3a8a; margin:25px 0;\">\r\n<a href=\"\" onclick=\"Calendly.initPopupWidget({url: 'https:\/\/calendly.com\/geo-stata\/china-market-entry-strategy-session'});return false;\">Book a China Market Entry Strategy Sessionme<\/a><\/p>\r\n<!-- Calendly link widget end -->\n","protected":false},"excerpt":{"rendered":"<p>Huawei\u2019s successful post-training of the DeepSeek-V4-Pro model on domestic Ascend 910C chips marks a significant milestone in China\u2019s efforts to achieve AI self-reliance, overcoming longstanding challenges in complex AI model training under US sanctions.<\/p>\n","protected":false},"author":1,"featured_media":1677,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"gspe_hreflang":"{\"en-US\":\"https:\/\/www.geo-strata.com\/cms\/huawei-advances-chinas-ai-training-capabilities-with-ascend-910c-chips-amid-us-export-controls\/\",\"de-DE\":\"https:\/\/www.geo-strata.com\/cms\/de\/huawei-und-chinas-aufstieg-zur-ki-selbststandigkeit-fortschritte-bei-der-modell-feinabstimmung-mit-heimischer-chiptechnologie\/\",\"pt-BR\":\"https:\/\/www.geo-strata.com\/cms\/br\/avanco-da-huawei-com-chips-ascend-910c-impulsiona-autonomia-chinesa-em-treinamento-de-ia\/\",\"zh-CN\":\"https:\/\/www.geo-strata.com\/cms\/cn\/%e5%8d%8e%e4%b8%ba%e8%8a%af%e7%89%87%e5%8a%a9%e5%8a%9bdeepseek%e6%a8%a1%e5%9e%8b%e5%90%8e%e8%ae%ad%e7%bb%83-%e5%ae%9e%e7%8e%b0%e4%b8%ad%e5%9b%bdai%e8%87%aa%e4%b8%bb%e5%8f%91%e5%b1%95%e7%9a%84%e5%85%b3\/\",\"ru\":\"https:\/\/www.geo-strata.com\/cms\/ru\/huawei-%d0%b4%d0%b5%d0%bc%d0%be%d0%bd%d1%81%d1%82%d1%80%d0%b8%d1%80%d1%83%d0%b5%d1%82-%d0%bf%d1%80%d0%be%d1%80%d1%8b%d0%b2-%d0%b2-%d0%be%d0%b1%d1%83%d1%87%d0%b5%d0%bd%d0%b8%d0%b8-%d0%b8%d0%b8-%d0%bc\/\",\"hi-IN\":\"https:\/\/www.geo-strata.com\/cms\/hi\/%e0%a4%b9%e0%a5%81%e0%a4%b5%e0%a4%be%e0%a4%b5%e0%a5%87-%e0%a4%95%e0%a5%87-%e0%a4%9a%e0%a4%bf%e0%a4%aa%e0%a5%8d%e0%a4%b8-%e0%a4%a8%e0%a5%87-%e0%a4%9a%e0%a5%80%e0%a4%a8-%e0%a4%95%e0%a5%80\/\",\"x-default\":\"https:\/\/www.geo-strata.com\/cms\/huawei-advances-chinas-ai-training-capabilities-with-ascend-910c-chips-amid-us-export-controls\/\"}","footnotes":"","_links_to":"","_links_to_target":""},"categories":[80],"tags":[807,808,22,371,276,89,337],"class_list":["post-1678","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technology-ai","tag-ai-training","tag-ascend-910c","tag-china","tag-deepseek","tag-huawei","tag-semiconductors","tag-us-export-controls"],"_links":{"self":[{"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/posts\/1678","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/comments?post=1678"}],"version-history":[{"count":1,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/posts\/1678\/revisions"}],"predecessor-version":[{"id":1679,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/posts\/1678\/revisions\/1679"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/media\/1677"}],"wp:attachment":[{"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/media?parent=1678"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/categories?post=1678"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.geo-strata.com\/cms\/wp-json\/wp\/v2\/tags?post=1678"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}