{"slug": "chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips", "title": "China’s Meituan says its new AI model was trained on domestic chips", "summary": "Meituan launched LongCat-2.0, a 1.6-trillion-parameter AI model trained entirely on domestically developed chips, claiming it as the first of its scale to achieve end-to-end training on Chinese silicon. The move directly challenges US export controls on advanced semiconductors and signals China's progress in building frontier AI models without Nvidia hardware.", "body_md": "*LongCat-2.0, a 1.6-trillion-parameter model, is the first of its size to be trained end-to-end on home-grown silicon, the company says, in a pointed answer to US export controls.*\n\nThe most striking claim about Meituan’s new artificial intelligence model is not how large it is, though it is large, but what it ran on.\n\nThe Chinese delivery-and-services giant launched LongCat-2.0 on Tuesday and said it was the first model of its scale to be trained entirely on domestically developed chips, a milestone aimed squarely at the export controls Washington has used to keep its best silicon out of Chinese hands.\n\nThe specifications are serious. LongCat-2.0 carries 1.6 trillion parameters and a context window of one million tokens, and Meituan says its performance is comparable to Google’s Gemini 3.1 Pro, released in February.\n\nThe company describes it as *“the industry’s first trillion-parameter model to complete end-to-end training and inference on a 50,000-chip domestic compute cluster.” *\n\nThe model has been open-sourced, putting the weights in the hands of anyone who wants to run or scrutinise them.\n\nThe crucial detail is the phrase “end-to-end.” Plenty of Chinese models run inference on domestic hardware, the comparatively light task of answering a query once a model is trained.\n\nPre-training is the heavy part, the computationally brutal process in which a model digests vast data sets to learn its basic patterns, and it is where the most advanced chips have mattered most.\n\nMeituan’s claim that LongCat-2.0 was both pre-trained and served on domestic silicon is what makes the announcement more than a marketing line.\n\nIf the claim holds, it speaks directly to the strategic question hanging over China’s AI sector: whether it can build frontier-scale models without Nvidia.\n\nWashington restricts exports of the most cutting-edge chips on national-security grounds, and Beijing has responded by pouring resources into a domestic alternative, accelerating efforts to design and manufacture silicon that can carry the load the American hardware was carrying.\n\nThat effort has produced a steady run of milestones. China recently claimed the [supercomputing crown without US chips](https://thenextweb.com/news/china-lineshine-supercomputer-top500-no-us-chips), and a clutch of domestic challengers has emerged to contest Nvidia’s dominance, with Alibaba’s T-Head unit pushing its [Zhenwu M890 GPU](https://thenextweb.com/news/alibaba-zhenwu-m890-t-head-china-ai-chip-nvidia) as a home-grown accelerator.\n\nLongCat-2.0 is the software counterpart to that hardware push, a large model designed to prove the domestic stack works at scale.\n\nMeituan is an unlikely flag-bearer for the cause, which is itself part of the story. Better known for food delivery than frontier AI, the company is one of several Chinese internet giants that have moved aggressively into model development, treating it as core infrastructure rather than a side project.\n\nOpen-sourcing a 1.6-trillion-parameter model is also a competitive move, seeding adoption among developers and signalling confidence that the underlying chips can keep up.\n\nFor a company that runs one of the world’s largest on-demand logistics operations, the appeal of cheaper, domestically secured AI is concrete rather than abstract: routing, demand forecasting, and customer service all run on compute, and a model trained on home-grown silicon insulates that compute from the next turn of the export-control screw.\n\nThe independent verification will come from the open-source community, which can now run LongCat-2.0 against the benchmarks Meituan cites and test whether it genuinely matches a model like Gemini 3.1 Pro.\n\nThe training-hardware claim is harder for outsiders to confirm directly, since it rests on Meituan’s account of its own infrastructure, and that caveat is worth holding in mind alongside the company’s confidence.\n\nWhat is not in doubt is the direction of travel. The race for AI dominance between China and the United States has become, at its base, a race over chips, and each model trained without American hardware narrows the gap the export controls were meant to widen.\n\nMeituan’s announcement is one more data point in a contest that Washington designed its restrictions to win, and that Beijing is determined to prove it can run on its own terms.\n\n## Get the TNW newsletter\n\nGet the most important tech news in your inbox each week.", "url": "https://wpnews.pro/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips", "canonical_source": "https://thenextweb.com/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips", "published_at": "2026-06-30 12:42:44+00:00", "updated_at": "2026-06-30 13:33:51.360205+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-chips", "ai-policy", "ai-research"], "entities": ["Meituan", "LongCat-2.0", "Google", "Gemini 3.1 Pro", "Nvidia", "Alibaba", "T-Head", "Zhenwu M890 GPU"], "alternates": {"html": "https://wpnews.pro/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips", "markdown": "https://wpnews.pro/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips.md", "text": "https://wpnews.pro/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips.txt", "jsonld": "https://wpnews.pro/news/chinas-meituan-says-its-new-ai-model-was-trained-on-domestic-chips.jsonld"}}