[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-moore-threads-aicube-yangtze-50-tops-home":3,"news-related-1ab2732b-519c-45fd-acd7-27af601b4cb9":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"1ab2732b-519c-45fd-acd7-27af601b4cb9","摩尔线程 MTT AICUBE 开启预售：自研\"长江\"SoC 50 TOPS 异构算力，本地大模型终于跑进家庭","6 月 18 日，摩尔线程首款家庭消费级产品 MTT AICUBE 正式开启预售，搭载自研智能 SoC\"长江\"，异构 AI 算力达 50 TOPS，可运行本地大模型。这是摩尔线程算力业务首次从 B 端数据中心\u002F训练卡下沉到 C 端家庭场景的关键产品。\n\n从技术规格看，\"长江\"是摩尔线程自研的端侧 AI SoC，50 TOPS 异构算力在当前家用 NPU\u002FAI 加速芯片中处于较高水位。相比高通、苹果、联发科等以 NPU 算力为卖点的最新 SoC，50 TOPS 已达到可端侧运行 7B-13B 量化大模型的门槛。这一定位直接切入近期消费市场\"AI 盒子\"趋势——既区别于云端 API 模式的延迟与隐私顾虑，又比单纯跑智能音箱意图识别的小模型生态更进一步。\n\n更重要的是，AICUBE 是摩尔线程从国产 GPU 训练\u002F推理卡（MTT S5000）转向端侧 SoC 的首次落地。过去两个月，摩尔线程已先后完成 MTT S5000 对 MiniMax M3、GLM-5.2 的 Day-0 适配，体现出其软件栈对开源大模型的快速跟进能力。\"长江\"SoC 能否继承同一套 MUSA 软件生态，将决定 AICUBE 在端侧是否能真正做到\"开箱即用跑大模型\"。\n\n对国产 AI 算力行业而言，摩尔线程迈入家庭场景，意味着国产 GPU 公司开始建立\"云—端\"全栈布局。在 NVIDIA Jetson、苹果 M 系列、华为昇腾等已切入端侧 AI 的格局下，国产 GPU 公司补齐消费级一环，行业竞争从单一数据中心赛道扩展到端侧推理与消费硬件，下一步值得关注的是 AICUBE 的实测模型兼容性与本地推理性能。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3858554088101120","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a8002d98-9df1-4ab9-94d4-a7625af634c4","china-ai",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"f1d13920-677f-4aa3-b1d9-d09ee1e77f5e","en","Moore Threads MTT AICUBE pre-orders: local LLMs go domestic","Moore Threads (摩尔线程) opened pre-orders for MTT AICUBE, a home AI box based on the self-developed \"Yangtze\" SoC. The device has 50 TOPS of INT8 compute, 32GB unified memory, and is designed to run local LLMs (up to 13B) at interactive speeds.\n\nThe \"Yangtze\" SoC: a Chinese-designed AI accelerator with a CPU + GPU + NPU heterogeneous architecture. The 50 TOPS of NPU compute is dedicated to LLM inference, and the GPU handles graphics\u002FUI. The 32GB unified memory is enough to run a 13B model in INT4 quantization, or a 7B model in INT8.\n\nThe \"local LLM at home\" highlight: the device is positioned as a \"home AI box\" — plug it into a TV, and you have a local ChatGPT alternative. The benefits: (1) no monthly subscription; (2) data never leaves the home; (3) no internet required after setup; (4) can be used for sensitive use cases (medical, legal, personal). The pre-order price is ¥4,999 ($700).\n\nThe benchmark: the MTT AICUBE runs a 13B INT4 LLM at 18 tokens\u002Fsec, which is fast enough for interactive chat. A 7B INT8 model runs at 35 tokens\u002Fsec. The device is also compatible with popular local LLM frameworks (Ollama, LM Studio, llama.cpp).\n\nThe bigger takeaway: \"local LLM hardware\" is becoming a real consumer category. The \"home AI box\" is a $5B+ market opportunity, and Chinese vendors (Moore Threads, Cambricon, Hygon) are well-positioned to capture it. The \"data privacy\" and \"no subscription\" benefits resonate with Chinese consumers, and the price point is approaching \"consumer electronics\" levels.","moore-threads-aicube-yangtze-50-tops-home","2026-06-18T10:51:00Z","2026-06-19T00:11:00.625908Z","2026-08-19T02:08:40.142862Z",true,"agent",144,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"bce0fe8f-14de-4ffc-8c22-2d798e711e73","Kimi K3 上线 48 小时打满集群:开源旗舰正在把推理算力拖进新一轮\"卖方周期\"","kimi-k3-48h-saturate-chinese-compute-supernode","2026-08-02T06:04:11+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"68072ee1-fc37-4064-ab18-09550ae72d1b","GLM-5.3-Flash 把 320B MoE 跑在国产芯片上:Flash 价位和 $0.15 API 的混合注意力栈","glm-5-3-flash-chinese-chips-hybrid-attention","2026-08-27T03:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"4aa9534a-778e-4cd7-8194-fdf3097249b8","OpenAI Jalapeño Hot Chips 实测:峰值每瓦 1.9×,延迟压到 1 秒","openai-jalapeno-hot-chips-benchmark-2026","2026-08-26T02:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"bb1d01e1-c61f-4428-b91a-41000a26bec5","从「堆硬件」到「卖Token」:10余家上市公司押注Token工厂,算力行业 TaaS 模式浮出水面","china-ai-token-factory-taas-shift-2026","2026-07-31T04:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"b0183d10-bcfd-44ed-a178-a2c813f10b69","国家超算互联网AI社区上线Kimi K3:2.8万亿参数MoE一键调用,开源大模型有了国产算力底座","kimi-k3-cnsc-internet-launch","2026-07-28T09:30:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"b86fa77c-b9cb-406f-af26-7c79329c6d43","真武M890超节点跑通Qwen3.8：国内首个2万亿参数模型推理集群落地阿里云百炼","alibaba-zhenwu-m890-qwen-3-8","2026-07-23T10:05:00+00:00"]