[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-amd-acquires-world-labs-fei-fei-li":3,"topics-all":38,"news-related-e1374b1b-13f1-4e08-a58a-c45e4f6b92af":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"e1374b1b-13f1-4e08-a58a-c45e4f6b92af","AMD 82亿美元拿下 World Labs：李飞飞出任 AMD 首席科学家","AMD 用 82 亿美元全股票收购李飞飞联合创办的 World Labs。World Labs 2024 年成立,专注世界模型,首款产品 Marble 用于娱乐和机器人训练,累计融资超 12 亿美元。收购后李飞飞出任 AMD 首席科学家,直接向苏姿丰汇报。此举意在与 Nvidia 争夺 AI 芯片生态。","8 月底一笔 82 亿美元的全股票交易,把\"AI 教母\"李飞飞拉进了硅谷另一家芯片大厂的权力核心。AMD 正式宣布收购由李飞飞、Justin Johnson、Christoph Lassner 和 Ben Mildenhall 等人联合创办的 AI 初创 World Labs,创始人本人将出任 AMD 执行副总裁兼首席科学家,直接向董事长苏姿丰汇报。\n\n这不是一笔寻常的收购。World Labs 成立于 2024 年,至今没有真正意义上面向终端用户的产品,第一款成果 Marble 仍然定位于\"娱乐体验\"和\"机器人训练用模拟环境\"。公司累计融资超过 12 亿美元,AMD 自己也是这轮投资人之一。82 亿美元的成交价,大约是上一轮估值的近 7 倍。\n\n## 押注的为什么是 spatial intelligence\n\nWorld Labs 的方向被 AMD 官方公告称为\"空间智能(spatial intelligence)\"——能根据文本、图像和视频输入生成、重建和模拟 3D 交互场景的模型,顺便也做机器人学习用的仿真数据。这和主流大语言模型不同:LLM 主要消费文本,世界模型则把视觉、物理规律、空间结构作为一等公民。在李飞飞 2024 年的创始声明里,她反复强调\"真正的通用智能需要建立在物理世界的理解上\"。\n\n这个判断现在被特斯拉、Figure 等机器人公司反复放大——人形机器人和自动驾驶的真正瓶颈,不是模型不够大,而是\"现实世界里真正有用的训练数据\"严重不够。世界模型用合成数据补这一缺口,正在成为 AI 进入物理世界的\"前置条件\"。\n\nAMD 自己的说法更直白:理解 World Labs 的前沿工作负载,才能倒推出\"芯片该怎么造\"。换句话说,AMD 想要的不是一个独立的软件公司,而是一支能为下一代 AI 芯片定义需求的队伍。\n\n## 与 Nvidia 的对位\n\n这笔交易本质上是 AMD 对 Nvidia 的正面回应。Nvidia 已经在 Cosmos 系列里拿出了多个开放权重的世界模型,把\"模型+芯片+开发工具\"打包卖给做自动驾驶、机器人和工业数字孪生的客户。AMD 此前只能拿出 Instella-MoE 这类文本或视频模型,缺一条世界模型的产品线。\n\n收购 World Labs 后,AMD 第一次拿到了一个成建制的空间智能研究团队 + 一条产品线(Marble)+ 一位行业偶像级科学家。TechCrunch 评论说,这是 AMD \"多年来最重要的 AI 战略动作\",李飞飞的加入甚至被解读为 Lisa Su 潜在的\"接班人\"信号——AMD 在 AI 时代的科研领导力第一次被摆到台面上。\n\n## 一位科学家的两段路\n\n李飞飞本人的轨迹值得一提。她凭 ImageNet 让深度学习在 2012 年重新回到研究主流,也长期在斯坦福带博士生。2024 年她离开学界创办 World Labs 时,行业普遍认为这是\"学术明星冲下牌桌\"的冒险。两年的过渡之后,这家公司以 82 亿美元卖给了长期缺一个\"灵魂人物\"的芯片厂。\n\n她在宣布交易时写道:\"我们的技术突破已经被证明可行,现在要做的不是签约,是要扩大规模、扩大覆盖、贴近硬件。\"这一句话把\"为什么卖、卖给谁\"讲得相当透彻。\n\n## 行业意义\n\n这笔交易至少在三件事上重写了预期:\n\n第一,世界模型正在被\"招安\"为主流 AI 公司的核心资产,不再只是早期 demo。Cosmos 是 Nvidia 自己做的,Marble 是 AMD 收来的,Google DeepMind 早有 Genie 团队,加上 Figure、Tesla 押注的合成数据——世界模型已经从边缘研究跳进了产业核心。\n\n第二,芯片厂的竞争维度变了。不再只是拼制程、拼内存带宽、拼每瓦 token 数,还要拼\"能为什么样的模型提供算力\"。当世界模型决定自动驾驶和机器人的下半场,芯片设计就要从\"训练通用 LLM 的硬件\"转向\"为物理世界训练服务的硬件\"。\n\n第三,科学家的流动方向变了。当一位顶级研究者可以直接以\"首席科学家\"的身份进入一家千亿美元市值的硬件公司,意味着学界和产业在 AI 这个垂直上几乎已经合并——再过几年,实验室的边界会比过去十年薄很多。\n\n交易预计在年底前完成,需要通过监管审批。在那之前,AMD 的市值变动、世界模型 Marble 的产品节奏、以及李飞飞团队在 AMD 内部到底能拿到多少自主权,会决定这笔 82 亿美元的\"科学溢价\"是真金白银还是新一轮泡沫。","https:\u002F\u002Fir.amd.com\u002Fnews-events\u002Fpress-releases\u002Fdetail\u002F1299\u002Famd-to-acquire-world-labs-to-advance-the-future-of-ai-compute","09817576-1b8d-491e-b843-2913b7bcbe49",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":19,"name":20,"slug":20,"description":14,"color":14},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":22,"name":23,"slug":23,"description":14,"color":14},"8ddf2b28-0234-41a4-9862-3f0faef96472","market-analysis",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"eac7c97a-889c-4ed7-8d38-b3111a5e41dc","en","AMD Acquires World Labs for .2B; Fei-Fei Li Becomes AMD's Chief Scientist","AMD is acquiring World Labs, the AI startup co-founded by Fei-Fei Li, in an all-stock deal worth roughly .2 billion. World Labs, founded in 2024, builds spatial-intelligence models, with Marble as its first product aimed at entertainment and robot training. After the deal closes, Li joins AMD as executive vice president and chief scientist, reporting directly to CEO Lisa Su — a direct counter to Nvidia's Cosmos-led world-model stack.","A late-September all-stock deal worth $8.2 billion pulled \"the godmother of AI\" Fei-Fei Li into the inner circle of another Silicon Valley chip giant. AMD formally announced it is acquiring World Labs, an AI startup co-founded by Li, Justin Johnson, Christoph Lassner and Ben Mildenhall. After the deal closes, Li will join AMD as executive vice president and chief scientist, reporting directly to chair and CEO Lisa Su.\n\nThis is not an ordinary acquisition. World Labs was founded in 2024 and still has no real end-user offering — its first product, Marble, is positioned for \"entertainment experiences\" and \"simulated environments for robot training\". The company had raised more than $1.2 billion in total funding; AMD itself was among the investors. The $8.2 billion deal is roughly seven times the previous valuation.\n\n## Why bet on spatial intelligence\n\nAMD's official release frames World Labs' direction as \"spatial intelligence\" — models that generate, reconstruct and simulate interactive 3D environments from text, image and video inputs, with adjacent technology for robotic learning and simulation. This sits one layer below the dominant LLM paradigm: where LLMs primarily consume text, world models treat vision, physics and spatial structure as first-class citizens. In her 2024 founding statement, Li repeated a simple thesis: \"true general intelligence requires grounding in the physical world\".\n\nThat thesis is now being amplified by Tesla, Figure and others. The real bottleneck for humanoids and autonomous driving is not that models are too small — it is that useful real-world training data is severely scarce. World models fill that gap with synthetic data and are quickly becoming the precondition for AI to enter the physical world.\n\nAMD's own framing is even more direct: understanding World Labs' frontier workloads is what will let AMD decide \"how chips should be built\". In other words, AMD is not buying a software company — it is buying a team that can articulate the hardware requirements of the next AI era.\n\n## The Nvidia mirror\n\nThis deal is, in essence, AMD's direct response to Nvidia. Nvidia has already shipped multiple open-weight world models in its Cosmos family, packaging \"model + chip + developer tools\" for autonomous driving, robotics and industrial digital-twin customers. Before this deal, AMD could only point at text- and video-only models like Instella-MoE — it lacked an entire world-model product line.\n\nAfter the deal, AMD gets, in one swoop, a turn-key spatial-intelligence research team, a product line (Marble), and an industry-icon scientist. TechCrunch has called this \"AMD's most important AI strategic move in years\", and Li's appointment has even been read as a possible succession signal for Lisa Su — AMD's research leadership in the AI era is, for the first time, being put on a public stage.\n\n## A scientist's two chapters\n\nLi's own trajectory is worth a beat. She put deep learning back on the research map with ImageNet in 2012, then spent years mentoring PhD students at Stanford. When she left academia in 2024 to start World Labs, the consensus read was \"an academic star stepping onto the casino floor\". Two years later, the company she built sold for $8.2 billion to a chipmaker that has long lacked a single \"soul figure\" of its own.\n\nHer note announcing the deal said it plainly: \"our technical breakthroughs have been proven possible — what we need to do now is not sign them but scale them, widen our reach and get closer to the hardware.\" That single sentence captures both the why and the who.\n\n## What this means for the industry\n\nThe deal rewrites expectations in at least three places.\n\nFirst, world models are being \"co-opted\" as core assets of mainstream AI companies — no longer just early demos. Cosmos is built in-house by Nvidia; Marble is acquired by AMD; Google DeepMind has had its Genie team for a while; plus Figure and Tesla are betting heavily on synthetic data. World models have moved from the edge of research to the heart of industry.\n\nSecond, the competition among chipmakers has shifted axis. It is no longer just process node, memory bandwidth or tokens-per-watt — it is now \"what kinds of models can your silicon serve well\". Once world models decide the second half of autonomous driving and robotics, chip design has to pivot from \"hardware for training generic LLMs\" toward \"hardware for training models of the physical world\".\n\nThird, the direction of talent flow has shifted. When a top researcher can walk straight into a hundred-billion-dollar hardware company as \"chief scientist\", it means academia and industry have effectively merged inside the AI vertical — in a few years, the boundary of \"the lab\" will be thinner than it has been in a decade.\n\nThe deal is expected to close by the end of 2026, pending regulatory approvals. Until then, three signals will decide whether this $8.2 billion \"science premium\" is real money or just another bubble: AMD's market-cap response, Marble's release cadence, and how much autonomy Li's team is actually given inside AMD.","amd-acquires-world-labs-fei-fei-li","2026-09-29T04:00:00Z","2026-10-01T11:10:50.626616Z","2026-10-01T11:10:50.626637Z",true,"agent",659,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"552761c1-2fb8-4952-bee5-b2a35c712ccb","谷歌把 TPU 卫星送上天:Gemma 只能跑 15 分钟","google-project-suncatcher-trillium-tpu-gemma-orbit-prototype","2026-10-18T04:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"f6e7c791-0e6b-47c4-8d26-af7163609235","LoongArch LA664 原子指令丢更新:AI 两天定位,龙芯 14 天出固件","loongarch-la664-atomic-erratum-ai-debug","2026-10-07T03:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9db08b3d-e12c-4761-9a34-69451e126ae1","谷歌把冰箱大小的 Suncatcher 送上近地轨道","google-suncatcher-tpu-gemma-orbital-launch-v2","2026-10-06T00:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"a86df58c-c601-4949-b42d-ba26c2feccb2","AI 两天锁 LoongArch 原子加丢更新:谁干关键活?","loongarch-la664-atomic-add-lost-ai-debug-2026","2026-09-28T01:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"b398dc77-58a1-498e-a9ed-c045c83c90be","AI 抢走消费级 DRAM:一年涨价五倍,手机路由器全被拖下水","ai-dram-consumer-electronics-price-surge","2026-09-14T01:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"1464179a-2b7f-4369-b680-25868ddd9042","皮尤实测：超过三分之一 ChatGPT 后的英文网页已有 AI 写作痕迹","pew-research-ai-web-content-2026","2026-08-31T03:00:00+00:00"]