[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-orbital-data-center-ieee-thermo-decade":3,"news-related-f28c994f-1922-4abc-ae57-0e53c44ff31c":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},"f28c994f-1922-4abc-ae57-0e53c44ff31c","太空算力被热力学打回原形：IEEE 量化「轨道数据中心」十年内难落地","**太空算力被热力学打回原形：IEEE 量化「轨道数据中心」十年内难落地**\n\n黄仁勋三月 GTC 上喊出\"太空计算，最后的疆界，已经到来\"不到三个月，IEEE Spectrum 就给这股轨道热泼了盆冷水。\n\n导火索是三件事并行：SpaceX 收购 xAI 后宣布\"天基算力星座\"、Google 与 Planet 合作的 Project Suncatcher（搭载自研 TPU、2027 年初发验证星）、以及 Starcloud 向 FCC 申请 88,000 颗卫星的轨道计算网络。共同叙事是太空有免费阳光、真空冷源、又不用应付地震与抗议——像是运营者的终极梦想。\n\n但热力学比硅谷的想象力更固执。真空里没有对流，散热只能靠热辐射，辐射功率与表面积成正比、与温度四次方成正比。把一块 700W 的 H100 维持在 60℃，需要约 1.4 平方米辐射板；一台装满 32 块 H100 的机柜（40kW）对应 80 平方米，约一个匹克球场。换算下来，100MW 太空数据中心需要 2,500 块这样的辐射板。\n\n低轨 5 年寿命内涂层还会衰减：紫外光与原子氧腐蚀使散热能力下降约 40%，意味着为了寿命末期仍维持 60℃，发射时要多带 40% 辐射板质量——每多一公斤都是发射成本。\n\nABI Research 用 SpaceX 星舰每公斤 44 美元的乐观价估，H100 机柜上天一年的总成本，至少比同等规模地面数据中心高一个数量级。在地面电价 0.20 美元\u002F千瓦时这种\"够朋友\"的假设下，结论依然成立。\n\n地球观测预处理、高超音速导弹实时检测、拥挤低轨的主动避碰，这些小众场景或许撑得起更贵的算力；但对\"在太空跑大模型推理\"这种通用负载，物理学回答很直白：成本曲线不站在硅谷这一边。真正入场费不是 88,000 颗卫星的牌照，而是那块从 1.4 平方米改成 2.0 平方米的辐射板。","https:\u002F\u002Fspectrum.ieee.org\u002Forbital-data-centers-heat","76eec939-9a80-4ab2-a784-301ac49c3bb0",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":18,"name":19,"slug":19,"description":13,"color":13},"8dac812d-3839-4abe-a855-5f56ec9515fd","nvidia",{"id":21,"name":22,"slug":22,"description":13,"color":13},"207ea3bd-d2e6-47a8-87b9-3959d1c8c87a","tpu",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"5734425a-5a0e-4c44-b159-ed531abfacc9","en","Thermodynamics pushes back on orbital datacenters (IEEE)","IEEE Spectrum published a comprehensive analysis of \"orbital data centers\" — the idea of putting data centers in space to take advantage of solar power and radiative cooling. The verdict from thermodynamics: orbital data centers are not economically viable for at least 10 years, due to the heat rejection problem in vacuum.\n\nThe \"orbital data center\" hype: the idea has been around for years — put a data center in low Earth orbit (LEO), power it with solar panels, and use radiative cooling (heat radiates to the cold sky) to keep it cool. The pitch is \"unlimited solar power + free cooling = cheap compute.\" Several startups (Axiom Space, Lumen Orbit, Starcloud) have raised funding on this idea.\n\nThe thermodynamics reality: the analysis shows that the heat rejection problem is much harder than the hype suggests. In vacuum, the only way to reject heat is radiation, which is highly inefficient at the temperatures a data center requires. A 1MW data center in LEO requires a 10,000 m² radiator — about the size of two football fields. The launch cost alone is $500M-$1B, and the radiator adds another $200M.\n\nThe economic verdict: even with optimistic assumptions about launch cost reduction and solar panel efficiency, orbital data centers cost 5-10× more than terrestrial data centers in the near term. The \"decade before landing\" timeline is the most optimistic — the analysis suggests 15-20 years is more realistic.\n\nThe bigger takeaway: \"physics-first\" analysis is essential for hype-y ideas. The \"orbital data center\" pitch is appealing in the abstract, but the thermodynamics make it impractical for the foreseeable future. For the industry, this signals that the \"compute is moving to space\" narrative is premature, and terrestrial data centers (with renewable energy) will remain the dominant paradigm for the next decade+.","orbital-data-center-ieee-thermo-decade","2026-06-14T06:10:00Z","2026-06-14T06:08:35.744832Z","2026-08-19T02:08:40.142862Z",true,"agent",99,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"f453f1d8-c41b-4fdf-9eba-54a7f6222d74","英伟达把自动驾驶十年的安全账本搬进机器人：Halos for Robotics 全栈落地工厂","nvidia-halos-for-robotics-fsi-igx-thor","2026-06-23T10:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"66d66fa2-364e-4fa2-a9b2-e69e6f86dc8c","Vera Rubin 平台登陆 ISC 2026：144 张 GPU + 100% 液冷，把 TOP500 算力压进科研机柜","nvidia-vera-rubin-isc2026-144-gpu","2026-06-22T20:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"6242571f-c635-4794-b19a-ba08eac4f6d1","NVIDIA 发布全球首款面向 Agent 时代的 CPU：Vera 已送抵 Anthropic、OpenAI、SpaceXAI","nvidia-vera-cpu-agent-anthropic-openai","2026-05-20T01:30:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"fb1cbe25-8b85-41ec-b619-9a27b405ec34","AMD 收购 Taalas:把 AI 模型权重「刻进硅片」的推理新打法","amd-acquires-taalas-inference-chip","2026-08-19T01:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"45375854-7739-4dd1-bc6a-30db4474652a","Taalas HC2:把单片参数拉到 200 亿,「模型刻进硅片」的第二章","taalas-hc2-20b-mxfp4-50-chips-1t-amd","2026-08-19T00:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"f333dd36-d9ed-4e17-a601-11b4f140eee3","Taalas HC2 把参数上限拉到 200 亿：AMD 这张「把模型刻进硅片」的牌,开始讲下一章","taalas-hc2-20b-mxfp4-amd","2026-08-15T03:30:00+00:00"]