科技巨头设想将数据中心搬上太空
内容总结：
随着人工智能产业爆发式增长，其背后的数据中心正面临严峻的能源与环境挑战。数据显示，到2030年，AI数据中心的电力需求可能激增165%，目前超半数电力依赖化石燃料，不仅加剧电网压力，更可能逆转全球气候治理进程。
面对质疑，科技巨头提出颇具科幻色彩的解决方案：将数据中心部署至太空。OpenAI首席执行官山姆·奥特曼公开表示，尽管意识到地面数据中心的环境负面影响，仍认为其扩张不可避免，甚至提出围绕太阳构建"戴森球"式太空数据中心的设想。亚马逊创始人贝佐斯、前谷歌CEO施密特等人也对此领域表示关注。
太空数据中心的理论优势显著：可全天候利用太阳能供电，避免对地面社区的水资源消耗和污染问题。加州理工学院太空太阳能项目专家阿里·哈吉米里指出，随着发射成本下降（现已降至约每公斤1500美元）和太阳能技术提升，太空发电成本有望降至每度电0.1美元。但关键技术瓶颈依然存在——太空辐射对设备的破坏、数据延迟传输、设备更新维护难题等尚未解决。
尽管初创企业已展开实践探索（如Lonestar公司在月球部署微型数据中心、Starcloud计划发射搭载英伟达芯片的卫星），哈佛大学经济学家马修·温齐尔仍强调，其经济可行性尚未得到验证："除非能在成本和服务质量上与地面中心竞争，否则难以实现规模化应用。"
值得关注的是，太空领域尚未完善的监管体系可能成为企业布局的动机。密西西比大学航空航天法中心主任米歇尔·汉伦指出："太空目前没有邻避问题，也没有严格的审批流程——这对企业具有独特吸引力。"但随着商业化进程推进，相关立法监管必然提上议程。
当前，在弗吉尼亚等数据中心聚集区，电力需求预计十年内翻倍。在经济效益主导下，企业仍倾向于优先扩建地面设施。太空数据中心能否从科幻走向现实，不仅取决于技术突破，更需跨越经济性与可持续性的双重考验。
中文翻译：
《连线》杂志推荐的所有产品均由编辑独立遴选。然而，我们可能通过零售渠道或产品购买链接获得补偿。了解更多。
本文原载于Grist网站，系"气候专题"合作内容。
那些疯狂建设和租赁数据中心的企业心知肚明：它们正在加剧电网负荷、推高碳排放、吞噬水资源。到2030年，人工智能数据中心的电力需求增幅可能高达165%。这些庞设施超过一半的能源来自化石燃料，可能逆转应对气候危机的既有进展。
人工智能领域的巨头们声称已有解决方案：将这些巨型计算机集群送入太空。OpenAI首席执行官萨姆·奥尔特曼向男性话题播主西奥·冯表示，数据中心的急剧扩张不可避免。"我预测未来世界将遍布数据中心"（这实则并非必然，而是超级企业选择投入天量资金的结果。奥尔特曼曾透露可能投入数万亿美元，OpenAI正参与耗资5000亿美元的"星门"项目）。
不过奥尔特曼也意识到可能引发的争议："我与环保人士交流过"，随后提出设想，"或许可以把数据中心安置在太空。虽然目前缺乏具体方案，但我们正在探索。"
将占地面积超百万平方英尺的数据中心送入轨道看似不切实际，但持此想法者不乏其人。杰夫·贝索斯与埃里克·施密特同样看好这个构想。奥尔特曼甚至提议建造环绕太阳的"戴森球"数据中心——这种假想中的恒星能源收集装置存在明显缺陷：建造所需资源超过地球储量，且可能令星球不再宜居。但更现实的方案正在推进，Starcloud、Axiom、Lonestar等初创公司已获数百万美元融资开展研发。
美国现有5400多个数据中心，从微型设施到拥有数千服务器的超大规模中心，数量仍在快速增长。预计到2028年，这些设施将消耗全美12%的电力。太空部署看似一劳永逸：通过全天候太阳能解决能耗问题，让社区摆脱空气、噪音和水污染困扰。
该设想具有科学依据。加州理工学院空间太阳能项目教授阿里·哈吉米里早在2016年就为"太空超大规模并行计算系统"（即数据中心）申请专利。如今发射成本已降至每公斤约1500美元，太阳能板更轻效。哈吉米里团队近期提出轻型天基太阳能系统，发电成本可降至每度电0.1美元，大幅低于地表同类系统。理论上这种技术能为轨道数据中心供电，但哈吉米里不确定何时能实现OpenAI要求的规模："我从不断言某事不可为，但这确实存在挑战。"
首先，太空数据处理速度远逊于地面系统。它们将持续遭受辐射冲击，且"设备淘汰会成为难题"——维修升级极其困难。哈吉米里认为太空数据中心终将可行，但难以预估时间表："数年內实现确有可能，关键在于效能与成本效益。"
这个构想不仅限于技术人员的随想或学者的深究。在亚马逊等企业计划建设数据中心的城市，连民选官员也提及此方案。8月，亚利桑那州图森市议会全票否决本地数据中心建设提案时，议员尼基·李充满诗意地阐述："许多人说沙漠不该建数据中心，但若这真是国家优先事项，就该将联邦研发资金投入太空数据中心研究。这听起来或许荒诞如科幻，但确实正在发生。"
事实如此，不过目前仍处于实验阶段。Starcloud公司原定8月发射装载英伟达芯片的冰箱大小卫星，但已延期。Lonestar数月前在月球着陆的微型数据中心（载有Imagine Dragons乐队歌曲等珍贵数据）最终倾覆失效。更多发射计划正在筹备，但哈佛大学研究太空市场经济学家马修·温齐尔表示："难以预测其经济可行性的实现速度。天基数据中心或有特定用途，如处理太空数据、增强国安能力。但要与地面中心真正竞争，仍需在成本与服务品质上取胜。"
目前太空数据中心的成本远高于地面建设（如弗吉尼亚数据中心谷——若无管制其电力需求十年内将翻番）。只要地球运营更经济，逐利企业就会选择地面扩张。然而有个因素可能促使OpenAI等公司仰望星空：太空几乎无监管。密西西比大学航空航天法中心主任米歇尔·汉伦指出：地球建设需市政许可，可能因居民担忧耗水、电费上涨或气候影响而受阻。"在太空没有邻居投诉。若是美国企业想部署太空数据中心，最好赶在国会立法监管之前行动。"
英文来源：
All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links. Learn more.
This story originally appeared on Grist and is part of the Climate Desk collaboration.
The companies frantically building and leasing data centers are well aware that they’re straining grids, driving emissions, and guzzling water. The electricity demand of AI data centers in particular could increase as much as 165 percent by 2030. Over half of the energy powering these sprawling facilities comes from fossil fuels, threatening to reverse progress toward addressing the climate crisis.
Some of the biggest names in artificial intelligence say they have a solution: Just stick these colossal computer clusters in space. OpenAI CEO Sam Altman told manosphere podcaster Theo Von that he considers a massive expansion of data centers inevitable. “I do guess a lot of the world gets covered in data centers over time,” he said. (This is not, in fact, inevitable, but the result of unfathomably wealthy companies choosing to invest unfathomably large sums of money. Altman has speculated that he would quite literally put trillions into it, and OpenAI is part of the consortium behind the $500 billion Stargate project.)
Altman is aware, however, that some people might not like this. “I’ve spoken with environmentalists,” he said. Then, he offered a suggestion. “Maybe we put [data centers] in space,” he said. “I wish I had, like, more concrete answers for you, but like, we’re stumbling through this.”
Now, the idea of hurling data centers, the largest of which can cover over a million square feet, into orbit may seem impractical. But Altman’s not alone in considering it. Jeff Bezos and Eric Schmidt are also betting on the idea.
Altman has proposed creating a Dyson sphere of data centers around the sun, referring to a hypothetical megastructure built around a star to capture much of its energy. The rather glaring downside to this is that building it would likely require more resources than exist on Earth, and could make the planet uninhabitable. But somewhat more realistic plans are inching closer to reality. Startups like Starcloud, Axiom, and Lonestar Data Systems have raised millions to develop them.
There are at least 5,400 data centers in the United States, ranging from micro-size to thousand-server “hyperscalers,” and the number is growing fast. These facilities are expected to consume up to 12 percent of the nation’s electricity by 2028. Putting them in space, then, can seem like a panacea: solving the energy-use problem with 24/7 solar power, and freeing communities from the burden of air, noise, and water pollution.
There’s some real science behind this. Ali Hajimiri, an electrical engineer and professor with Caltech’s Space Solar Power Project, sought a patent for a “massively parallel computational system in space”—as in, a data center—back in 2016. Since then, launch costs have gone down (to around $1,500 per kilogram, by one estimate) and solar panels have gotten lighter and more efficient. Hajimiri and his colleagues recently proposed a lightweight space-based solar power system that could generate electricity at 10 cents per kilowatt-hour, significantly cheaper at scale than comparable systems here on Earth. Such technology theoretically could power orbital data centers like those Altman imagines, though Hajimiri is still not sure when they could be built at the kind of scale companies like OpenAI demand. “I never want to say something cannot be done,” he said. “But there are challenges associated with it.”
For one thing, the systems he imagines process data relatively slowly compared to those on terra firma. They’d be constantly bombarded by radiation, and “obsolescence would be a problem” because making repairs or upgrades would be confoundingly difficult. Hajimiri believes that data centers in space could, someday, be a viable solution but hesitates to say when that day might come. “Definitely it would be doable in a few years,” he said. “The question is how effective they would be, and how cost-effective they would become.”
The idea of simply putting data centers in orbit is not limited to the offhand musings of techies or the deeper thought of academics. Even some elected officials in cities where companies like Amazon hope to build data centers are raising the point. Tucson, Arizona, councilmember Nikki Lee waxed poetic about their potential during an August hearing, in which the council unanimously voted down a proposed data center in their city.
“A lot of people are saying data centers don’t belong in the desert,” Lee said. But “if this is truly a national priority,” then the focus must be on “putting federal research and development dollars into looking at data centers that will exist in space. And that may sound wild to you all and a little science fiction, but it’s actually happening.”
That’s true, but it’s happening on an experimental scale, not an industrial one. A startup called Starcloud hoped to launch a refrigerator-sized satellite housing a few Nvidia chips in August, but the launch date was pushed back. Lonestar Data Systems landed a miniature data center, carrying precious information like an Imagine Dragons song, on the moon a few months ago, though the lander tipped over and died in the attempt. More such launches are planned for the coming months. But it’s “very hard to predict how quickly this idea will become economically feasible,” said Matthew Weinzierl, a Harvard University economist who studies market forces in space. “Space-based data centers may well have some niche uses, such as for processing space-based data and providing national security capabilities,” he said. “To be a meaningful rival to terrestrial centers, however, they will need to compete on cost and service quality like anything else.”
For now, it’s much more expensive to put a data center in space than it is to put one in, say, Virginia’s Data Center Valley, where power demand could double in the next decade if left unregulated. And as long as staying on Earth remains cheaper, profit-motivated companies will favor terrestrial data-center expansion.
Still, there is one factor that might encourage OpenAI and others to look toward the heavens: There isn’t much regulation up there. Building data centers on Earth requires obtaining municipal permits, and companies can be stymied by local governments whose residents worry that data center development might siphon their water, raise their electricity bills, or overheat their planet. In space, there aren’t any neighbors to complain, said Michelle Hanlon, a political scientist and lawyer who leads the Center for Air and Space Law at the University of Mississippi. “If you are a US company seeking to put data centers in space, then the sooner the better, before Congress is like, ‘Oh, we need to regulate that.’”