字号·· | 护眼
欧洲新闻台

乌克兰打击俄罗斯数据中心,重创莫斯科的AI能力Ukraine hits Russian data centres in blow to Moscow's AI capabilities

点「原文对照」整页切到原文,或双击某段只看那段的原文。

2026年10月20日,萨索沃的Yandex数据中心在遭到乌克兰无人机袭击后的景象。Yandex data centre in Sasovo after a strike by Ukrainian drones, 20 October 2026.

乌克兰无人机于10月袭击了俄罗斯最大科技公司的五个数据中心,导致云服务、支付系统和导航应用在全国范围内大面积故障。

Ukrainian drones struck five data centres belonging to Russia's largest technology companies in October, causing widespread failures in cloud services, payment systems and navigation apps nationwide.

乌克兰武装部队袭击了三个Yandex(扬德克斯)设施:10月7日位于梁赞州萨索沃,10月9日位于卡卢加,10月11日位于弗拉基米尔。10月14日,萨拉托夫州巴拉科沃的一座Sber(斯伯银行)数据中心遭袭,10月18日,圣彼得堡附近列宁格勒州的一座VKontakte(VK)数据中心遭到攻击。

The Ukrainian armed forces hit three Yandex facilities: in Sasovo in the Ryazan region on 7 October, in Kaluga on 9 October, and in Vladimir on 11 October. On 14 October a Sber data centre in Balakovo in the Saratov region was struck, and on 18 October a VKontakte data centre in the Leningrad region near St Petersburg came under attack.

Yandex、Sber和VKontakte共同运营着俄罗斯消费级互联网基础设施的大部分,涵盖打车应用、银行卡支付处理、云存储和即时通讯等。

Yandex, Sber and VKontakte together run much of Russia's consumer internet infrastructure, from taxi apps and card payment processing to cloud storage and messaging.

这些公司均未披露损失规模。俄罗斯民众反映,支付购物、叫车、使用导航服务和访问云存储持续遇到困难。

None of the companies has disclosed the scale of the damage. Russians have reported persistent difficulties paying for goods, ordering taxis, using navigation services and accessing cloud storage.

互联网保护协会负责人米哈伊尔·克利马列夫告诉美联社,专家表示恢复基础设施可能需要数月甚至数年,且这些袭击已使俄罗斯的人工智能发展倒退。俄罗斯持续打击乌克兰IT行业这些袭击发生之际,俄罗斯本身在这个秋天一直在系统性地打击乌克兰数据中心。

Experts say restoring the infrastructure could take months or years, and the strikes have set back Russia's artificial intelligence development, Mikhail Klimarev, head of the Internet Protection Society, told AP. Russia continues to target Ukraine's IT sectorThe attacks come as Russia has itself been systematically targeting Ukrainian data centres this autumn.

10月19日,一次俄罗斯袭击彻底摧毁了乌克兰运营商Datagroup在基辅最大的设施之一,该设施被宣布无法修复。次日,一次俄罗斯袭击损坏了基辅的GigaCenter。

On 19 October a Russian strike completely destroyed one of the largest facilities of the Ukrainian operator Datagroup in Kyiv, which was declared beyond repair. The following day a Russian attack damaged Kyiv's GigaCenter.

俄罗斯在9月和10月袭击了乌克兰的其他几个数据中心。此前在7月,俄罗斯对基辅一个数据中心的袭击导致全市互联网接入、移动通信和电视服务中断。自俄罗斯全面入侵乌克兰的第一天起,莫斯科就以乌克兰的IT系统为目标:2022年2月24日,巡航导弹袭击了一个政府服务器设施;3月1日,一枚俄罗斯导弹击中了基辅市中心的国家数据备份中心。此后,乌克兰将大部分数据转移到了欧洲国家的服务器上,限制了后续袭击的影响。

Russia hit several other Ukrainian data centres in September and October. An earlier Russian strike on a Kyiv data centre in July disrupted internet access, mobile communications and television across the city. Moscow has targeted Ukraine's IT systems since day one of Russia's full-scale invasion: cruise missiles struck a government server facility on 24 February 2022, and a Russian missile hit the National Data Backup Centre in central Kyiv on 1 March. Ukraine has since moved most of its data to servers in European countries, limiting the impact of subsequent attacks.

分析人士称,俄罗斯的数据中心比乌克兰的更脆弱。根据旨在建立所谓“主权互联网”的立法,俄罗斯要求公司将所有数据存储在该国境内的服务器上。这种集中化使得这些设施更容易成为攻击目标。

Russian data centres are more vulnerable than Ukrainian ones, analysts say. Under legislation designed to create a so-called "sovereign internet," Russia requires companies to store all data on servers inside the country. That concentration makes the facilities easier to target."

“俄罗斯的问题在于,它必须为广阔的领土提供防空保护,但没有足够的资源做到这一点,”欧亚集团高级地缘技术分析师尼克·赖纳斯说。“由于所有基础设施和所有数据都位于俄罗斯境内,要使它们瘫痪相对容易一些。”

Russia's problem is that it has to provide air defence over a vast territory, and it does not have enough resources for that," said Nick Reiners, senior geo-technology analyst at Eurasia Group." Since all the infrastructure and all the data are located inside Russia, it is relatively easier to disable them."

为什么防空系统没有保护俄罗斯数据中心,或者未能做到这一点,目前尚无解释。怕火分析人士将这些设施描述为软目标:装满服务器、电力设备和冷却系统的巨大大厅,即使是少量的炸药也能造成严重破坏。被袭击的Yandex设施的卫星图像显示,大火可能已经摧毁了微芯片和其他易碎组件。

Why air defence systems were not protecting Russian data centres, or failed to do so, has not been explained. Sensitive to fireAnalysts describe the facilities as soft targets: vast halls filled with servers, power equipment and cooling systems where even a small explosive charge can cause serious damage. Satellite imagery of the struck Yandex facilities shows fires that may have destroyed microchips and other fragile components."

“遭受此类袭击后,很难指望有什么东西能幸存下来:数据中心设备脆弱、挑剔且昂贵。火灾产生的烟灰对电子设备极其有害,”克利马列夫告诉美联社。

After such a strike it is hard to expect anything to have survived: data-centre equipment is fragile, demanding and expensive. Soot from fires is extremely harmful to electronics," Klimarev told AP.

对Yandex的破坏对俄罗斯的AI雄心尤为重要。该公司的三台超级计算机用于开发AI模型,位居世界前500名。

The damage to Yandex is particularly significant for Russia's AI ambitions. The company's three supercomputers, used to develop AI models, rank among the world's top 500.

最强大的那台超级计算机位于萨索沃设施,曾位列全球第19位,但此后已跌出前100名。“乌克兰的无人机袭击可能会进一步推迟俄罗斯人工智能的发展,”外交关系委员会的康纳·马丁表示。在袭击发生前,俄罗斯在人工智能领域已落后美国和中国数年。

The most powerful, housed in the Sasovo facility, once ranked 19th globally but has since fallen below the top 100." Ukrainian drone strikes will probably delay the development of Russian artificial intelligence even further," Connor Martin of the Council on Foreign Relations said. Before the strikes, Russia was already several years behind the US and China.

由于西方制裁和全球芯片短缺,恢复硬件将变得复杂。Yandex的超级计算机运行在英伟达A100处理器上,而俄罗斯无法合法进口这些处理器。

Restoring the hardware will be complicated by Western sanctions and the global chip shortage. Yandex's supercomputers run on Nvidia A100 processors, which Russia cannot legally import.

克利马列夫表示,这些袭击可能会给俄罗斯互联网政策带来意想不到的后果。“具有讽刺意味的是,对数据中心的袭击使俄罗斯互联网免于孤立,因为公司现在将被迫转移其计算资源,最有可能转移到欧洲,”他说。“俄罗斯公司很有可能正在研究欧洲市场,寻找租用算力的机会,”克利马列夫说。他补充称,俄罗斯当局可能会放松收紧互联网管控的尝试,以限制经济损失。用于恶意目的的基础设施多年来,莫斯科积极利用其IT部门监视用户,并在其他地方开展恶意混合行动。

Klimarev said the strikes may carry an unintended consequence for Russian internet policy." Paradoxically, the attacks on data centres have saved the Russian internet from isolation, because companies will now be forced to move their computing resources, most likely to Europe," he said." It is quite likely that many Russian companies are now studying the European market in search of opportunities to rent computing capacity," Klimarev said. He added that Russian authorities might ease their attempts to tighten internet controls to limit the economic damage. Infrastructure for malign purposesMoscow has for years actively used its IT sector for spying on its users, as well as conduct malign hybrid operations elsewhere.

根据俄罗斯的SORM立法,包括Yandex、VKontakte和Sber在内的所有国内互联网公司均被法律要求安装联邦安全局(FSB)后门,授予安全机构访问用户数据和通信的权限。

Under Russia’s SORM legislation, all domestic internet companies including Yandex, VKontakte and Sber are legally required to install FSB backdoors giving the security services access to user data and communications.

这些数据中心融入了俄罗斯的监控体系,并非纯粹的商业基础设施。VKontakte具体而言被广泛用于国家资助的宣传和招募。

These data centres are integrated into Russia’s surveillance apparatus, not purely commercial infrastructure.VKontakte specifically has been extensively used for state-sponsored propaganda and recruitment.

乌克兰于2017年部分基于这些理由将其禁用。克里姆林宫已被证实在其虚假信息行动中使用了AI工具,例如重写真实新闻文章,植入亲克里姆林宫的倾向性内容和捏造的说法,并将其发布在数十个旨在模仿合法国际媒体(包括欧洲新闻台)的网站上。它还利用AI生成的视频和音频捏造归因于西方官员的言论。冒充欧洲政客的AI克隆声音也曾流传,尽管其中一些在大范围传播前被识破。

Ukraine banned it in 2017 partly on those grounds. The Kremlin has demonstrably used AI tools in its disinformation operations, such as rewriting real news articles with pro-Kremlin slants and fabricated claims, publishing them across dozens of sites designed to mimic legitimate international outlets, including Euronews. It has also used AI-generated video and audio to fabricate statements attributed to Western officials. AI-cloned voices impersonating European politicians have also circulated, though some were detected before they spread widely.

最近,它向互联网灌输重复性的亲克里姆林宫内容,以扭曲AI聊天机器人检索的内容,并最终影响其训练方式。《新闻卫士》的压力测试发现,在对抗性提示下,聊天机器人在大约三分之一的回答中重复了虚假说法。然而,卡内基国际和平基金会的研究人员指出,训练数据投毒过程缓慢、经过严格过滤,且越来越多地被新型模型内置的行为规则所遏制。

More recently, it has flooded the internet with repetitive pro-Kremlin content to skew what AI chatbots retrieve and eventually how they are trained. NewsGuard stress tests found chatbots repeated false claims in roughly a third of responses when prompted adversarially. Researchers at the Carnegie Endowment have however stated that training-data poisoning is slow, heavily filtered and increasingly countered by behavioural rules built into newer models.EU sanctions prohibit the supply of IT services, software and cloud capacity to Russian companies.

欧盟制裁禁止向俄罗斯公司提供IT服务、软件和云算力。寻求规避限制的企业利用了中介商、匿名托管服务商或外国企业架构。布鲁塞尔一直在收紧对这些规则的执法力度。

Firms seeking to circumvent the restrictions have used intermediaries, anonymous hosting providers or foreign corporate structures. Brussels has been tightening enforcement of those rules.

阅读媒体原文 ↗