内容来源：https://www.wired.com/story/uncanny-valley-podcast-wired-roundup-tech-politics-trends-2025/

内容总结：

【年终特稿】2025年回顾：AI狂潮、政府效率部争议与未解之谜

2025年是科技与政治深度交织、充满戏剧性转折的一年。从人工智能产业重塑全球经济格局，到备受争议的“政府效率部”（DOGE）引发联邦机构震荡，一系列关键事件不仅定义了这一年，也为2026年的趋势埋下了伏笔。

AI数据中心：繁荣背后的隐忧
人工智能基础设施投资在今年达到惊人规模，Meta、谷歌、微软等巨头纷纷加码AI数据中心建设。这些中心不仅是AI体验的物理基石，更在拉动经济、推高能源价格、引发环境担忧等方面产生多重影响。有分析指出，高达60%的数据中心建造成本用于购买GPU，且需每三年更换，其商业模式可持续性存疑。部分投资者，如因《大空头》闻名的迈克尔·伯里，已公开押注“AI泡沫”即将破裂。尽管行业预测泡沫完全显现可能还需时日，但数据中心对地方资源（如水电）的巨量消耗及就业承诺与实际就业机会的落差，已开始引发多地社区反弹，未来或面临政治压力与海外转移趋势。

AI伴侣：情感陪伴与伦理困境
随着AI推理需求激增，聊天机器人伴侣及AI人际关系在今年显著兴起。这类产品在缓解部分用户孤独感的同时，也引发了多起悲剧事件，例如有报道称其可能与自杀案例相关。业界普遍认为，AI伴侣的社会心理影响尚处早期阶段，亟需更多前瞻性研究与安全保障措施。这不仅是技术问题，更折射出社会连接缺失的深层议题。

中美AI竞赛：开源模式与战略博弈
中国AI实验室深度求索（DeepSeek）在今年初发布开源模型R1，引发全球市场震动，甚至导致英伟达股价单日创纪录暴跌。该事件凸显了中国在开源AI模型领域的快速进展及其潜在影响力。开源（开放权重）模式允许全球开发者协作改进，正成为中国AI发展的优势路径。相比之下，美国企业正趋向闭源，Meta等此前开源先锋也暗示未来模型可能转为专有。这种路径分歧可能带来战略差异，并引发关于重复研发造成资源浪费的讨论。尽管中国模型存在内容审核限制，但出于成本与性能考虑，其吸引力并未减弱。围绕对华芯片出口管制的政策辩论，也因此变得更加复杂。

“政府效率部”（DOGE）：混乱与持久影响
由埃隆·马斯克在特朗普政府支持下主导的“政府效率部”，今年一度深入联邦机构推行改革，目标包括大幅削减预算、整顿所谓“低效”部门。其行动导致了大规模联邦雇员裁员、美国国际开发署（USAID）功能严重萎缩等后果。尽管未能实现其宣称的万亿削减目标，且马斯克近期访谈中流露出悔意，但DOGE的遗产已然显现。其推动的政府数据壁垒拆除项目，将社保、税务、国土安全等多部门数据整合，极大增强了移民与海关执法局（ICE）的监控能力，对移民政策产生了深远影响。这一数据融合实践可能永久改变美国政府的隐私与数据管理格局。

爱泼斯坦文件：政治承诺与“真相”迷雾
特朗普总统竞选时承诺公开的“爱泼斯坦文件”，在上台后经历反复，成为年度政治悬念之一。尽管部分文件已被国会批准公开，但其中内容（涉及诸多名流联系）未能满足所有阴谋论的期待。司法部发布的爱泼斯坦狱中视频被WIRED揭露存在近3分钟剪辑，尽管可能有技术原因，但官方此前声称视频“未经编辑”的说法加剧了公众疑虑。这一事件凸显了在高度两极化的信息环境中，建立共识性“真相”面临的巨大挑战。

展望2026年，AI发展的经济与社会影响将持续深化，政府效率部改革的数据监控遗产或将进一步发酵，而围绕信息真实性的信任危机，预计仍将是公共舆论场的核心议题。

中文翻译：

无论好坏，2025年可谓包罗万象——从塑造全球经济与生活的AI产业，到埃隆·马斯克主导的所谓“政府效能部”接管美国联邦机构。在本期节目中，主持人佐伊·希弗与执行主编布莱恩·巴雷特将共同回顾今年的一些关键事件，并探讨这些事件如何为我们预示即将到来的2026年。

本期节目提及的文章：

• 《AI数据中心热潮正在扭曲美国经济》
• 《美国需要开源AI干预以击败中国》
• 《DOGE就是深层政府》
• 《ICE计划组建全天候社交媒体监控团队》
• 《FBI的杰弗里·爱泼斯坦监狱录像被删减近3分钟》

您可以在Bluesky上关注佐伊·希弗（@zoeschiffer）和布莱恩·巴雷特（@brbarrett）。欢迎发送邮件至uncannyvalley@wired.com与我们联系。

收听方式
您可以通过本页面的音频播放器收听本周节目，若想免费订阅以获取每期更新，请参考以下方式：
如果您使用iPhone或iPad，请打开“播客”应用，或直接点击此链接。您也可以下载Overcast或Pocket Casts等应用，搜索“uncanny valley”。我们的节目亦可在Spotify收听。

文字记录
注：此为自动生成的字幕稿，可能存在误差。

佐伊·希弗：欢迎收听《连线》杂志的“诡异谷”播客。我是佐伊·希弗，《连线》商业与产业板块总监。在本期节目中，我们将通过回顾定义2025年的趋势与事件，为今年的新闻系列报道画上句号。而最合适的对谈者，莫过于在幕后孜孜不倦工作的执行主编布莱恩·巴雷特了。

布莱恩·巴雷特：佐伊，谢谢你的邀请。很高兴能从我的幽暗巢穴里暂时现身。

佐伊·希弗：从那个黑漆漆的洞穴里出来啦。谢谢你能来。

布莱恩·巴雷特：这一年真是波澜壮阔，而我无比兴奋它终于快要结束了。

佐伊·希弗：天啊，我也是。好吧，因为从新闻角度看，这一年确实非同寻常，尤其在科技和政治领域。老实说，挑选今天要讨论的趋势有点棘手，但我们最终确定了五个故事，它们很好地概括了这一年，并为我们窥见2026年的动向提供了线索。第一个我想谈的话题与我心心相印，那就是AI数据中心。我们都知道，对数据中心的投资金额惊人，Meta、谷歌和微软等公司今年在AI基础设施上的支出增加了两倍。但这不仅仅是花钱的问题，更关乎资金投向何处，以及它已经对科技行业乃至整个经济产生的连锁反应。

布莱恩·巴雷特：很高兴我们讨论数据中心，因为它们听起来——也应该——很无聊。在一个理想的世界里，它们确实如此。但如今它们却处于许多事件的核心。它们在很多方面独自支撑着经济，造成了巨大的环境扰动，推高了人们的能源价格，同时驱动着那些时而精彩、时而糟糕的AI体验——这取决于使用者和使用方式。如果五年前有人说科技界最大的故事将是数据中心，我想没人会相信。

佐伊·希弗：我的意思是，如果当时告诉我，我们会有多名记者的报道领域直接与数据中心相关，我肯定会觉得无聊透顶，跳过不谈。但我们总在寻找与彼得·蒂尔的共识点，我想这就是其中之一。彼得·蒂尔说过，美国确实没有其他大型登月项目了，我们现在没有曼哈顿计划，我们只有人工智能。我觉得他说这话时是把它当作一件坏事：我们应该有其他重大倡议。但我觉得更具体地说，在AI领域，核心就是AI数据中心。我们听到业内很多人，比如萨姆·奥尔特曼和其他高管，都在谈论一个事实：看起来确实有一个AI泡沫正在形成并可能扩大。

而数据中心正处于泡沫的中心。我认为很多这类金融交易是通过特殊目的实体进行的，所以这些支出并不直接体现在公司的资产负债表上。另外，建设数据中心成本的60%基本上都花在GPU上，而且每三年就需要更换这些GPU。所以很多人看到这种情况都相当担忧，他们说：“这笔账看起来算不平。”

布莱恩·巴雷特：是的。迈克尔·伯里，他在房地产危机中成名，对吧？他是《大空头》的核心人物。他下了相当大的赌注，认为这里的会计处理有问题，正如你所说，这是一个泡沫，而且会破裂。当然，他以前也下过一些没有成功的赌注。

佐伊·希弗：我正想说这个。

布莱恩·巴雷特：他并非永不犯错，但他的观点代表了你能看到的一种论点，而且他是有过成功捕捉泡沫经历的人。佐伊，2026年会是这一切达到顶点的一年吗？连萨姆·奥尔特曼都说这可能是个泡沫，会有赢家和输家。他显然在赌自己会成为AI泡沫中的谷歌——挺过泡沫，生存下来，发展壮大，并主导市场多年。你觉得我们2026年就会走到那一步，还是说我们还有更多时间？

佐伊·希弗：实际上我认为我们还有不错的一年——所谓“不错”取决于你在这个领域的位置。但我猜——这主要是基于感觉，所以请对此持保留态度——我们可能要到2027年才会真正迎来顶点。我特别关注的一点是，我去参观过一些数据中心，也和那些非常支持在自己城市建数据中心的政治家聊过。目前看来，在很多地方，政治上付出的代价似乎并不大。当然，地方上有很多反对声音，但在很多地方，这些数据中心建在非常支持共和党的州，那里就业机会不多。所以它们被宣扬为一种经济福音：“看，这即将到来的经济繁荣将对本地经济非常有利，会带来所有这些工作岗位。”

但事实是，第一，建设数据中心需要成千上万的人，但维护数据中心所需的人手要少得多。所以一旦建成运行，很多工作岗位就会消失。第二，它们资源消耗非常非常大。冷却数据中心需要大量水，显然还会消耗巨量能源。所以我认为我们会看到一个转变——这只是个预测——支持数据中心将在政治上变得难以为继，未来三年我们将看到向海外转移的推动力。

布莱恩·巴雷特：而且即使是建设数据中心的那些成千上万的岗位，也不总是本地人来做，因为很多是专业人员从外地过来。我住在一个红州，那里也有反对声音。他们想在我附近建一个数据中心，但反对声浪还不足以阻止任何事情，这东西还是会建起来。但是，是的，我确实认为，当人们开始把各种事情联系起来——“哦，我的电费现在更高了”，“哦，我以为会有的工作岗位并没有出现”——我同意，我们会看到更多反对。而那时，我们可能就要在太空建数据中心了。

佐伊·希弗：我正想说在火星上建呢。我目前居住的圣巴巴拉没有数据中心，但我们住的地方非常靠近SpaceX的火箭发射场。对于一个平时对政治并不热衷的社区来说，哇，人们真的会团结起来反对那些火箭发射，因为这会吓到他们的狗和马等等。所以我很想看看如果他们试图在这里建数据中心会发生什么。

布莱恩·巴雷特：那会很有意思。

佐伊·希弗：好了。我觉得如果不谈谈建这些数据中心是为了什么，那就是我们的失职，因为它们是为了聊天机器人，为了ChatGPT，是为了推理——不仅仅是训练模型，更是为了支持数百万、数千万人向聊天机器人提问并得到答案。我觉得今年我们真切感受到的一个趋势是聊天机器人伴侣和AI关系的兴起。我很好奇你现在对此有何感受，以及这种感受从年初到现在是否发生了变化？

布莱恩·巴雷特：这很有趣。一方面，发生了很多聊天机器人伴侣出问题的个别事件。据称有互动导致或促成了自杀，还有一些看似非常不健康的关系。但另一方面，也有一些案例显示人们似乎真的更快乐了，它们似乎在填补某种空虚。我不知道。就我个人而言，我一直有种本能的反应：“哦不，我们别这样。”但我意识到我需要转变想法：“好吧，这不适合我，但这并不意味着它没有存在的空间。”我认为最重要的是，这一切都还处于非常早期的阶段。我们需要做更多工作来弄清楚这些关系对人产生的影响是好是坏，还是中性。

我希望一些AI公司能在前期就做这些工作，而不是简单地说“去和你的新AI男友或女友玩得开心吧”，却没有真正理解后果是什么，因为目前还没人真正知道。现在还为时过早。所以我很高兴看到很多公司在多个层面设置了更多保障措施，但这似乎还是太快、太多了。我们需要先弄清楚到底是怎么回事。

佐伊·希弗：这是我一直在思考的关于“AI躁狂症”或所谓“AI精神病”的问题。因为很多时候，至少在我作为科技记者的经验里，手头的问题——无论是错误信息还是其他什么——都有技术上的原因，但感觉需要多管齐下的解决方案，而我们却只指望科技公司来解决这个影响我们民主制度的整个问题。而对于AI精神病或AI躁狂症，我不知道，这确实感觉像是一个技术问题。如果聊天机器人告诉你你发现了物理学的新前沿，或者它一再地肯定你，而你相信了，那似乎是聊天机器人的问题。但对于AI关系，我想知道，当它成为问题时，情况是否更复杂。

如果社会层面正在发生其他事情，而聊天机器人关系只是其症状而非原因，那么我认为我们绝对需要更多的保障措施。同时，社会层面也需要做很多事情，以便我们与他人有更多联系，更容易建立这些联系，有更多推动我们彼此共处的事情。

布莱恩·巴雷特：佐伊，我们要复兴保龄球联盟。我们要复兴保龄球联盟。还有什么比这更好的呢？

佐伊·希弗：今年《连线》密切关注的另一个趋势是前沿AI模型的全球竞争。我保证，我们会谈到不以AI为中心的故事和趋势，但假装AI不是今年我们行业的核心故事就太虚伪了。其中一个让这一点变得非常清晰的时刻发生得很早。如果你还记得一月份，中国AI研究实验室深度求索发布了R1开源模型，当时感觉就像炸开了锅。

布莱恩·巴雷特：是的。它真的像是凭空冒出来的——我想这么说，但它并非如此。它来自中国，我们不应该对中国能做出如此出色的工作感到惊讶。

佐伊·希弗：而且它确实产生了市场影响，对吧？投资者有点被吓到了。

布莱恩·巴雷特：是的，特别是英伟达，它目前确实是AI行业的风向标。深度求索发布后，英伟达在1月27日市值蒸发了近6000亿美元，这是历史上单只股票最大的单日跌幅。

佐伊·希弗：然后英伟达就完了。

布莱恩·巴雷特：是啊，然后我们就再也没听到他们的消息了。

佐伊·希弗：不，他们挺好的。

布莱恩·巴雷特：他们做得很好。不，但这首先说明了这些天价股票有多少——我不想说“水分”——但从这里开始波动的空间有多大；其次，也说明了一个模型发布的影响力。事实上，一个模型发布——市面上有几十个模型——一个来自中国的模型能产生如此大的影响，确实说明了大家认为这件事有多重大。而他们是对的，这确实重大。对我来说，深度求索R1最大的特点是其开放性。它是一个开源模型，任何人都可以使用。在此之前，基本上只有美国，只有Meta在大规模推行这种策略。

现在突然出现了一个我认为与Llama竞争的模型。Llama也有点被甩在后面了。所以现在的情况是，中国在任何人可以使用的开源模型领域真正处于领先地位，这将是一件大事，因为如果在付费和免费之间选择，很多人会选择免费版本，而这些模型将在未来两三年、五年、十年内深刻影响人们使用AI的方式。我认为我们将看到中国模型在这方面产生很大影响。

佐伊·希弗：是的。在我们深入讨论之前，应该说明一下，“开源权重”意味着模型的权重参数是公开的。所以任何人都可以在个人设备上下载该模型并进行修改。ChatGPT你无法做到这一点，但深度求索可以。你可以了解它的工作原理，并按照自己的喜好进行调整。这对AI公司有吸引力的原因是，与其仅仅让你那300名研究人员和开发人员研究模型，不如发布一个开源权重模型，这样全世界所有摆弄这个模型的人都有可能做出改进，然后你可以采纳、整合并改进你自己的模型。这样，你理论上就能接触到比你现有团队大得多的研究社区。这对中国来说是一个巨大的优势，因为他们正在大力推进开源AI、开源权重AI，这使他们能够非常非常快速地进步。

而在美国，我们实际上变得更封闭了。即使是Meta，正如你所说，作为首批构建先进开源AI的公司之一，也已经暗示他们的下一个系列模型很可能是专有的。因此，除了其他原因，人们觉得这是一种战略劣势。而且，我们正在重复相同的训练过程，就像我们使用所有这些能源、资源、研究和算力，本质上在做完全相同的事情。每个实验室都必须重复完全相同的过程，而不是在另一个实验室的见解和创新基础上进行构建。

布莱恩·巴雷特：佐伊，你对深度求索在中国构建有什么看法？有时中国模型会受到某些审查限制。我认为深度求索和其他中国模型一样遇到过这种情况。这会限制它的潜在优势吗？还是人们根本不太在意这个？我怀疑是后者，但是——

佐伊·希弗：我很好奇《连线》非常出色的AI记者威尔·奈特会怎么说。但到目前为止，在我与他的交谈中，我得到的印象是人们并不在意。即使是那些在私下里鼓吹美国与中国竞争的美国公司，似乎也在使用深度求索。如果没有其他原因，那显然是因为这样做便宜得多，而且它非常先进，能力也很强。我还认为有一个有趣的动态：关于如何处理出口管制的辩论一直在政治领域进行。我们是切断中国获取先进GPU和芯片的渠道以减缓他们的进步，还是让他们获得这些GPU，从而希望他们依赖美国芯片？我认为当深度求索出现时，感觉这几乎是一个信号，表明切断他们的渠道并不是一个好主意，因为看，这反而刺激他们在所有其他方面取得进步，因为深度求索是以一种非常廉价高效的方式训练的。

现在我们已经看到特朗普政府说：“好吧，等等，他们可以获得某些尖端芯片。”实际上中国方面也在说：“如果你是运营在这个国家的公司，我们不希望你使用那些美国芯片。”他们正试图将中国AI与中国硬件更紧密地绑定。接下来我们谈谈AI之外的话题。今年定义我们杂志的另一个趋势是所谓的“政府效能部”（DOGE）的创建和运作。该从何说起呢，布莱恩？

布莱恩·巴雷特：是啊。

佐伊·希弗：这是一个持续不断的故事，而且理由充分。我们最近了解到，该组织的成员仍在工作，我们相信他们很大程度上在联邦政府内无人监督地活动。所以我觉得现在正是评估为什么DOGE今年仍然如此重要的好时机。

布莱恩·巴雷特：就在这周，我回顾了我们早期的一些DOGE报道，提醒自己那几个月是多么疯狂。所以给那些上半年在沉睡中度过的朋友们提个醒——第一，我嫉妒你们；第二，我尊重你们的选择。

佐伊·希弗：我正想说，你真行。

布莱恩·巴雷特：政府效能部是在埃隆·马斯克和唐纳德·特朗普联手后成立的，基本上特朗普给了埃隆·马斯克在联邦政府内自由行事的权力——我在这里并没有太夸张。于是，马斯克的盟友接管了各个政府机构，包括人事管理办公室（相当于整个政府的人力资源部门）和总务管理局（基本上是政府的科技IT部门）。然后他们以此为基础，扩展到各个机构，并为我们在这届政府早期看到的许多混乱负责：大规模裁员、大幅削减美国国际开发署的预算、削减法规（并不总是好的）、要求每个联邦政府员工每周写一封包含五个要点的电子邮件汇报工作并发送出去却无人阅读。DOGE最终并没有实现其既定目标。

他们的想法是从预算中削减一万亿美元，但这实际上做不到，除非削减福利项目，而第一，DOGE对此没有控制权；第二，这在政治上是行不通的。那么，佐伊，他们到底做了什么？

佐伊·希弗：是的，我认为表面上的目标是根除欺诈、浪费和低效。

布莱恩·巴雷特：还有滥用，我想是欺诈、浪费和滥用。

佐伊·希弗：欺诈、浪费和滥用。我正想说，但我觉得我甚至不太清楚那具体意味着什么。

布莱恩·巴雷特：他们也不清楚。没关系。

佐伊·希弗：表面听起来不错，但我也觉得，这很快让人感觉像是埃隆·马斯克的一个政治项目。我想他也差不多这么说过。有趣的是，我记得当首次发布命名和确立DOGE的行政命令时，他们之前的谈论方式让人觉得这会是件大事。然后行政命令中的措辞却有点像“改善政府IT并使其现代化”之类的。我甚至记不清了，就是那种“好吧”的感觉。但后来当埃隆和DOGE相关人员再次谈论它时，他们使用的语言类似于“我们的工作是执行总统的意志”。当你这样定义自己的工作时，真的感觉没有什么事是你不能做的。

我认为影响首先波及了联邦工作人员，我们对此进行了细致的记录。现在我认为我们开始看到涓滴效应，比如疾控中心目前大约有四分之一的人离职了。我想大约有30万联邦工作人员离开了政府。美国国际开发署的关闭据称导致了估计数十万人的死亡。所以我认为我们将继续看到连锁反应。这让我想到一个稍微轻松一点的话题，那就是埃隆·马斯克最近在凯蒂·米勒的播客上谈论DOGE。斯蒂芬·米勒在特朗普政府中地位很高，显然，凯蒂·米勒曾为埃隆及其公司以及政府工作过。她在政府工作时基本上充当了DOGE的通讯人员。她现在有一个播客。她最近采访了埃隆，问他：“你会重来一次吗？你认为它成功吗？”

他有点支支吾吾，但最终我从中得到的信息是，埃隆说：“我本应该专注于我的公司。在某种程度上，我本应该只专注于我的公司。”这感觉像是一种默认的承认，即DOGE并不成功，我认为这相当明显，但我也认为事情没那么简单。当然，他们削减的数额远未达到目标，但我确实认为他们深刻地改变了联邦政府。

布莱恩·巴雷特：是的。不仅如此，我认为埃隆·马斯克的个人品牌也受到了相当大的打击。如果你销售电动汽车，其市场大概是那些更偏左、更关心环境的人，而你却成了这种破坏联邦政府的煽动性代理人，这中间存在距离。特斯拉在欧洲的销量暴跌。在美国它们还算撑得住，但他甚至不得不转向机器人出租车和人形机器人作为未来。我认为正如你所说，他们改变了政府，而其影响，我认为今天在2025年另一个不幸的重大主题——移民和移民打击——中感受最为明显。至少从夏天开始，DOGE一直在进行的一个大项目是整合美国政府各部门的不同数据块，无论是社会保障数据、税务数据还是国土安全数据，进行交叉比对、汇集，而这些数据历史上一直是分开保存的，然后利用所有这些综合信息来监视和追踪移民。

明确地说，这就是目的，并且我认为这真正改变了：第一，每个人的数据在美国政府中的保存方式，因为现在每个人的数据都以从未有过的方式混杂在一起；第二，这确实给了ICE（移民和海关执法局）一笔巨大的数据财富，再次强调，他们本不该拥有、历史上也从未拥有过这些数据，这给了他们工具来真正推动他们的任务。所以我认为这是一个我们将无限期感受到的后果。一旦你把所有那些数据混在一起，就无法再分开了。这就是我们现在所处的世界。谢谢你，DOGE。

佐伊·希弗：是的，那是另一个发布的行政命令。我想它的标题就是关于消除数据孤岛之类的。另一件表面听起来不错的事情，数据孤岛，低效，听起来不好，但将这些信息分开保存有很好的

英文来源：

For better or for worse, this year had it all—from the AI industry shaping the global economy and our lives, to the so-called Department of Government Efficiency taking over US federal agencies under Elon Musk’s leadership. In today’s episode, host Zoë Schiffer and executive editor Brian Barrett get together to reflect on some of this year’s key moments—and how they give us important clues as to what we can expect this upcoming year.
Articles mentioned in this episode:

• The AI Data Center Boom Is Warping the US Economy
• The US Needs an Open Source AI Intervention to Beat China
• DOGE Is the Deep State
• ICE Wants to Build Out a 24/7 Social Media Surveillance Team
• The FBI's Jeffrey Epstein Prison Video Had Nearly 3 Minutes Cut Out
  You can follow Zoë Schiffer on Bluesky at @zoeschiffer and Brian Barrett on Bluesky at @brbarrett. Write to us at uncannyvalley@wired.com.
  How to Listen
  You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how:
  If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for “uncanny valley.” We’re on Spotify too.
  Transcript
  Note: This is an automated transcript, which may contain errors.
  Zoë Schiffer: Welcome to WIRED's Uncanny Valley. I'm Zoë Schiffer, WIRED's director of business and industry. Today on the show, we're wrapping up our news episode series by reflecting on the trends and stories that shaped 2025. And who better to do that with than Brian Barrett, our executive editor who works tirelessly in the shadows?
  Brian Barrett: Zoë, thank you. Thank you for having me. Happy to emerge from my shadowy lair.
  Zoë Schiffer: From the dark, dark cave. Thank you.
  Brian Barrett: What a year it's been, and I'm so excited for it to be almost over.
  Zoë Schiffer: Oh, my gosh. Me too. OK. Because it's been quite a year news-wise, safe to say, especially in tech and politics. Honestly, it was a little bit tricky to pick which trends we should discuss today, but we settled on five stories that kind of encapsulate this year pretty well, and I think give us clues as to what is going to be unfolding in 2026. The first one that I want to talk about is dear to my heart, and it's about AI data centers. So we all know that the investment, the amount of money being spent on data centers is absolutely staggering with companies like Meta, Google, and Microsoft tripling down on AI infrastructure spending this year. But it's not just about the money that's being spent. It's also about how that money is being invested and the domino effect that it's already having on the rest of the tech industry and honestly our entire economy.
  Brian Barrett: I'm glad we're talking about data centers because they sound boring, they should be boring. In a just world they would be, but they're at the center of so much that is going on right now. They are sort of single-handedly propping up the economy in a lot of ways. They are responsible for so much environmental upsetment. They are driving up energy prices for people, and they are powering AI experiences that are sometimes really great and sometimes really not, depending on who's using them and how. If you had said five years ago that the biggest story in tech would be data centers, I don't think anybody would've believed it.
  Zoë Schiffer: I mean, if you had told me that we were going to have multiple reporters whose beat was directly related to data centers, I would've said absolutely boring. Skip. Next. But we're always looking for points of agreement with Peter Thiel, and I think this is one of them. Peter Thiel says—he says that the US really doesn't have another big moonshot project. We don't have a Manhattan Project right now. All we have is artificial intelligence. And I think when he's talking about this, he's saying it as a bad thing. We should have other big initiatives that we're doing. But I think even more specifically within AI, it's like it's AI data centers. And we're hearing a lot of people in this industry, Sam Altman, other kind of executives talk about the fact that it really looks like there's an AI bubble that is forming and expanding possibly.
  And the data centers are right at the center of that. I think that a lot of these financial deals are being set up through special purpose vehicles. So the money that is being spent isn't directly on these companies' balance sheets. It's also true that 60 percent of the cost of building a data center is basically just on the GPUs and you need to replace those GPUs every three years. So I think a lot of people are looking at this and getting pretty worried saying, "The math doesn't look like it's going to work out."
  Brian Barrett: Yeah. So Michael Burry, who made a name for himself around the housing crisis, right? He was the center of The Big Short. He has made a decently big bet exactly that, that the accounting here is kind of funky, that it is a bubble, as you said, and it's going to burst. Now, he's made bets before that have not panned out.
  Zoë Schiffer: I was going to say.
  Brian Barrett: He is not infallible, but his argument is indicative of the argument that you see, and he is someone who has caught a bubble before. Is 2026, Zoë, the year that this kind of comes to a head, even Sam Altman says it's probably a bubble, there's going to be winners and losers. He's obviously betting that he's going to be to the AI bubble, what Google was to the internet bubble. It made it through, it survived, it grew, it dominated the market for years. Do you think that's where we're heading 26 or do we have a little bit more time than that?
  Zoë Schiffer: I actually think we have another good year, good depending on who you are in this space. But I would guess, and this is based mostly on vibes, so take it with a giant grain of salt, but that we have until 2027 until this really starts to come to a head. And one thing that I'm going to be looking at specifically is, I've visited some of these data centers. I've talked to politicians who are really supportive of having data centers in their city and right now it looks like there's not a lot of cost politically speaking in a lot of places. Obviously there is a lot of local pushback, but in a lot of places, these data centers are being built in very red states. There's not a lot of job opportunity. And so they're being heralded as kind of like, "Look, this economic boon that's coming, it's going to be really great for the local economy. There's going to be all these jobs."
  But the truth is that, well, one, you need thousands of people to build a data center, but you need far fewer to maintain a data center. So once it's already up and running, those jobs, a lot of them will evaporate. Two, they are very, very resource intensive. It takes a lot of water to cool a data center. It obviously uses up an absolute ton of energy. And so I think we're going to see a switch. I mean, this is just a prediction, but where it's going to be politically untenable to support a data center and we're going to have an offshoring push in the next three years.
  Brian Barrett: And even those thousands of jobs of people who build the data center aren't always local because a lot of these are specialized people come in. I live in a red state and there is pushback. They're trying to build a data center near me. It's not enough to actually do anything. The thing's going to get built. But yeah, I do think as people start to put the piece together of, "Oh, my energy bill's higher now. Oh, the jobs aren't here that I though we're going to be." I agree. I think we're going to see pushback. And that's when we're going to build data centers in space.
  Zoë Schiffer: I was literally just going to say on Mars. We don't have a data center in Santa Barbara where I currently live, but we do live very close to a SpaceX rocket launch site. And for a really politically unengaged community, wow, do people rally against those rocket launches because it freaks out their dogs and horses and such. And so I would just love to see what happens if they try and build a data center here.
  Brian Barrett: Amazing.
  Zoë Schiffer: OK. So I feel like we would be remiss not to talk about what these data centers are being built for because it's chatbots, it's ChatGPT, they're being built for inference, meaning not just to train the models, but actually to support millions and millions of people asking chatbots questions and wanting to get a result. One trend that I feel like we really saw pretty acutely this year was the rise of chatbot companions and AI relationships. I'm curious how you feel about that now and how has that feeling, I guess, changed since the start of the year?
  Brian Barrett: It's interesting. I think on the one hand, you've got so many individual incidents where things have gone really wrong with chatbot companions. You've got interactions allegedly leading to suicide or contributing to suicide. You've got some seemingly pretty unhealthy relationships, but at the same time, you also have cases where people seem genuinely happier. They seem like they are filling a void. I don't know. I, personally, it's one of those things where I have been sort of reflexively like, "Oh no, let's not do that." But I recognize that I need to come around like, "Well, it's not for me, but that doesn't mean it's not a place." I think really, most of all, it's still so early in all of this. And I think there needs to be so much more work done to figure out what these relationships are doing to people for good or bad or neither.
  And I wish that some of that work had been done on the front end by AI companies themselves before just sort of saying, "Go have fun. Go have fun with your new AI boyfriend or girlfriend," without really understanding what the consequences are because no one really knows yet. It's too early. So I'm glad to see there are more safeguards in place on a lot of levels from a lot of these places, but it does seem too much too fast. Let's figure out what's going on.
  Zoë Schiffer: So this is something I've been thinking about with mania or what's being called AI psychosis because I think a lot of the time, at least in my experience as a tech reporter, it's like the issue at hand, whether it was misinformation or what have you, there was a technical cause to that, but it felt like it needed a multi-pronged solution and we were really just looking at tech companies to fix this entire issue that was impacting our democracy. And with AI psychosis or AI mania, I don't know, that really does feel like a tech issue. If the chatbot is telling you that you've discovered some new frontier of physics or whatever, or if it's validating you again and again, and you believe it, that seems like a problem with the chatbot. With AI relationships, I wonder if the issue when it is an issue is more complex.
  If there's something else that's going on kind of at a societal level and that chatbot relationships are a symptom of that rather than a cause, I do think we need more safeguards in place. Absolutely. I think also a lot of other things need to be done at the societal level so that we have more connections with people. It's easier to make those connections. There are more things pushing us to be in community with one another.
  Brian Barrett: Zoë, we're going to bring back bowling leagues. We're going to bring back bowling leagues. What could be better?
  Zoë Schiffer: So another trend that we followed really closely this year at WIRED is the global competition in making frontier AI models. So we will get into stories and trends that are not centered on AI, I promise, but it would be disingenuous to pretend that AI hasn't been the defining story of our industry this year. And one of the moments in which that became really clear happened pretty early on. If you remember back in January when the Chinese AI research lab, DeepSeek released the R1 open model, and it felt like all hell broke loose.
  Brian Barrett: Yeah. It really came out of, I want to say it came out of nowhere, but it didn't. It came out of China and we shouldn't be surprised that China's doing really great work like this.
  Zoë Schiffer: And there was an actual market impact, right? Investors got kind of freaked out.
  Brian Barrett: Yeah, especially, I mean, Nvidia is really the bellwether for the AI industry at this point. After DeepSeek came out, it lost nearly $600 billion in market cap on January 27. It is the largest single day loss for a single stock in history.
  Zoë Schiffer: And that was the end of Nvidia.
  Brian Barrett: Yeah. And then no one ever heard from them again.
  Zoë Schiffer: No, they were covered just fine.
  Brian Barrett: They're doing great. No, but that is A, tells you I think how, I don't want to say inflated, but how much room there is for these sky-high stocks to sort of vacillate from here, but B, the impact of just a model release. The fact that one model release, there are dozens of models out there. The fact that one from China can have that big of an impact really tells you how big a deal everyone thought this was. And they're right, it is. The big thing about DeepSeek R1 to me was the openness of it. It's an open-way model, anyone can use it. Really only US, only Meta had really been pursuing that strategy on a pretty big scale.
  Now all of a sudden you've got a model that I think was competitive with Llama. Llama also kind of fell by the wayside. So now you're in a world where China is really leading on these open way models that anyone can use, which is going to be a pretty big deal because if you have a choice between paying for something or it being free, a lot of people are going to go for the free version and these models are going to really inform a lot of how people are using AI in the next two, three, five, 10 years. And I think we're going to see a lot of influence from Chinese models doing that.
  Zoë Schiffer: Yeah. And we should say before we get into this more that open weight means that the weights of the model are published. So anyone can download the model on a personal device and they can modify it. You can't really do that with ChatGPT, but you could do it with DeepSeek. You could get an understanding of how it works and you could tailor it to your liking. The reason that this is attractive to AI firms is that instead of just having your, let's say, 300 researchers and developers working on the model, you release an open weight model and all of a sudden everyone in the world who's tinkering with that model could be making improvements that you can then take, co-op and improve the model yourself. So you get access theoretically to a research community that's much, much bigger than the one that you have. And this is a huge advantage for China because they're really going hard on open source AI, open weight AI, and it's allowing them to advance really, really rapidly.
  Whereas in the US, we've really become more closed source. Even Meta, like you said, one of the first firms to build advanced open source AI has signaled that their next series of models will likely be proprietary. And so among other things, people feel like it's a strategic disadvantage. Also, we're repeating the same training runs, like we're using all of this energy and resources and research and compute to essentially do the exact same thing. Each lab is having to repeat the exact same process versus building on the insights and innovations of another lab.
  Brian Barrett: Zoë, what do you think about DeepSeek being built in China? Sometimes Chinese models are subject to certain censorship situations. I think DeepSeek has run into that as of other Chinese models. Does that limit its potential upside or are people just kind of not going to care so much about that? I suspect the latter, but—
  Zoë Schiffer: I would be curious what Will Knight, one of our really amazing AI reporters at WIRED would say about this. But so far in my conversations with him, the sense that I get is that people don't care. And even US firms that have been kind of championing the US race against China behind closed doors appear to be using DeepSeek. If for no other reason then it's obviously just a lot cheaper to do, and it's really advanced and its capabilities are good. I also think there's this interesting dynamic where this debate has been playing out in politics around how to handle export controls. Do we cut off China's access to advanced GPUs and chips and so try and slow their progress or do we give them access to these GPUs and then hopefully make them dependent on US chips? And I think when DeepSeek came out, it felt like almost a signal that cutting them off was not a good move because look, it was spurring them to advance in all of these other ways because DeepSeek was trained in a really cheap and efficient way.
  Now we've seen the Trump administration say, "OK, wait, they can get access to certain cutting edge chips." And actually China is coming in and saying, "Well, if you're a company operating in this country, we don't want you to be using those American chips." They're trying to tie Chinese AI more closely to Chinese hardware. Moving on from AI. So the next trend that really defined the magazine this year was the creation and the workings of the so called Department of Government Efficiency or DOGE. Where to even begin, Brian?
  Brian Barrett: Yeah.
  Zoë Schiffer: This is the story that kept on giving and for good reason. We recently learned that members of the group are still working, largely we believe unsupervised across the federal government. So I feel like it's as good a time as any to take stock as to why DOGE remained so important this year.
  Brian Barrett: I just this week was looking back at some of our earliest DOGE reporting and reminded of what a crazy couple of months that was. So just as a reminder for folks who were slumbering through the first half of the year, A, I'm jealous. I respect that.
  Zoë Schiffer: I was just going to say, good for you.
  Brian Barrett: Department of Government Efficiency came about when Elon Musk and Donald Trump got together and basically Trump gave Elon Musk kind of free rein to do whatever he wanted, and I'm not really exaggerating here, within the federal government. So Musk allies took over various government agencies, including the Office of Personnel Management, which is sort of the human resources for the whole government, the General Services Administration, which is its tech IT department basically. And from there, kind of fanned out across agencies and were responsible for a lot of the chaos that we saw in this early administration, massive job cuts, massive cuts to USAID, regulations being slashed, not always for good, having everyone in the federal government, having to write an email with five points of what they did that week and sending it to never be read. DOGE didn't end up doing what it set out to do.
  The idea was to cut a trillion dollars from the budget, which you literally can't do unless you cut into entitlement programs, which A, DOGE had no control over, and B, politically would be untenable. So Zoë, what did they do?
  Zoë Schiffer: Yeah, I think the goal ostensibly was to root out fraud and waste, to root out inefficiency.
  Brian Barrett: And abuse, fraud, waste and abuse, I think.
  Zoë Schiffer: Fraud, waste and abuse. I was going to say it and I was like, I don't even know exactly what that means,
  Brian Barrett: But they didn't either. It's fine.
  Zoë Schiffer: It sounded good on the surface, but it also, I think, very quickly felt like a political project for Elon Musk. And I think he said as much. It was interesting. I remember this moment when the executive order codifying and naming DOGE first came out and the way that they had been talking about it previously, it felt like it was going to be this big enormous thing. And then the language in the EEO was kind of like improve the government IT and modernize it. I don't even remember. It was very like, OK. But then when Elon talked about it subsequently and when DOGE affiliates talked about it subsequently, the language they used was something akin to like, "It is our job to enforce the will of the president." And when that's how you conceive of your job, it really does feel like there's nothing you can't do.
  I think the impact hit federal workers first and we documented that pretty meticulously. Now I think we're kind of seeing the trickle-down effects, like a quarter of the CDC is gone at this point. I think around 300,000 federal workers are no longer in the government. USAID shutting down has led to an estimated hundreds of thousands of deaths reportedly. So I think we're going to keep seeing the ripple effect. It does bring me to a slightly lighter note, which was Elon Musk talking about Doge on Katie Miller's podcast recently. Stephen Miller, who's quite high up in the Trump administration, obviously, Katie Miller has worked for Elon and his companies as well as in government. And she kind of operated as the Doge comms person when she was working in government. She now has a podcast. She sat down with Elon recently and she asked him, "Would you do it all again? And do you think it was successful?"
  And he kind of hems and haws, but ultimately what I got from it was Elon saying, "I would've worked at my companies. In some ways, I should have just focused on my companies." It felt like a tacit admission that Doge was not successful, which I think feels fairly obvious, but I also think it's not as simple as that. Of course, they didn't cut anywhere close to what they wanted to, but I do think they changed the federal government pretty profoundly.
  Brian Barrett: Yeah. Well, and not only that, I mean, Elon Musk's personal brand I think took a pretty big hit. If you sell EVs whose market is presumably people who lean a little bit more left and care more about the environment, and you become this sort of demagogical agent of destruction to the federal government, there's a distance there. Tesla sales have plummeted in Europe. They're holding out OK in the US, but he's even had to switched over to robo taxis and humanoid robots as the future. And I think to your point, those changed the government and the effects are, I think today, most acutely being felt in another unfortunately big theme of 2025, which is immigration and the immigration crackdown. A big DOGE project that has been going on since at least the summer is to combine different pockets of data from throughout the US government, whether that's Social Security data, tax data, homeland security data, cross reference it, pool it when it has always historically been kept separate and use all of that sort of combined information to surveil and track down immigrants.
  Explicitly, that's the purpose and has really transformed, I think, A, how everyone's data is held in the US government, because everybody's data is now intermingled in ways that it was never supposed to be. And B has really given ICE an incredible wealth of data that again, they should not have, historically would not have had and has given them tools to really fuel their mission here. So I think that is a consequence that we're going to be feeling indefinitely. Once you comingle all that, you can't separate it out again. So that's the world we're in now. Thank you, DOGE.
  Zoë Schiffer: Yeah, that was another executive order that came out. I think it was literally titled something around eliminating data silos. Another thing that on its surface sounds good, data silos, inefficient, sounds bad, but there's a really good reason to keep a lot of this information separate. You don't necessarily want the federal government to be able to knit all of that together to track you across any platform to know your financial data, your health data, your whereabouts, all of that.
  Brian Barrett: Yeah. I mean, the Trump administration now wants to vet, what is it, five years of social media if you're coming into the country just to see what you've been saying. And also they're looking to strip more foreign-born Americans of citizenship. They're moving towards denaturalization. The predictions for 2026 and immigration aren't great. It's that it's going to get worse and more. Or let's say if you were an engineer born in a different country and you had an option of going to the US or Canada or China or somewhere else in Europe, it's a lot harder question to answer now than it was before all this happened, I think.
  Zoë Schiffer: Yeah, no, I think that that's exactly right. And we're going to have, I think, some really good reporting on that in 2026. So we're nearing the end of our 2025 recap and we would be remiss not to mention the Jeffrey Epstein saga. So this is a story that has had so many lives across many, many years, but it felt like it really came to a boiling point this year, politically speaking. So Brian, I guess, can you just refresh our memory on what even happened this year?
  Brian Barrett: I get the easy job. Without going into the whole Jeffrey Epstein everything for so many reasons. Obviously he's a disgraced financier, convicted sex offender, also the locus of so many conspiracy theories from QAnon up to the White House and its supporters. What we saw this year was a real movement towards releasing the so called Epstein files, which the White House Donald Trump basically ran on. He said, "Look, if I get in that office, I'm going to release the Epstein files. You're going to have them right as soon as I get there." And then we got to that point and he got there and he said, "You know what? Nevermind." And there was people saying there are no Epstein files. Actually, the Epstein files are a hoax. Actually, no, they are here. It has been a mess.
  The upshot of all this, maybe by the time this podcast airs, the Epstein files will be released or maybe sometime after there's a deadline for Friday, December 19th that has been set by Congress. So we'll see what happens what's actually in there. Personally, I think it's just going to add more fuel to various conspiratorial fires. I think the people who are fans of Donald Trump are going to find ways to decide that, oh no, he was actually an FBI informant the whole time. I think people who are not are going to have a more rational and sane view of whatever it is. I understand why people want to know what's in there. I'm also excited for everyone to get it and then move on.
  Zoë Schiffer: I love that you think we're going to move on. I don't think that's going to happen.
  Brian Barrett: Oh, I just said I'd love to. I said I'd love to. I didn't say we would. I said it's a beautiful dream.
  Zoë Schiffer: Yes, we would love to. We would love to. I know you don't like to speculate, Brian. Obviously you're a reporter at heart, but I am genuinely curious, what do you think was going on here with Trump? Do you think that he was saying what he needed to say on the campaign trial and just wasn't thinking too far ahead in terms of what happens if I win and then people still want these?
  Brian Barrett: So are you suggesting that our President Donald Trump did not think through consequences?
  Zoë Schiffer: I would never say that. I mean, explicitly outright. I'm just genuinely curious what was happening. What did you think was going to happen?
  Brian Barrett: I think that he says what he needs to say in the moment and he knows that that was a thing that got his base stirred up and then the check came due and it's actually caused real problems, which is rare for him because usually he's able to sort of work around it. But it appears to have basically fractured his relationship with Marjorie Taylor Green, who was one of his biggest supporters, not biggest supporter in Congress, but this was a real dividing line for her. Again, people will justify what Donald Trump says and does all day long, but I do think there are a significant number of MAGA adherence for whom this is a real wedge issue, for whom this is going to be depending on what happens when the files get released and the aftermath of that could actually impact the midterms next year.
  Zoë Schiffer: OK. We have to talk about the video. We can't not.
  Brian Barrett: Yes.
  Zoë Schiffer: Pam Bondi, Trump, they had talked about releasing the files and obviously then they win the election, they're in office. It appears they do not want to release it files, but they do release a video of Jeffrey Epstein's final moments in jail. And then, I mean, I hate to pat ourselves on the back, but we become pretty central to the story, Brian. No?
  Brian Barrett: We do. We broke the story. So the DOJ published a video. And the weirdest part about this was they said, "Look, here's the unedited video." They went out of their way to say that this is an unedited video. And so we took a look at it and it took us about 30 seconds to see that, oh no, actually it was edited. It appears to have been actually two different videos that were pieced together to form one video. There's about two and a half minutes that are missing. I don't want to say they're unaccounted for necessarily. What's weird about this is a lot of things could potentially explain it. I don't want to get too conspiratorial about it because it could be the kind of thing where, oh, the system changes over tapes at midnight and that's around when this happens. So we took the first tape and then we took the tape when it changed over and there's some overlap. There's a lot of technical explanations that could make this make sense, but at the end of the day, no one has given us those explanations.
  They went out of their way to misrepresent what they had released and there's two and a half minutes that they didn't include. Again, I'm not saying there's a conspiracy here. I am saying that they are presenting conspiracy vibes. They are doing things that one would do if you were in the middle of a big conspiracy. I think there's probably a rational explanation for it. It's more the way that they have dealt with it is only fueling this fire.
  Zoë Schiffer: Right. To quote Elon Musk, sunlight is the best disinfectant.
  Brian Barrett: And so it's again, why I work in the shadows.
  Zoë Schiffer: Exactly. Some of the files were already approved to be released by Congress last month. So I think that was roughly 20,000 documents and they're already kind of being dismissed by the MAGA base.
  Brian Barrett: Yeah. There's a lot of people who are just ignoring them or justifying them. A lot of the Democrats in Congress have been releasing images of Jeffrey Epstein's Island, a bunch of email exchanges and text message exchanges. Some of them mentioned Donald Trump explicitly. Larry Summers is in there, Bill Gates is in there a lot, but nothing has stuck. And I think Jeffrey Epstein, horrible guys seemingly smart enough not to literally write down, "Hey, do you want to do a sex crime with me?"
  So what you have is a lot of connections and innuendo and insinuations. I think where we get to an interesting point is in this new batch, if there are sort of more concrete financial relationships because the numbers don't really lie. If there is more concrete video, photographic evidence, what is in there that would actually get people to be on board with this? I don't know if there's anything. I think that for MAGA diehards, I think that you get to a point where it's like, "Well, that was just forged." That's the deep state. That's AI, to continue our year of AI. So I don't think anything is going to really turn the tide on it.
  Zoë Schiffer: Yeah, I think that's right. I feel like we don't have a shared understanding of the truth anymore. And so this is really catering to people who for the most part already have some level of belief or don't like Donald Trump or don't like the people who are being named in these files. And then there are other people who it really doesn't feel like there's anything. I will say, he's not literally explicitly talking about crimes in the emails we've seen so far, but he named his plane, The Lolita. I'm like, "This man wasn't hiding."
  Brian Barrett: No, absolutely. And the photos are creepy, if nothing else, right? Appropriate for 2025. I feel like it's the right note to go out on for this year.
  Zoë Schiffer: I know. I was going to say, I wish we could end on a different note, but I feel like this was this year. This year had a lot in it and boy, was it tough? We are ready to say goodbye. So Brian, thank you so much for joining me today.
  Brian Barrett: Zoë, thank you for having me.
  Zoë Schiffer: That's our show for today. We'll link to all the stories we spoke about in the show notes. Make sure to check out Thursday's episode of Uncanny Valley, where you'll get to hear some news from Mike and Lauren about the show. I'll leave you with a hint that you'll be hearing a lot more from me, Brian, and our colleague, Leah Feiger next year.
  Brian Barrett: Which feels like more than a hint that feels, but—
  Zoë Schiffer: A little spoiler.
  Brian Barrett: Little spoiler.
  Zoë Schiffer: A little tease.
  Brian Barrett: Tiny smile.
  Zoë Schiffer: Adriana Tapia and Mark Leyda produced this episode. Amar Lal at Macro Sound mixed this episode. Kate Osborn is our executive producer and Katie Drummond is WIRED's global editorial director.