内容来源：https://www.wired.com/story/uncanny-valley-podcast-4-things-you-need-for-a-tech-bubble/

内容总结：

近期，关于“人工智能泡沫”的讨论甚嚣尘上。谷歌、Meta和微软等科技巨头已宣布将加码对2026年人工智能领域的投资。在此背景下，科技媒体WIRED的播客节目《诡异谷》邀请专栏作家、《血洗机器》作者布莱恩·麦钱特，通过历史经验解析当前AI领域是否存在泡沫风险。

麦钱特引用学者戈德法布与基尔什的研究指出，判定技术泡沫需考察四大特征：首先是技术前景的不确定性。以电力发明为例，虽然其颠覆性显而易见，但商业化路径探索耗费数十年，当前AI技术同样面临应用场景模糊的困境。

其次是单一业务企业激增。这类企业将命运完全押注于某项技术，例如为AI提供算力服务的CoreWeave公司。值得注意的是，英伟达已从游戏芯片供应商转型为AI核心供应商，其市值占美股总市值约8%，形成“卖铲子的淘金者”效应。

第三是缺乏经验的散户涌入。当下通过股票软件投资AI概念的散户正重现互联网泡沫时期的投资热潮。更值得警惕的是，软银等机构投资者的巨额资本正通过芯片采购、数据中心建设等渠道形成循环投资网络，使普通投资者的养老基金等资产间接暴露于风险中。

最后是市场共识的集中爆发。当ChatGPT等现象级应用出现后，“AI将颠覆所有行业”的叙事促使资本盲目跟进。学者指出，当前AI领域已同时具备四大泡沫特征，在评估体系中达到最高风险等级8级。

尽管科技泡沫破裂后往往能沉淀实用技术（如互联网泡沫后宽带基础设施的留存），但专家警告本轮AI泡沫若破裂，其经济冲击可能远超互联网泡沫。与宽带等长期基础设施不同，AI芯片迭代速度极快，当前投入的硬件可能数年后即被淘汰。此外，若政府通过持有企业股份等方式介入救市，或将开创科技泡沫干预的先例。

在技术前景尚不明朗的当下，95%的企业承认其AI投入尚未产生收益。这种投入与产出的巨大落差，正为潜在的泡沫破裂敲响警钟。

中文翻译：

近来关于人工智能泡沫的讨论甚嚣尘上，谷歌、Meta和微软等科技巨头更是将2026年的AI投资额度翻倍。但历史上的分析师们是如何精准识别技术泡沫的形成？主持人迈克尔·卡洛尔与劳伦·古德邀请到《血洗机器》专栏作者、《连线》特约撰稿人布莱恩·麦钱特，共同剖析研究人员曾用于研判科技泡沫的四大标准，帮助人们未雨绸缪。

本期节目提及文章：
请填写听众调查问卷助力《诡异谷》节目改进。
您可以在Bluesky上关注迈克尔·卡洛尔（@snackfight）和劳伦·古德（@laurengoode）。欢迎来信至uncannyvalley@wired.com。

收听方式
您可通过本页音频播放器随时收听本周节目，若想免费订阅每期内容：
iPhone或iPad用户请打开播客应用，或直接点击此链接。也可下载Overcast、Pocket Casts等应用搜索"诡异谷"。节目亦在Spotify同步更新。

文字记录
注：本文为自动生成 transcript，可能存在误差。

迈克尔：劳伦，最近怎么样？
劳伦：还行，迈克。正值财报季，我们《连线》商业组的同事都在密切关注科技企业的财报和电话会议。说白了现在就是资本支出季。
迈克尔：CapEx？
劳伦：资本性支出。
迈克尔：你现在都说缩写了？
劳伦：既然成了商业线记者，就得习惯用术语。
迈克尔：你也成"那种人"了。
劳伦：难道我会在派对上拽专业术语？开玩笑的。不过我们确实注意到趋势——科技公司守着金山银山，却不是在坐吃山空。他们正宣布巨额支出计划，特别是AI基础设施领域。
迈克尔：比如数据中心。
劳伦：对，更多数据中心。虽然不限于此，但这确实是重头戏。
迈克尔：这些都在助推泡沫论调。两周前我们和莫莉·塔夫特同事简单讨论过，今天要深入探讨AI泡沫狂热。
劳伦：没错。今天的嘉宾对此颇有见解。作家布莱恩·麦钱特将加入对话。布莱恩，欢迎来到《诡异谷》。虽然话题不太令人愉快，但很高兴你能来。今天感觉如何？
布莱恩：还算不错。感谢邀请，很期待与二位深入探讨泡沫经济的方方面面。
迈克尔：这里是《连线》出品的《诡异谷》，一档聚焦硅谷人物、权力与影响力的节目。仿佛一夜之间，所有人都在讨论AI泡沫。自从今夏OpenAI的萨姆·奥尔特曼随口提及泡沫可能性，争论愈演愈烈。上周比尔·盖茨对CNBC表示确实存在泡沫，但补充说就像互联网时代，"深刻变革"终将显现。或许他指的是芯片巨头英伟达创始人将赚得盆满钵满——这家公司刚成为全球首家市值达5万亿美元的企业，相当于加拿大经济总量的2.5倍。AI投资仍在全速推进，毫无放缓迹象。科技巨头今年AI投资总额预计达4000亿美元，但这还不够。

谷歌、Meta、微软和亚马逊已承诺2026年追加投资。然而据某些报告显示，目前使用AI的企业中95%表示几乎未见回报。我们究竟是否身处泡沫？若是，该如何评估？泡沫破裂又可能引发什么后果？为厘清这些问题，我们请来布莱恩。他最近运用历史框架分析AI是否符合经典经济泡沫特征，以及这对我们意味着什么。我是迈克尔·卡洛尔，消费科技与文化版块总监。
劳伦：我是劳伦·古德，高级记者。
布莱恩：我是布莱恩·麦钱特，《血洗机器》专栏及书籍作者，《连线》特约撰稿人。
迈克尔：布莱恩，你着手研究AI是否真在催生泡沫。但首先要请你定义何为泡沫——有些纯粹派经济学家会否认任何泡沫的存在。
布莱恩：正因如此我才需要更系统的分析方法。目前关于AI的讨论多停留在直觉层面：看到英伟达市值突破5万亿——其实突破3万亿、4万亿时已令人瞠目——要知道首家万亿级公司出现才不过数年。资本积累与集中的速度近五到十年确实在急剧加速。眼见数据中心的疯狂扩张，人们直觉这是泡沫。但究竟意味着什么？最基础的理解是：当对某项技术/公司/商品的投资额远超其未来收益时就会形成泡沫。比如Pets.com这类早期互联网公司，玩具电商估值曾远超玩具反斗城，尽管营收规模天差地别。当投资血本无归，我们事后才确认这是泡沫。如今对AI的担忧在于，涌入英伟达等企业的巨额资金可能重蹈覆辙，且规模空前。
劳伦：你借鉴了两位学者布伦特·戈德法布与大卫基尔希关于泡沫的著作，并将其框架应用于AI。请谈谈这个分析框架。
布莱恩：说实话，鉴于当前泡沫规模，我很惊讶这本书未被更多讨论。2019年他们出版《泡沫与崩溃：技术创新的繁荣与萧条》，回溯梳理历史上众多科技泡沫，总结其共性要素：何种因素促使繁荣演变为泡沫，又是什么阻止了某些投资热潮泡沫化。他们提出构成科技泡沫的四大要素：首先是创新不确定性——这项技术能否盈利？商业模式何在？新技术能否带来收益是否明确？第二是纯游戏投资数量，即与创新技术商业成败深度绑定的企业数量。以当今的CoreWeave为例，其商业模式完全依赖芯片租赁与云计算服务，若AI泡沫破裂这类企业将瞬间崩塌。第三是新手投资者，即普通散户需要有投资渠道涌入新领域。比如现在用Robinhood应用买英伟达股票的人。历史上互联网泡沫就是典型——散户读了几篇《连线》文章就坚信互联网是未来，盲目投资上市企业却不懂基本面。最后是戈德法布所说的"信念协同"，即投资者对某项创新代表未来的集体共识。通常这需要现实用例支撑，例如ChatGPT的病毒式传播自然展现了用户兴趣。过去查尔斯·林德伯格跨大西洋飞行成功，在当时堪称史上最震撼技术演示，引发航空投资热潮最终形成泡沫。总结就是：不确定性、纯游戏企业、新手投资者和信念协同。
迈克尔：林德伯格的例子很经典。书中还有不少历史案例，能否通过19世纪电力应用的故事帮助我们理解不确定性？
布莱恩：戈德法布他们详述了电力与电灯的诞生。当时人们清楚这是颠覆性创新——街道照明、城市采购的巨型电塔火花四射虽混乱却震撼。但这项技术如何盈利？主要市场是家用灯泡？市政照明设备？何种方式最具成本效益？这些花了数十年才厘清。我们总以为爱迪生发明电灯后世界瞬间改变，其实从实验室技术到商业化盈利历经数十年。
迈克尔：可见即时应用场景并不明确。原始技术令人惊叹，但转化为实际业务需要过程，这正符合当前AI的发展态势。
布莱恩：正是。广播也是典型例子——当无线电首次公开展示时，所有人都知道这必将改变什么，但究竟用于广告宣传、戏剧转播还是其他用途并不清晰。因此与航空业同期也形成著名投资泡沫。AI同样如此：我们知道聊天机器人受欢迎，但商业路径仍不清晰。
劳伦：我们讨论了不确定性。泡沫第二要素是纯游戏企业，即公司命运完全系于单一创新。但AI涉及众多环节，英伟达固然是惊人创新（虽然黄仁勋可能强调不止于此），但各类模型存在多维度商业化可能。这如何符合纯游戏定义？
布莱恩：英伟达实质上已成为纯游戏公司。关注科技领域的人都知道，在AI热潮前英伟达主要生产游戏显卡。如今它将所有业务重心转向供应AI芯片，堪称"淘金热中的卖铲人"。即便AI泡沫破裂，芯片需求依然存在，这对投资者是最具体的投资标的，且作为上市公司可直接持股。因此它基本符合纯游戏投资定义——命运与AI热潮深度绑定。有趣的是AI时代IPO并不多，CoreWeave算是知名案例。通常IPO是投资纯游戏企业的途径，但本轮AI热潮的特点是资本集中在少数企业，外加房地产公司投资数据中心等组合投资。
迈克尔：如果资金多在私募领域流转，公众未深度参与，这些巨头企业的膨胀对公众有何风险？
布莱恩：风险正在积累。这种市场集中度很不寻常——要知道英伟达约占股市总市值8%。
迈克尔：多少？
布莱恩：8%。若英伟达崩盘，大量散户将受重创。更值得警惕的是泡沫期常见的循环投资现象：投资者通过基金投资数据中心扩建，普通人可能未察觉自己的投资组合已通过房地产基金间接涉足AI热潮。又如OpenAI与AMD达成的投资协议——AMD是上市公司，公众持有其股票，而通过股权结构设计，OpenAI现持有AMD部分股权。若AI泡沫破裂，普通散户、养老基金持有者等原本分散的投资组合将面临日益增长的AI风险暴露。虽然风险程度存在争议，但这类交易和漏洞确实亮起红灯。
劳伦：框架第三要素是新手投资者。但真正令人担忧的不仅是散户通过Robinhood涉足高风险投资，更是软银、英伟达、OpenAI收购AMD股权等机构层面的大规模循环投资，这才是潜在泡沫的最大隐忧吧？
布莱恩：没错。因为这些机构操纵着天量资金。戈德法布接受采访时指出，AI充满不确定性，在某种程度上所有人都成了新手投资者——毕竟没人能预知未来。
劳伦：所以该有人提醒萨姆·奥尔特曼他也是新手投资者。
布莱恩：他或许会坦然承认。记得OpenAI早期有人问如何盈利，约五六年前他认真回答："我们先造出通用人工智能，再问它怎么赚钱。"这就是商业计划——先实现AGI再让系统自己想办法。这才叫不确定性。
劳伦：没救了，布莱恩。
迈克尔：这情节我读过，叫《银河系漫游指南》。
布莱恩：说明我们完全身处未知水域。我认为不确定性已爆表，这也是我判断本次泡沫可能空前的原因。
迈克尔：这就引向第四指标——信念协同或叙事构建。众所周知，当前AI叙事是"万能论"：自动化工作、治愈癌症、照顾儿童、应对气候变化，最终迎来媲美人类的通用人工智能。无限潜力的承诺妙在无需明确目标。
布莱恩：多完美的故事！戈德法布说这是最离谱的叙事——光有故事不够，还需要一定可行性让投资者相信"这确实可能实现"。AI公司高调发布产品、夸大宣传、用演示强化系统能力认知，形成让投资者甘之如饴的温床。"能自动化所有工作？我们等了两百年！""能帮药企输入数据治愈癌症？""能解决气候危机？"每个投资者都能找到心动理由。过去三年投资阶层确实形成了"这是真风口/值得赌上数十亿"的集体信念，既怕错过"万能机器"的投资机会，又担心踏空。
劳伦：今年我为《连线》采访AMDCEO苏姿丰时，基于共同经历（父母住ICU）认同AI在医疗健康潜力，但在其他方面存在分歧。我问及规模扩展、推理与训练、内容审核等问题，提到当社会无法辨认真假时，她说："你似乎对AI很怀疑。"我回答："最乐观的往往是最大受益者。"并引用威廉·吉布森名言："未来已至，只是分布不均。"我认为我们正走向这样的未来。
布莱恩：很犀利的观点。显然即便泡沫破裂，AI热潮也会催生赢家。可能我比你对AI商业模式更怀疑。但可以肯定，就像人们用"互联网泡沫破裂后留下宝贵遗产"来反驳泡沫担忧者，AI也可能在某种程度上如此。不过本次泡沫规模可能导致远比互联网泡沫惨烈的经济灾难，而某些持续存在的AI应用——如自动化创作——可能长期以不利于劳动者的方式重塑经济。
迈克尔：关于"泡沫是否将破裂"这个根本问题，我们稍后休息后再继续讨论。

（休息）

劳伦：欢迎回到《诡异谷》。今天我们与布莱恩探讨AI是否泡沫及破裂后果。上周英伟达CEO黄仁勋刚表示完全不认同AI泡沫论，称这是从通用计算到加速计算的"自然转型"——这套说辞他坚持已久。他否认泡沫可以理解，但答案似乎因人而异。若泡沫破裂，经济冲击可能远超25年前的互联网泡沫。但若持续，AI将根本性重塑工作生活方式。长期来看两种情形可能并存。布莱恩，上半场我们讨论了戈德法布与基尔希的四要素框架，他们还用0-8分量表评估泡沫风险。当被问及AI时，他们给出了最高的8分——买方自负。
布莱恩：是的。他们发现所有要素都已具备，且某些要素含量极高。根据这个历史验证的框架，我们处于最高级别泡沫警戒。需要说明的是，该框架不预测灾难规模，只衡量四要素的具备程度来判断是否泡沫。因此他们可能止步于分析结论，不会如我般断言这可能是"终极泡沫"。
迈克尔：唱个反调：即便这是终极泡沫，历史表明技术创新不会因经济激励减弱而消失。互联网泡沫后网络依然蓬勃发展，25年后我仍在网络媒体工作。AI产业是否会走类似道路？
布莱恩：毫无疑问。再严厉的泡沫批评者也不会认为泡沫会让AI消失。我个人认为可能性极低，但期间可能经历巨大经济阵痛。AI已证明在特定用户群中具有吸引力，科技公司也看到了产品热度。他们仍需解决每个查询都因计算资源消耗而亏本的问题，但这很可能实现。不过通过这个量表分析，在我看来投资规模明显可能超过实际效用水平。不少强硬派投资银行家也得出类似结论，狂热可能掩盖真实财报。AI持续存在是必然的。有趣的是互联网泡沫后留存的是电信基础设施——光纤网络建设让互联网持续运行。但芯片情况不同，它们需要持续升级，现在的芯片十年后可能过时。所以我认为部分要素会留存，部分企业会胜出，但我的预测能力并不比你强。
劳伦：很高兴你比较宽带设施与芯片。读你文章时我产生一个可能不太成熟的疑问：生成式AI的持久性是否会更多体现在影响力而非商业化潜力？类比历史泡沫，AI的长期存续是否更接近广播、社交媒体等内容领域——替代脑力劳动、生成内容，而非底层基础设施？你怎么看？
布莱恩：我同意这个判断。我认为这是其主要效用。有趣的是，即便最坚定的AI拥护者也承认其输出质量并未远超人类，有时甚至更差。对许多企业而言这是成本考量，因此归根结底是自动化与劳动力问题。可以分两类：聊天机器人属于社交媒体范畴，可视为人际关系的自动化；另一类是劳动自动化，人们说"能自动化团队工作流程/降低人力成本"。但宣传口径很快变成"全能自动化"。我认为这是泡沫典型特征——宣传承诺与实际落地存在差距。听众可能记得最近这轮泡沫讨论始于MIT研究：投资生成式AI作为自动化工具的企业中，95%尚未获得回报或利润。这个数字引发了对实际成效的质疑。所以我倾向于认同你的观点，仍怀疑AI是否足够可靠，能成为日常数字生活的基础设施。我更认为它可能成为线上服务的一部分，但不会像互联网开启全民在线时代、广播成为生活标配、电力入户那样具有变革性。AI会存在，但如你所说，它本质是内容生产工具和降本工具。
迈克尔：所以你是说未来我们将观看AI生成的劣质内容，老板用AI虚拟人跟我们开会？
劳伦：已经实现了。你试过那些医保计划选择机器人吗？就是例子。
迈克尔：这听起来算是更健康的未来图景——这些技术逐渐渗透日常生活，而非如公司承诺的那般颠覆性。我在设想泡沫持续不破裂的可能性，或许风险值不是8而是6，存在转机。不过我是节目里的乐观主义者，才会这么说。
布莱恩：我们确实处于社会、技术、政治的多重特殊时刻。在经历泡沫的技术的社会政治建构中，有些路径鲜少被探索。例如从未有政府愿意持有美国科技公司10%股份：英特尔现由联邦政府持股10%，而政府明显青睐AI——随便看官方账号都在发布AI生成内容。他们认为这技术有多重效用。若泡沫开始破裂，我们可能看到政府进行经济干预，扶持企业或收购股权。我们在多个层面处于前所未有的境地，不能排除国家在AI企业面临金融危机时成为合作伙伴的可能性。
迈克尔：虽然这个话题结束得有点诡异，但我们必须进入休息环节。稍后回来。

（休息）

迈克尔：劳伦、布莱恩，感谢二位关于这个悬顶之泡沫的精彩讨论。相信未来数月乃至数年大家都会密切关注事态发展。读者可前往WIRED.com阅读布莱恩关于泡沫的文章，现在进入新环节"当红与过气"——新鲜事物当红，陈旧事物过气。劳伦先请？
劳伦：我的"过气"是开会。我认为只有两种情形需要开会：一是明确议程、逐项快速推进；二是无议程自由头脑风暴。最令人抓狂的是介于两者之间——既无效率又耗时间的会议。
迈克尔：所以你的"当红"是不开会？
劳伦：我的"当红"是察言观色。如今交流时人们总盯着屏幕，手机放桌上，面对面还开着笔记本电脑。试着注视对方脸庞，你会更懂对方情绪与意图，记忆也更深刻。虽然有人觉得困难，但请尽力在交流时保持专注。
迈克尔：扎实建议。
劳伦：这就是我的当红与过气。布莱恩，快用更精彩的答案拯救我们。
布莱恩："过气"给AI伴侣。这简直是社会毒瘤——毫无真实面孔交流。不针对使用者，但该收场了。我们已看到弱势群体如何被影响，见证过社交媒体的前车之鉴，而这简直是加强版社交媒体，一味迎合私欲。如劳伦所说，我们急需回归真实互动，摆脱自我中心的数字关系。所以AI伴侣过气——如果你是人均14个虚拟朋友之一，请丢掉好友挂件。"当红"我选卢德俱乐部——纽约兴起并蔓延全美的组织，他们拥抱劳伦所说的面对面联结需求。地铁里涂鸦"AI朋友"的活动正形成风潮。我研究这些，某种意义上算是"卢德主义正名者"，写了整本书为他们平反。我接触许多反抗科技巨头、争取真实相处时间、阻止AI吞噬生活的人们。所以卢德主义者当红。
迈克尔：精彩。
劳伦：迈克，你的选择？
迈克尔：当下正值换季时节，旧金山和纽约都是昼暖夜凉，该穿基础打底了。我的"当红"是卡佩莱内材质基础款，"过气"是羊毛基础款。羊毛虽好但易痒，还有伦理采购、纯素主义和过敏问题。我测试的卡佩莱内是100%再生涤纶，性能接近羊毛，更轻量透气，不像羊毛那么闷热。晨练或徒步时穿着效果很好，价格与羊毛相当，但穿整天或睡醒后气味更明显。建议尝试——
劳伦：我是卡佩莱内铁粉。
迈克尔：真的？太好了。
劳伦：虽然不常滑雪，但雪季必备。
迈克尔：感谢二位。布莱恩，谢谢做客。
布莱恩：非常愉快。
劳伦：期待再次邀请你。
迈克尔：等泡沫破裂时。
劳伦：天啊。
迈克尔：感谢收听《诡异谷》。若喜欢本期节目，请关注并评分。如有

英文来源：

Chatter about an AI bubble has been everywhere lately, and top tech companies like Google, Meta, and Microsoft have doubled down on their AI investments for 2026. But how have analysts in the past accurately identified forming tech bubbles? Hosts Michael Calore and Lauren Goode sit down with Brian Merchant, WIRED contributor and author of the newsletter Blood in the Machine, to break down the four criteria some researchers have used in the past to understand and brace for the worst.
Articles mentioned in the episode:
Please help us improve Uncanny Valley by filling out our listener survey.
You can follow Michael Calore on Bluesky at @snackfight and Lauren Goode on Bluesky at @laurengoode. 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.
Michael Calore: Hey Lauren, how are you doing?
Lauren Goode: I'm OK, Mike. It's earnings season, so a lot of us on the business desk here at WIRED have been tuning into tech companies earnings reports and their earnings calls. And I guess that basically means it's CapEx season.
Michael Calore: CapEx?
Lauren Goode: Capital expenditures.
Michael Calore: You say CapEx?
Lauren Goode: Yeah. Now that I'm a business desk reporter, I say CapEx.
Michael Calore: You're one of those.
Lauren Goode: I throw it around at parties. No, I really don't. But we are seeing a trend in how tech companies are sleeping on piles of money, but they aren't just sleeping on it. They're sharing big plans to spend on it, and especially to spend on AI infrastructure.
Michael Calore: Right. Data centers.
Lauren Goode: Yeah, more data centers. Not just data centers, but yes, that's a big part of it.
Michael Calore: And this is all partly what is fueling all of this talk about a bubble, which we touched on a little bit a couple of weeks ago with our colleague Molly Taft. But today we are going to be talking about AI bubble mania.
Lauren Goode: Yes, we are. And our guest today has a lot to say about it. The writer Brian Merchant is joining us. Brian, welcome to WIRED's Uncanny Valley. We're really thrilled to have you here, although we're not super thrilled about the topic we're talking about. How are you doing today?
Brian Merchant: Oh, I am as well as can be, I think. Yeah, thanks for having me. I am thrilled to get into the ins and outs of bubbledom with you both.
Michael Calore: This is WIRED's Uncanny Valley, a show about the people, power, and influence of Silicon Valley. It feels like suddenly everyone started talking about the AI bubble and whether we're in one right now. Since this summer, when OpenAI's Sam Altman casually mentioned the possibility of an AI bubble, the debate has only grown louder. Just last week, Bill Gates told CNBC that yes, we are in a bubble, but added that just like in the dotcom era, “something profound,” will come out of it. Perhaps he means an insane amount of money for the founders of chipmaker Nvidia, which just became the world's first $5 trillion company. For context that is 2.5 times the entire Canadian economy. Investments in AI are still flowing at full blast, and there are no signs of the money slowing down. Big Tech is already on track to spend $400 billion on artificial intelligence this year, and that is still not enough.
Google Meta, Microsoft, and Amazon are pledging even more money in 2026. And yet, according to some reports, 95 percent of businesses currently using AI say that they're seeing little to no return. So are we in a bubble or not? And if we are, how do we even assess that, and what is the potential fallout if the bubble bursts. To make sense of this, we have Brian on the show today. Recently, Brian used a historical framework to analyze whether AI fits the classic signs of an economic bubble. And if so, what that means for all of us. I'm Michael Calore, director of consumer tech and culture.
Lauren Goode: I'm Lauren Goode. I'm a senior correspondent.
Brian Merchant: And I am Brian Merchant, author of the newsletter and book, Blood in the Machine, and a contributor to WIRED.
Michael Calore: So Brian, you have set out to understand if AI is really causing a bubble, but I think we have to start by asking you to define what a bubble even is, because some purist economists would say that bubbles of any kind don't really exist.
Brian Merchant: I think that's why it was so important for me to try to actually sink my teeth into something a little more analytical or something a little more methodical, because so far it's been a lot of vibes-based commentary when we're looking at what's going on with AI and we see Nvidia cross these thresholds ... I mean 5 trillion is just the most recent. It was pretty astonishing when it crossed 3 and 4 trillion, and it's happening so quickly. I mean, folks might not remember that it's only been really a handful of years since we even had a $1 trillion company. And so the rate of capital accumulation and concentration is really accelerating and has accelerated over the last five to 10 years.
And so we look at what's going on and we look at the data center and it's like, it feels like a bubble. It seems like a bubble. It's got to be a bubble, right? But but what does that mean? And the most basic way to understand what a bubble is, is basically to think about the level of investment in a technology or a company or a commodity. And if more is being invested in that technology than it will ever return in its revenues and its profits down the line. You think of Pets.com or some of those early dot com companies as good examples where there's way more money invested in these companies. Toys.com was worth at one point, way more than Toys R Us was ever worth, despite it doing far smaller revenue figures. When it tanked and those investors never got paid out, we can say in hindsight that this was a bubble, that dotcom was a bubble, because people invested more than was ever returned. And now the fear with AI is that this will happen on a truly gargantuan scale with all the money that's pouring into Nvidia and the like.
Lauren Goode: So you ended up turning to two scholars, Brent Goldfarb and David Kirsch, who wrote a book about bubbles, and you applied their framework to AI. Talk a little bit about this framework.
Brian Merchant: Yeah, so honestly, I'm surprised that this book hasn't figured into more discussion about the AI bubbles given the enormity of said bubble. But in 2019, Goldfarb and Kirsch published this book, Bubbles and Crashes: the Boom and Bust of Technological Innovation. And what they did is they attempted to sort of go back and identify as many historical tech bubbles as they could and then identify exactly what made those bubbles and what sort of things they had in common, what kind of factors played into them, what prevented one boom from becoming a bubble and what pushed other investment booms into bubble territory.
And so they come up with four main factors that determine what makes a tech bubble tick. And those are the presence of uncertainty in innovation. So whether or not this technology or innovation is going to make money, how it's going to make money, what's the business model? Is it clear that people can turn a profit from this new technology that's emerging? Number two, the number of pure-play investments. That is the number of companies that are inextricably tied to the business success of a innovation or technology. So you think of something like CoreWeave in modern terms, a company that is completely dependent on the AI boom playing out, because its entire business model revolves around chips, around renting out space for cloud compute. So if the AI bubble bursts, then a company like CoreWeave is gone. So that's a pure play.
Number three, novice investors. Retail investors, sort of non-expert investors, need to have access to an investment vehicle that allows them to sink money into a new innovation. And so in this case today, anybody who has a Robinhood app and is investing in Nvidia is an example of retail investors taking to this innovation. Historically, we talked about the dotcom boom, and that's a really famous example of retail investors reading an article or two maybe on WIRED and saying, “Oh, the internet is the future.” And then there's all these companies that were going public that they could just plow their money into without recognizing maybe what the fundamentals of a given company are.
Finally, you need what Goldfarb and Kirsch call a coordinating belief or an alignment of belief among investors that a particular innovation is going to be the future. And typically this is demonstrated by a real-world use case of a technology. So you think about Chat GPT going viral and sort of organically demonstrating its consumer interest levels. In the past you've had things like Charles Lindbergh's transatlantic flight when the aviation industry was starting to take off and he successfully flies a plane across the Atlantic. And at the time, that was the biggest tech demo probably in history, and investors all wanted to get into the aviation game, and then that became a bubble. So those are the four: uncertainty, pure plays, novice investors, and alignment or coordination of beliefs.
Michael Calore: So Brian, that historical example of Lindbergh's cross-Atlantic flight is a good one because there are a lot of historical examples in the book and in your story, and I think it would help us understand uncertainty if you told us the anecdote about how we came to understand the possible uses of electricity in the 19th century.
Brian Merchant: Goldfarb and Kirsch talk a lot about the advent of electricity and electric light. And when it first comes on the scene, it's clear that this thing is a game changer, right? Electric lights. Suddenly you can have streets that are illuminated. You have these giant towers that cities are buying up that are sort of sending sparks everywhere and it's really sort of chaotic, but it's a powerful demonstration of this technology. But it's not immediately clear how that technology is going to make money, what the key market for it is going to be. Is it going to be light bulbs in the home? Is it going to be municipalities buying these giant tower lighting fixtures? What is going to be cost-effective? And it really takes decades and decades for all that to get sorted out. We kind of think of Edison inventing the light bulb and then the world changed. It's decades. It's decades between this technology getting figured out in the lab and transitioning into something that can actually make companies money.
Michael Calore: So there's no obvious application right away. There's the raw technology and it's awe-inspiring, but translating that into a real business takes a while, and that is certainly what it seems like we're looking at with artificial intelligence right now.
Brian Merchant: Yeah, exactly. And another great example is radio. Radio was broadcast radio when that first started being demonstrated publicly, it was a really big deal and everybody kind of knew, “Well, this has to be something,” but it wasn't clear whether it was going to be used as a marketing tool or to broadcast plays or what exactly. So it was another kind of famous bubble because it attracted a ton of investment throughout the 1920s around the same time as aviation. And there were no clear vectors for that business model right away. And again, with AI right now, we still don't know. We know that the chatbot is popular, we know that there are a lot of people using AI. But we still don't know.
Lauren Goode: OK, so we've touched on uncertainty. The next principle of a bubble you talked about is pure-play, which, as I understand it from your story, is when a company's fate rises and falls entirely on a single innovation. So I'm wondering how relevant that is for AI, which has so many moving parts to it. There's the Nvidias of the world, which is OK, that's a pretty impressive single innovation, although I'm guessing Jensen Huang and others at Nvidia would say, “It's not just a single innovation,” but let's just say for this sake it is, OK. But then you look at these models and there are so many different parts of them and possible ways that they could go and be commercialized. So how is this pure play?
Brian Merchant: So Nvidia has effectively become a pure-play company. We all know, or those of us that follow the space know, that before the AI boom, Nvidia made chips for graphical processing units for gaming. That was one of its big business segments.
Lauren Goode: Brian, we remember the days well, when at CES, Nvidia would have a press conference every year talking about their gaming inventions. And we would be like, “Who's going to go cover the Nvidia press conference?”
Michael Calore: Cards? We're still talking about cards?
Brian Merchant: And now it's right there on top of the mountain.
Lauren Goode: Right.
Brian Merchant: Because it has essentially foregrounded all of its business into supplying the chips that make the AI boom possible. It's the classic case of selling shovels during the gold rush. It's the most certain bet you can probably make, is that even if the AI boom starts going bust, you can imagine that we're still going to need all these chips and the chips are being bought. And so it's the most tangible sort of aspect for a lot of investors, and it's also a public company, so you can invest directly into it. And so it has become what we would consider a pure-play investment where its fate is tied to the AI boom more or less.
It is a little bit interesting because there have not been a ton of IPOs in the AI era. There have been some, famously CoreWeave, and usually that's one of the ways that people can invest in pure-play companies. But the AI boom is interesting in that it's all being concentrated into a handful of companies, and then a few other kinds of portfolio investments like real estate companies investing in data centers and that kind of thing.
Michael Calore: So if much of this money is just changing privately between investors and companies, and if the public is not really involved, then what is the risk for the public that these companies are getting so huge?
Brian Merchant: I mean, it is growing, and it's a weird case again because we are seeing such market concentration, which is unusual, but we have to remember that Nvidia is something like 8 percent of the entire stock market.
Michael Calore: Excuse me?
Brian Merchant: Yeah.
Michael Calore: I did not know that.
Brian Merchant: So if Nvidia goes bust, then it is a big deal to a lot of retail investors, but increasingly what's happening is that we're seeing a lot of this sort of circular investing that is sort of common in bubble eras where, as I mentioned, investors are starting to find ways to put money into data center expansion, for example. And that might be part of a portfolio that people aren't watching really closely, and they might not recognize that their fund manager has a company that's investing in real estate that's going to the data center boom, that's now sort of part of their portfolio.
We have these investment deals like OpenAI cut with the chipmaker AMD, where now that is a public company. And so people have investment in AMD, and its fate is increasingly tied to OpenAI due to the structure of some of these deals where OpenAI now owns a chunk of AMD. And so if that goes bust ... Basically we have a growing number of things that expose regular retail investors, regular pension holders, people with portfolios that are otherwise diversified. There's a growing exposure to an AI bust, and I think there's a lot of argument over how severe that exposure is, but to me, I look at all of these kinds of deals and these kinds of vulnerabilities and it does raise some red flags for sure.
Lauren Goode: So within this framework, the third of the four pillars is novice investors. It sounds like the real concern here is, yes, the novice investors having exposure to risky investments because everyone can just go to the Robinhood app now and do that, but actually it's the much larger institutional investments, the fact that the SoftBanks of the world and Nvidia and OpenAI taking a 10 percent stake in AMD and all of these circular, large-scale investments that are actually the biggest cause for concern in this potential bubble, right?
Brian Merchant: Yes. Because they're the ones wielding these phenomenal sums of money. And as Goldfarb told me in an interview, AI is so loaded with uncertainty, so unknowable on some level that it basically leaves everybody with the status of a novice investor, right? Because nobody knows what this future is going to be.
Lauren Goode: So somebody tell Sam Altman he's a novice investor.
Brian Merchant: Sam Altman may be one of the first to admit that he's kind of a novice investor. This is a guy who, when they were at the early stages of OpenAI and somebody asked him, “Well, how is this company going to make money?” I think this is five or six years ago. And he said, completely seriously, “We're going to build AGI and we're going to ask it.” That's our business plan, to establish artificial general intelligence and then ask the system itself how it's going to give … So talk about uncertainty
Lauren Goode: We're cooked. Brian.
Michael Calore: I read this book. It's called Hitchhiker's Guide to the Galaxy.
Brian Merchant: Well, that is to say we are just in totally uncertain waters here. I mean, I think the uncertainty is off the charts, and that's part of why I make the case that this is a bubble beyond other bubbles in the past.
Michael Calore: And that does bring us to the fourth indicator, which is the alignment of beliefs or the narrative. And as we all know, the current narrative in the world of AI is that AI is going to do everything. It's going to automate jobs, it's going to cure cancer, it's going to babysit your kids, it's going to fight climate change. And all of this will usher in an era of artificial general intelligence that's going to be able to do anything that a human can do. And the promise here is of limitless potential, which is really convenient because you just don't have to define what your goal is.
Brian Merchant: Talk about a story, right? It's the story to end all stories. And when I spoke to Goldfarb, he's like, “This is the one that's furthest out of the park,” because you can't just have a story. There has to be some level of feasibility that is present for investors to say, “You know what? This is feasible enough to happen.” And so the confidence with which these AI companies have come out with their products, the claims that they've made, the demo that sort of reinforces the belief in the system's capacities, it creates this stew where they can tell investors who really want to hear a lot of this stuff, right? “Oh, you can automate every job ever?” Like, “We've been waiting 200 years for something like that.” “Oh, it's going to cure cancer and we can just sort of put input into these systems for our pharma companies and technologies?” And, “Oh, it's going to solve climate change?”
There's something for everyone. There's something that every single investor or corporate partner wants to hear in this and could feasibly buy into, because for the last three years now, there has been this sort of coordination of beliefs, certainly among the majority of the investor class that, OK, this is the real deal, or it's a real enough deal that we're willing to put billions of dollars on the line because we don't want to get caught if it doesn't pan out. We don't want to have missed out on the investment opportunity for the do-literally-everything machine.
Lauren Goode: I had a conversation earlier this year for a WIRED story with the AMD CEO, Lisa Su, and we went back and forth of course for a while about AI, and we did both agree based on experiences we'd had with a parent being in the ICU that we thought that there was potential for AI to really help in health and medicine. But other than that, we sort of diverged. And I was asking her questions about the future of scaling and inference versus training and also content moderation and what happens when we get to this point in society where AI is like, people actually can't interpret what's real from what's fake. And she said, “You seem pretty skeptical about AI.” And I said, “Well, I just think the people who are most positive about it right now are those who stand to benefit the most.” And I quoted William Gibson, “The future is here, but it's not often evenly distributed.” And that's what I think we're headed towards.
Brian Merchant: Yeah, I think that's a really good point, and I think it's clear that there are going to be some victors from the AI boom even if the bubble does burst, and I think I may be even more skeptical about some of these elements of the AI blueprint or the AI business model than even you are. But it's clear that even if the bubble bursts, there's going to be utility that's found. And that's one thing you often hear to counteract those worried about a bubble, they say, “Well, the dotcom bubble burst and then we got all of this good stuff out of it.” And that could be true on a number of levels. It could also be true that again, the scale of this bubble could prove truly disastrous on an economic level that makes the dotcom burst pale in comparison. But it could also be that some of these use cases for AI that persist, like automating work or automating especially sort of the production of creative goods, are things that more permanently shape our economy in ways that aren't good for workers in the long term.
Michael Calore: All right, so the fundamental question of are we headed for a pop, we're going to put a pin in that question and we're going to come back to it after the break.
[Break]
Lauren Goode: Welcome back to Uncanny Valley. Today we're talking with Brian Merchant about whether AI is a bubble and what happens if that bubble bursts. Just last week, Nvidia CEO Jensen Huang said he doesn't believe we are in an AI bubble at all. It's just a quote-unquote “natural transition” from general-purpose computing to accelerated computing, which is something he's been saying for a long time. It makes sense that he wouldn't think of AI as a bubble, but it also consistently seems like, depending on who you ask, the answer changes.
If the bubble bursts, the economy could take a hit harder than the dotcom crash of 25 years ago. But if it holds, AI could fundamentally reshape how we work and live, and both could end up being true in the long run. So Brian, in the first half of this show, we talked a lot about the four-point framework that was created by Goldfarb and Kirsch. They also use a scale of 0 to 8, whether or not we're in a bubble. And when you ask them about AI, they came back with a big fat 8, a buyer beware.
Brian Merchant: Yeah. They found that all of the ingredients, so to speak, were present and some of them in quite large doses, and that according to this historically informed framework, we've got the maximum level of bubble alert here. It's worth saying that their framework doesn't necessarily say a lot about the potential scale of calamity, that if the bubble bursts, it's just like how much of each of these four different factors do we have? And then based on that, can we say that this is a bubble? So that I think is where they would probably want to say that their analytical framework ends, and they wouldn't necessarily say what I have said, what I use their framework to conclude, which is that this could be a bubble to end all bubbles.
Michael Calore: Well, just playing devil's advocate, even if this is the bubble to end all bubbles, history shows that these technological innovations don't just go away once the economic incentives around them lessen. I mean after the dotcom crash, the internet just kept growing and growing and I work for a dotcom 25 years later, could the same path be true for the AI industry?
Brian Merchant: Oh yeah, no doubt. I don't think even any of the fiercest bubble critics ... Well maybe the very fiercest bubble prognosticators would say that a bubble would just sort of banish AI from the world. I think that's completely unlikely, personally. I just think that there could be a lot of economic pain that happens in the interim. I think that AI has proven too popular among a certain user base for it to go away altogether. And tech companies have seen how popular these products are, and they still may need to find a way so that every query doesn't end up costing them money due to their compute power and resources that it drains. And it's very well likely that they could. But that said, I think it's obvious to me anyways, after going through this scale that the enormity of the investment sort of outpaces the level of utility that we're probably likely to see.
And I'm not alone there. There's a lot of more hardline investment banker types who are making similar conclusions, and some of the mania may be obscuring the actual balance sheets here, but it's entirely possible that AI continues to be a factor. It's interesting because one of the things that stuck around after the dotcom burst was the telecom infrastructure, right? Was like the build out of fiber-optic cables and that continue to let the internet run. The chips are in kind of a different situation because the chips are constantly being upgraded, and the chips today may not be super useful 10 years from now. So it's a little bit of a different story. I certainly think that pieces of this are going to stick around, and certainly companies will emerge victorious, but my crystal ball is just as good as yours.
Lauren Goode: Yeah. Brian, I'm glad you brought up broadband infrastructure versus chips because one of the questions that I started to have as I was reading your article, and maybe I'm so far off, but I was hoping maybe to concretize it, is I was wondering whether what ends up being enduring about generative AI is maybe its influence rather than its real potential for commercialization. If we compare it to previous bubbles, what if AI's long-term durability is more comparable to something like content, radio, social media, basically generating content, replacing knowledge work, maybe it's not as comparable to underlying infrastructure. What are your thoughts on that?
Brian Merchant: I would agree with that. I think that's its primary utility. I mean, the interesting thing about generative AI is that I think even its most full-throated proponents would agree that the quality of the content that it's putting out isn't miles better than what humans are doing. And in some cases it's even worse. Again, it's a cost proposition for a lot of companies. That's why I think that a lot of it does come down to this automation question, this question of labor. I think you've got really two buckets to look at. You've got the chatbot that sort of maps onto the social media space where it's a similar product where people are going to spend their time with this, and you can even think of that as kind of the automation of a human relationship. And then you have the labor automation bucket where a lot of people are saying, “Well, I can automate this part of my team's workflows or maybe I can cut down on labor costs over here.”
And again, the promise is that it will soon be able to do all of it. And I think that's one of those instances that I think is also indicative of a bubble, where what the sales pitch is and what's actually happening on the ground. And I think listeners may remember that a lot of this bubble talk, this most recent round of it got started when MIT published this study that said 95 percent of the firms that have invested in generative AI as an automation tool have not yet seen returns or profits as a result of their AI investment yet. So that's a pretty big number, and that started raising questions about how it is actually fairing on the ground.
So I tend to agree with you, I am still very skeptical that AI is reliable enough to be replacing the broad swath of functions in our daily digital lives where it will become an infrastructure-level thing. I more see it as potentially becoming one of the things that we do online and that is integrated into a lot of services maybe, or some services, but that isn't necessarily completely transformative in the way that the dotcom boom sort of inaugurated this era where everybody's online all the time or radio become a distinct feature of most Americans' lives or electricity being piped into every household. I think AI is probably going to be there, but I think what you're saying is right, it's a content production tool and it's a sort of a wage reduction tool.
Michael Calore: So you're saying that we're all going to be watching slop feeds and all of the YouTube videos that we're watching are going to be AI-generated, and our bosses are going to be using AI avatars to talk to us on.
Lauren Goode: They already do. Have you ever gone through one of those, like, choose your health care plan bots? It's there.
Michael Calore: So this does sound like a sort of healthier path towards the future, the fact that these things are encroaching on our daily lives and not being as transformative as the companies are promising possibly. I'm just trying to imagine a future here in which this bubble actually sustains and does not result in a crash, does not pop. It's not an 8 on the scale, maybe it's more of a 6 and there's a way to turn it around. But then again, I'm the optimist on the show, so I would say that.
Brian Merchant: Look, I think we are in a very unique moment right now, both socially, technologically, and politically, right? There are also a number of sort of avenues that have not often been explored in sort of this socioeconomic and political formation of a technology that is experiencing a bubble.
So for example, we have not really seen an administration that's been willing to take a 10 percent stake in an American technology company before. So Intel is now 10 percent owned by the federal government, and we know that the administration is a big fan of AI. Just look at its Twitter feed at any given time, you'll see AI-generated content, and that it finds this technology useful on a number of levels. So I think one possible outcome is if the bubble does start to burst, then we might see an administration that's willing to economically intervene in that bursting and maybe prop up some of the firms or buy stakes in them, which would, again, we're in an unprecedented moment on a number of levels, and I certainly wouldn't discount that from happening, from the state becoming a partner in these AI companies should they face a financial crisis.
Michael Calore: Well, that is a very strange place to end the conversation, but we do have to take a break. So let's take a break and we'll come right back.
[Break]
Michael Calore: Lauren and Brian, thank you so much for a great conversation about this bubble that we're all sitting on top of. I think everybody will be watching closely to see how it plays out over the next few months and years. In the meantime, you can read Brian's story on WIRED.com about the bubble, but for now we're going to dive into our new segments called WIRED and TIRED. Whatever is new and cool is WIRED. Whatever passé thing is on the way out is TIRED. So Lauren, do you want to go first?
Lauren Goode: OK, I'm going to start with my TIRED. My TIRED is meetings.
Michael Calore: OK.
Lauren Goode: I think there are probably two specific reasons you should have a meeting and then otherwise you should not have a meeting. And the first form of meeting that I think works really well is when you have an agenda and you bullet point it out and you get through it as quickly as possible because there are just things that you need to say to people in person or on a Zoom or something like that. But you should not go in with an agenda and meander. You should be very clear and explicit about the agenda.
And then the other kind of meeting that can be useful are the free-flowing, open brainstorms where you get together specifically without an agenda. But the idea is that you're going to benefit some of your team or your group of people. Your people are going to benefit from just having that free-flowing conversation.
Michael Calore: The no-bad-ideas meeting.
Lauren Goode: The no-bad-ideas, the yes and meeting, it's the in between that just kind of drives me crazy and we don't have time for it, folks. We don't have time. We got to get stuff done. It's like, sorry to be the Q4 person, but it's like getting to the end of the year.
Michael Calore: Sorry to be the—
Lauren Goode: Sorry to the KPI person.
Brian Merchant: It's your business desk instincts taking over again.
Lauren Goode: Yeah, it's all right. It's just in your personal life too, right? It's like, let's not have the in-between where you say we have something to accomplish, but then we're going to just completely monopolize the time.
Michael Calore: Your TIRED is all meetings.
Lauren Goode: A certain category of meeting.
Michael Calore: The meeting without the agenda.
Lauren Goode: But ends up being the meeting that everyone has.
Michael Calore: OK.
Lauren Goode: Yeah. So either be agenda or brainstorming and nothing in between.
Michael Calore: What's your WIRED? No meetings.
Lauren Goode: My WIRED is read people's faces better. I think when we communicate, this is like Lauren's life hacks. I think now these days when we talk to people often in person, we're still consumed by screens. Our phones are on the table in front of us. Maybe we're taking meetings with laptops in person. Just look at people's faces when they talk to you. It's remarkable how much more you can understand about what they're feeling and what they're trying to convey, and you actually remember what they say better when you're just looking them in the eye. And it's hard for some people. I do understand that, but just try to do the best you can to actually look at someone and be present with them when you're talking to them.
Michael Calore: Solid.
Lauren Goode: Those are my WIRED and TIRED or TIRED and WIRED or whatever we're calling it. I literally just came up with those. Brian, please do something better. Please save us. Save me.
Brian Merchant: I'm going to say TIRED are AI companions because talk about no face time at all. I've seen enough, right? This is a toxic development for society in general. No shade on anyone who is using an AI companion, but I think it's time to wind it down. We've seen what these things are doing to people in vulnerable states. We've seen what happened with social media in decades past, and this feels like just like social media on steroids and just catering to your every whim. And as Lauren said, I am a firm believer we got to get back at looking at people's faces and crawl out of these solipsistic digital relationships. So TIRED, AI companions—get rid of your Friend pendant, if you are one of the 14 people who have one.
And WIRED, I will say, again, calling back to what Lauren said are the Luddite clubs, these kids in New York City growing chapters around the country who are embracing, I think that exact tendency that you're talking about, the need to see more people's faces, more person-to-person connection. That great funny campaign in New York City where people are scrawling on the AI friend graffiti on the subways, and it's turning into this real thing. I write about this stuff. I am the Luddite Rehabilitator in Chief in some ways. I wrote a whole book about why the Luddites were actually right, and we've got them all wrong. So I hear from people and people who are trying to stand up to Big Tech to try to get more in-person time and to sort of stop AI from overrunning their lives completely. So WIRED, the Luddites.
Michael Calore: Nice.
Lauren Goode: All right, Mike, what's your WIRED/TIRED?
Michael Calore: OK, so we're in shoulder season right now, right? It's between the summer and the fall, at least here in San Francisco. And I think it's kind of the same in New York where it's like it's still nice during the day, but it's chilly at night and it's time to start wearing your long unders, your base layers. So I'm going to say for right now my WIRED is capilene base layers and my TIRED is wool base layers. Now I love wool. Wool is great. It's a little itchy. Some people have problems with it for ethical sourcing reasons, for veganism reasons, some people are allergic to it. So I have this great alternative which I've been testing, which is a capilene base layer.
Capilene is a fabric that is 100 percent recycled polyester that has been dyed and treated and everything and performs very close to wool. It's also lighter in weight and it is not as stuffy as wool is. So when you need a base layer, which is like when you're going out exercising early in the morning or you're going to be out hiking and you don't know what the weather is going to be like, you want to wear a good base layer. If you wear capilene, you can get something that's a little bit lighter, performs about the same as wool without the intense heat trapping and the intense itchiness that wool can bring you. So they are about as expensive as wool. They're a little bit stinkier than wool is after you wear them all day or if you sleep in them when you wake up in the morning. But they're really nice. So give capilene a shot if that's—
Lauren Goode: Huge fan of capilene.
Michael Calore: Are you? That's great. Great to hear.
Lauren Goode: I mean, I don't do snow sports that often, but when I do, great base layers.
Michael Calore: Yeah. All right, well thank you both. Brian, thank you for joining us.
Brian Merchant: It was my pleasure. This was a lot of fun.
Lauren Goode: We'll have to have you on again soon
Michael Calore: When the bubble bursts.
Lauren Goode: Oh, dear.
Michael Calore: Thank you for listening to Uncanny Valley. If you liked what you heard today, make sure to follow our show and rate it on your podcast app of choice. If you'd like to get in touch with us with any questions, comments, or shows, suggestions you can write to us at uncannyvalley@wired.com. Today's show is produced by Dara Lookpots and Mark Leyda, Amar Lal and Macrosound mixed this episode. Mark Leyda is our San Francisco studio engineer. Daniel Roman fact-checked this episode. Kate Osborn is our executive producer. Katie Drummond is WIRED's global editorial director. Chris Bannon is Condé Nast's head of global audio.