一年时间，华强北以及周围辐射的代工厂，已经没人不知道AI眼镜了。
深夜一点，东莞长安镇一处厂房仍亮着成片白光。
4条产线同时运转，几十个工人戴着防静电手环，以每小时220台左右的速度，把电池等元器件组装起来。车间角落的一间实验室内，机器在做着镜腿弯折和高低温测试，料盒里一半是镜腿、镜架，另一半还混着没来得及清走的蓝牙耳机主板。
这家靠做耳机起家的工厂，今年开始，已经全面转型。
而在去年，长安镇还是另一番光景。一家又一家的工厂，密集地支撑起了这个中国南方重镇的生产制造行业。大家做的产品各不相同，从整条镇子的主街穿过，你几乎可以看到横跨过去近十年来国内所有的热门消费单品。平衡车、头显、口袋相机、空气炸锅……
但如今，如果问他们要生产什么，答案几乎只剩下了AI眼镜。
一位工厂主欣喜地发现，做白牌AI眼镜，仅今年6月，订单就有7万副。而且这还是外贸带来的单量。过去的几个月里，数以百万计的AI眼镜，被出口到了欧美、东南亚和中东等地。
同样在半夜未入睡的还有60公里外的华强北商家张磊，就在盘完货准备休息时，收到了一个外商的订单邮件，客户白天在档口体验过AI眼镜之后，决定先拿200个回到自己的国家试试水。
今年春节过后，这种零散的订单逐渐多了起来，各种品类的AI眼镜也被张磊放在了档口最显眼的位置。一时间，AI眼镜成了市场中最炙手可热的单品之一。
据张磊估计，华强北大概一多半的商户都在做AI眼镜，这半年时间，整体销量比去年下半年翻了一倍不止。对于他的同行们来说，这就是现在的风口，物理AI的元年，眼镜是第一个爆品。
只是，这样的热闹并不陌生。过去几年，一个个曾经被视为“风口”的消费电子产品，都经历过类似的狂热，比如VR头显。如今，这个熟悉的剧本，似乎又一次出现在AI眼镜身上。
起势
AI眼镜赛道，高涨的热情已经持续一年多的时间，让人恍惚间，好像回到了元宇宙时期。
全行业都在赶产能，好像慢一步就会被风口甩掉。 去年开始，一向慢节奏的AI眼镜行业，开始跑了起来。
首先是投资人开始发觉，电话变得越来越多。FA开始主动找项目，原本还在观望的产业资本也陆续下场。被推荐过来的企业中，很多连产品都还没做出来，甚至连个像样的PPT都没有。
一名投资人杨玺记得，其中有一位创业者，此前在深圳有一条流水线做美容仪，如今也准备转型做AI眼镜，“这完全是不同的两类东西”。他和对方聊了半个多小时，“连第一笔钱都还没进来呢，他已经在想下一轮融资和量产后的销量了”。
而很多原本就身在局中的创业者也第一次感觉熬出了头。在XR行业干了6年研发的张帅发现，身边不少人敢拿着刚出品的Demo就去约投资人，参加各种路演，连基础常识还没搞清楚，就开始讲下一代人机交互入口的故事，“完全就是一群草莽的作风”。
这股风来得太快，以至于不少身处其中的人还没来得及弄清楚，消费者究竟为什么需要一副AI眼镜，资本和供应链已经先替他们回答了这个问题。
不过，草莽玩家们做货的过程一言难尽。张帅所在的品牌原本只为新品准备了2000副货，但预订量超预期10多倍。为了能够按时交货，他和老板天天追着供应链求他们扩产，一家一家敲门找愿意开新产线的代工厂。但和品牌方的狼狈相比，供应链端显然更加理智，“你说的百万量级销量，到底能不能兑现，你必须要想办法给我证明和保障”。
对方的态度并非毫无道理，“要跟供应商锁定产能，就要一次性投入好几个亿，工厂那边也要同步投入几亿甚至是十几亿，一旦没有长期订单支撑，那就完蛋了。但我们又没有这个能力买断产线，只能是一遍又一遍去说服那些老板。”
那是一段痛苦的经历。原本只需要搞定销量，现在不仅要用产品和市场数据去证明自己，还得在酒桌上让客户点头。张帅记得，有一次，他和主管市场的副总裁好不容易约到一家代工厂负责生产的总监出来吃饭，在酒桌上，副总裁上来就把分酒器里的一壶酒干了。最后带来的三瓶茅台，也有一大半都进了他和老板的肚子——把能攀上的关系都攀了一遍，才终于让对方点头，“答应第二天会上去跟老板推推”。
但产线搭好也只是第一步。不少AI眼镜创业者称，起初的良品率低的可怜，很多时候只有个位数。而且工人的分工也不合理，效率不高，更有人还会把镜腿、镜片装反。
为了赶产能，张帅的同事干脆直接住进了工厂。但理念和立场不同，导致他和工人经常争吵，有次他一激动把门踢坏了，最后还是老板上门赔罪。那时候的车间，灯是彻夜亮着，但所有人依旧焦灼——期货、瑕疵、交付慢，成了AI眼镜行业背负的沉重行囊。
移远通信XR产品线总经理马国文有着同样的感受。在他看来，想要做一款AI眼镜，不仅需要做光学方案，显示模组、芯片平台、AI算法、语音交互，系统、内容、应用也要整合，最后到了SMT贴片、整机生产，又要找工厂。
但即使过了工厂这一关，也只是过了及格线，保证交付质量，是此刻最重要的战斗。
几乎每一个零件和环节都有可能出现问题，即使样机最终做出来，软件还有可能掉链子，“软件一跑，卡了，那到底是算法、驱动还是散热的问题？后面又得几个月的排查”，马国文说。
他曾经手过一个案例。一家智能硬件初创公司，产品上市时间已经对外公布，样机却始终做不出来——第一版样机花了8个月，虽然反复调试，但底层兼容问题始终无解。最后倒数60天时找到移远，两边连轴排障，才在发布会前3天完成交付。而这已经是幸运的——更多品牌的资金链，还没撑到最后样机阶段就断了。
但今年，事情开始发生变化。订单越来越多，供应商越来越多，进入的企业也越来越多。曾经只有少数厂商能够供应的核心零部件，也开始出现更多玩家。
张帅记得，早期光机供应商屈指可数，一颗光机的价格可以达到三四百元，而当时AI眼镜基本都是双光机方案，需要配置两颗光机，成本因此更高。随着更多厂商进入，今年部分光机的价格已经降到200元左右，甚至有厂商为了抢客户，一度把价格压到99块钱。
蓝思科技、歌尔、瑞声科技等头部消费电子企业开始加码，移远、龙旗等OEM/ODM厂商纷纷建设专用产线，传统眼镜厂们都在寻找新的增长空间。原本属于TWS、手机和其他消费电子产品的成熟供应链，开始快速向AI眼镜迁移。
Omdia最新数据显示，2026年上半年全球AI眼镜出货量达到420万台，同比增长127%，中国智能眼镜出货量52万台。与此同时，上半年几乎平均每三天就有一款AI眼镜新品发布。这样的上新密度，很难不让人想起十年前国产手机的“机海时代”。据不完全统计，目前中国AI眼镜品牌数量已经接近400个，这之中不乏阿里、华为等大厂。
AI眼镜正式进入放量时代，而要理解这波热潮，得先读懂物理AI的全面爆发。
大模型落地到硬件，是今年以来行业的主要方向。手机、机器人、智能家居，各家都在寻找下一代人机交互的入口，物理AI是其中的一条路径——让AI直接感知现实环境，并对现实中发生的事情做出回应。
在这一轮探索中，眼镜被普遍视为一个合适的载体。原因大概在于，它的摄像头与人眼视角基本一致，麦克风天然靠近嘴边，扬声器则直接对着耳朵。换句话说，人类戴上之后不需要额外操作。相比之下，其他电子设备的使用则麻烦不少，手机需要掏出解锁，耳机只能收音不能拍摄，手表屏幕过小，VR头显则太重，也过于封闭。
也因此，眼镜成为物理AI中最早进入大众消费市场的品类。
定制冠军
所有人都觉得，产能的紧箍咒终于摘掉了，行业爆发的时刻已经到了。
国家补贴政策的出台，以及电商平台的让利，更是添了把火。今年618期间，单单京东平台的销量就同比增长超3倍。
一同燥热的，还有猎头圈。过去，XR领域的人员主要是在圈内流通，现在很多公司开始从大厂挖人，除了高薪之外，还拿出丰厚的期权作为筹码。
刘笑就是被招募的一员，当收到猎头的消息时，他直接愣住了。提供岗位的是一家他没有听过名字的公司。而且他也确实对这个行业不甚熟悉。之所以猎头找到他，是因为招聘单位点名要有华为或者阿里背景的人，而且很急，要不也不会半夜联系自己。
“酬劳确实很吸引人，给的title也很高，直接汇报给CEO，而且他们一直催着回复，甚至是CEO多次想要约着线下见面。”不过，当刘笑看到618前后的各种报告之后，他意识到不对劲，最终放弃了这次总包涨幅40%的机会。
之所以放弃，并不是说AI眼镜的终端数据不好，恰恰相反，市场总量在高速增长。让他开始警惕的，是增长背后越来越奇怪的一件事：几乎所有品牌，都在说自己是第一。“这不是健康的市场爆发，更像是混沌期的虚火。”
回看618期间的战报，大家用的机构来源各不相同，有的是IDC，有的是Counterpoint，还有其它的机构或是电商平台。口径也有所差别，比如出货量、销售量、销售额等。并且，由于现在对AI眼镜的定义并未形成统一的标准，音频眼镜、拍摄眼镜、带显示的眼镜，不同的定语下，冠军也多了起来。
更耐人寻味的是围绕“第一”衍生出的灰色生意。一名创业者称，今年开始，行业里有不少数据机构主动找上门，兜售定制化的排名报告，只要愿意付费，就能量身定制一套统计口径，给品牌安上一个“细分赛道第一”的头衔。
作为某头部XR品牌的前市场负责人，李倩的邮箱里经常收到这样的邮件。几乎毫无例外地，邮件上都标注着明码标价，私人定制。对方的话术并不直接，几乎没人敢明晃晃地把“造一个第一”表明出来，而是换一种更专业的说法——市场洞察、行业研究、权威认证、第三方数据支持。
但聊到最后，核心诉求总能统一：你想不想要一个“第一”？如果答案是肯定的，就可以根据品牌需求重新设计统计范围和口径。
灰色生意做得越来越长尾，有些机构卖的并不只是报告。他们还可以继续提供“第一”的传播服务，比如品牌宣传物料、销售材料。
李倩觉得这样的“第一”毫无价值，所以她从没回复过那些邮件。但很快她就有点摇摆了，因为她看到了一份市场报告，那个"细分第一"的位置，被替换成了她的友商。
不过促使她最终放弃的最核心原因，则是榜单中others（前五名之外的品牌）的占比。
目前，中国AI眼镜市场“others”的份额高达54%。对比之下，手机、平板行业的这一数据则是在17%左右。
一个成熟的大众消费品市场，前五甚至是前三的市场份额应该在六七成之上，才算健康。54%的others，意味着大量的白牌占据了市场。“也就是华强北看到的那些产品。一个靠供应链整合的行业，暂时看不到更高的上限。”刘笑说。
这一点，也经常被XREAL创始人徐驰拿来吐槽竞对：“自信一点讲，今天除了XREAL，其他家都是供应链整合，这不是一件丢人的事。”但同样，XREAL也因为暂时没有AI眼镜产品而被质疑，很多榜单上更是坐不上主桌。
即便如此，“争第一”依然是每家品牌的必修课。因为对仍在烧钱的赛道而言，“第一”的头衔并不是给用户看的。它争夺的已经不只是消费者的注意力，更是资本的信心。
不同于具身智能可以变着花讲不同的故事，AI眼镜的底层逻辑依旧是硬件思维，销量是一切故事的起点。“第一”的背后不仅是品牌层面的竞争，更是对资本的一种示好和证明，每一个数据都是融资桌上的筹码。
算账，全链失血
行业的火热，暂时还没有偃旗息鼓的迹象，但并非所有人都选择继续Buy in，首先就是一级市场的故事不好讲了。
杨玺最近听到同行议论，之前拿着PPT都能约到一波投资人，而现在创业者发过去BP，对方的回复却越来越直接：什么时候能盈利？
如果单从融资金额来看，今年以来，雷鸟、影目、XREAL、Viture、Rokid、逸文以及多家产业链相继在上半年完成了一笔或多轮融资，有的规模甚至超过了10亿元。根据洛图科技（RUNTO）不完全统计：2026上半年，全球智能眼镜产业链融资、并购、战略投资至少有64起，披露的总金额约为90亿元。
但剥开数字外壳，融资结构的变化早已悄悄暴露了行业的尴尬。一笔笔大钱里，70%集中在了上游光学、微显示供应链端，落到终端品牌的资金里，绝大多数也来自产业资本，而非纯财务投资。
用股权换供应链资源与产能优先级，这并不是一件错事，但财务投资的熄火，似乎也在说明一级市场已经逐渐失去耐心。
一位深耕消费电子领域的FA人士透露，2025年是AI硬件的投资狂热期，只要沾着AI的标签就能拿到钱。但今年大家都慢慢回过味来了，AI眼镜很多时候只是套了个AI的壳子，走的仍是硬件的发展体系，“现在就算给你100亿的现金，未来的这一仗也是不够的”。
他认为，“未来行业的注意力一定会转向具身智能、确定性更高的AI光模块以及存储芯片等上游算力产业链。”
这背后是硬件投资的冰冷规律——消费电子的迭代周期长、商业验证慢，远不如算力赛道的爆发性强。在没拿出像样的销量答卷前，资本自然不会持续砸钱。
真格基金投资合伙人徐梧认为，AI眼镜接下来的路，或许没有想象中那么顺。赛道时间长，短期市场空间没想象那么大，言下之意，这是一个十年起步的赛道。
更核心的痛点在于市场的天花板：即便头部品牌年销量冲到几十万台，翻倍增长，但放在成熟消费电子赛道里连零头都算不上。就算是华为和小米这样的号召力，第一个月销量也才几万台，暂时撑不起一级市场期待的高估值与快回报。
更为残酷的现实是，现在多数终端品牌几乎是卖一台亏一台，靠着拉新融资去回血。一方面是前期高额的研发和产业投入，另一方面则是较低的毛利。
张帅透露，AI眼镜存在着性能、重量与续航的不可能三角。拿重量来说，几乎是每减少1g就要花掉一两千万的成本，现在的优化标准都是以0.1g为单位，而且后面的成本会越来越高。
不仅仅是硬件，软件、系统以及AI也要亲自去做。雷鸟创始人李宏伟曾透露，光AI团队的人数就差不多有30人。
“这样算下来，一款AI眼镜的毛利要到六成以上，才有可能收回成本。”这个数据也是硬件行业的底线，但实际情况远没有那么乐观。张帅发现，“大家真的是在贴着成本线定价，观影类的AR眼镜可能还能赚钱，像显示类的AI眼镜，毛利也就20%多，关键这部分利润也都给经销商了。”
眼镜作为一款强线下体验的产品，通过开设体验店以及铺货到更多的代理手中，是行业目前最常见的做法。但即便是手里握着20%的利润空间，大多数代理商也没赚到钱。
不同于张磊做外贸白牌生意，李象（化名）代理的是Rokid、雷鸟这样的大品牌，线上线下同步在卖。虽然一个月好的时候也能卖个两三百台，但算上投流、赠品等费用，最后也很难赚钱。
“我们不卖白牌机，那就是一次性交易，但确实成本低，有钱赚。我这忙前忙后，加上退货的成本，不说是白干，但也差不多，投入产出比太低了。”
通常情况下，传统眼镜电商的退货率是15%~30%，做得好的是可以控制在10%以内。对比之下，AI眼镜的退货率今年依旧是六七成，没有太大改观。
李象所经历的，正是整个AI眼镜的一个缩影，尤其是高退货率，远超手机、智能平板等成熟的3C产品。“线下还好，都是深度体验后下单的，我这线上的退货率能有50%以上，关键有时候品牌那边还压货不给换新，成本只能自己担着。”
高退货率的背后，不只是产品质量或消费者冲动消费的问题，更折射出AI眼镜当下的一个尴尬：行业已经开始用智能手机的方式讲述下一代终端的故事，但体验上远没有出现那个“Aha Moment”。
“相信AI眼镜会是下一个超级终端的人，会尝试多次购买。”一名用户称，基本每次有新款AI眼镜上市他都会买来试用，“赌的就是新一代产品已经解决上一代的问题”。但结果就是带来新的失望。
目前市面上的新产品，症结依旧固定在框架不新颖、长时间佩戴不舒适、续航差、性能妥协——这是多年前就被反复提及的老问题，至今依旧很难通过迭代来解决。
用户愿意花几千元去买单，“但起码得是传统眼镜的佩戴舒适度加上智能手机80%的体验才行，现在顶多是个‘功能机’”。
马国文认为，目前智能眼镜退货率偏高的一个重要原因，是行业存在“拔苗助长”的现象。目前订单尚未出现快速爬坡，智能眼镜仍处在类似智能手机的“诺基亚时代”。
一个典型的现象是，小米在今年5月给车主发了AI眼镜六折购买券，但很多都被车主挂在闲鱼上，以几十元的价格就卖掉了。
给汽车充满一次电的钱，最终就是AI眼镜的价值。一名接近小米的人士告诉我们，小米AI眼镜的项目在内部没达到预期，亏损金额大概在2亿左右，而且目前新品节奏也在放缓。
更底层的矛盾在于，这个行业赚钱的逻辑不是做增量，而是降本。最知名的例子就是XREAL，为了实现正向现金流，过去三年这家公司一直在做的是把销售费用降下来，但这本身就不是朝阳产业该有的样子。
当然，也有在风口赚到钱的，比如开发者。但是，他们赚钱不是靠的用户付费，而是参加一些品牌组织的开发大赛，以及一些B端文旅项目等。
Lisa就是其中的一员。她曾是苹果vision OS的第一批开发者，但前后赔了十几万，做出的软件在推广时遇到了很大阻力，只有部分极客用户感兴趣，后来转向了Rokid等国内AI眼镜软件生态的开发。
“C端肯定是赔钱的，但品牌会介绍一些工业、文旅等B端项目，算是能维持着生存。现在行业这现状只能先等待，最起码你得有100万用户，我们这些生态开发者才能真正跑出来。”
上游的日子也并非高枕无忧。JBD这样的龙头企业，凭借初期大量的订单赚到了不少钱，但亏钱的更多。李熙透露，自己朋友里十个有7个都是亏的，有一个专门开了条新的线去做AI眼镜，但良率低，最后卖了1亿多，算总账还亏了几千万。
现在她只做外贸订单，因为她发现，“中国有一个怪象，那就是创业者无论做什么，都会变得特别卷”。
AI眼镜的今天，让她偶尔会恍惚想起当初智能音箱时代——2017年前后，阿里、小米、百度等厂商先后进入智能音箱市场，依靠补贴把价格压到百元以内，出货量迅速放大，但一直没有形成稳定盈利。风口过去后，大部分品牌退出，只留下少数几家。
上市，续命赛跑
当一级市场的粮草难以为继，奔赴二级市场就成了头部玩家的共同选择。今年4月，XREAL向港交所递交招股书。在此之前，Rokid、雷鸟均被传出要冲刺IPO。
很少有人知道的是，这股上市潮里还藏着一个没成行的计划。当3、4月份行业里各种IPO消息传出来的时候，一家头部品牌也动了心思。一名熟知内幕的行业人士称，品牌方内部也曾讨论过，是否要找媒体放出即将上市的消息，吸引一波热度，最终再出公告进行否认。但这个危险的计划最终被老板叫停。
对方之所以要这么做，大概率是为了暂时缓解资本方的压力。今年春节之后，陆续有投资人承认，对于AI眼镜行业来说，未来的一两年内或许是上市最后的机会了。毕竟AI眼镜还没真正进入大众消费市场，用户规模、商业模式和刚需属性都尚未得到充分验证。一旦行业热度退潮，资本担忧它重蹈元宇宙的覆辙。
但这种“抓住最后窗口”的判断，并非所有行业参与者都认同。关于IPO，徐驰给出的答案更为克制，他并不认为这背后有谁在推动，投资人也不都是短视的，上市是水到渠成的事情。
不过，在部分创业者眼里，上市这阵热潮中，能被看到的更多是仓促。据张帅了解，XREAL递交招股书时，甚至在国内还没有一个专职的IR，而另一家对外放话两年内IPO的公司，很长时间内连一个专职的PR都没有。两家都是奔着二级市场去的公司，目前还没有搭起专职班子——这足以说明很多问题，创始人们未必真的准备好了，只是箭在弦上。
一名创业者最近听说的消息是，一家原本打算28年左右再做上市的公司，现在也提前了，“这里面未必都是创始人的个人决策，大多数情况下，投资方的立场无法忽略”。行业里此前还传过的消息是，Misa（Rokid创始人）被劝着有机会被大厂收购就赶紧接受，变现退出。
在一次社群的讨论中，谈及扎克伯格要“收购”Rokid，Misa曾回复说：“有点后悔，早知道这么难，我就答应了。”
这显然有开玩笑的意味。
但IPO背后，看似是行业突围的光明出路，行业最敏感的一面也同样暴露——多数公司尚未盈利，二级市场的投资者会先于一级市场看到这份答卷。利润和现金流，恰恰是整个AI眼镜行业目前最难回答的两个问题。
压力之下，有的公司已经选择退出。
今年年初，蜂巢科技主动砍掉了AI眼镜新品线，这家公司去年在国内有10%的份额，销量也处于上升期。在创始人夏勇峰看来，眼镜这条路或许会成功，但未必能直接通向下一代AI硬件。
“及时止损，不要在乎沉没成本，他这么做是对的。”在投资人老邵看来，大家明显变得更清醒，本以为行业会达成的共识，现在反倒是变成了分歧，“遮羞布总有被撕开的那一天”。
这种认知或许源于今年年中的一场全员大会。他投过的一位创始人，在那场会议上宣布，公司要进行全面的收缩，“因为全行业的产品都不过关，包括他们自己”。
但行业的曲折注定无法影响到所有人。华强北的档口里，张磊依然每天拆包、发货、处理退货。他不关心“下一代入口”这个判断，只看品类每天的出货情况。
这是华强北商人的生存哲学，也是整个行业的缩影。但可以确定的是，所有的爆发、名次、资本故事，最终都要回到那个最朴素的问题：产业链何时止血，高退货率何时有解。在那之前，所有的“元年”和“爆发”，都缺少依据。
而AI眼镜正在经历的这一切，其实并不陌生。
过去几年，VR头显等一个个曾经被寄予厚望的消费电子品类，都经历过类似的轨迹——当所有人开始讨论下一代入口的时候，市场逐渐回归现实。
AI眼镜，似乎也无法绕开这段让人熟悉的路。当然，它可能真的成为下一代AI硬件的入口，但这中间隔着的，从来都不只是技术。（文中相关人物均为化名，文 | 志读科技）
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
In a year, no one knew about Ai's glasses at the radiation plant in and around China.
Later in the night, a factory in East Chang An town was still lit in white light.
Four production lines operated simultaneously, and dozens of workers were wearing electrostatic bracelets to assemble, for example, batteries at a rate of approximately 220 units per hour. In a laboratory in the corner of the workshop, the machine was performing mirror leg bends and high-temperature tests, half of the material boxes were mirror legs, mirror frames and the other half were mixed with the bluetooth headset master plate that could not be removed.
The factory, which started with headphones, has been completely transformed since this year.
Last year, Chang An was another sight. A factory after a factory has heavily supported the manufacturing industry in this major town of southern China. The products that you make are different, going through the main street of the whole town, and you can almost see all the hot-food items that have been in the country for almost a decade. Balance car, head, pocket camera, air fryer...
But now, if you ask them what they want to produce, the answer is almost all that's left of them.
A factory owner was pleased to find that in June of this year alone, there were 70,000 orders for white-carded AI glasses. And this is a single amount of foreign trade. Over the past few months, millions of AI glasses have been exported to Europe and America, South-East Asia and the Middle East.
Also up to 60 km away from sleep, Zhang Zhang Zhengbei, a Chinese businessman, received an order from an outsider when he was on board for a break, and the client decided to take 200 of them back to his country and try water during the day after he had experienced the Ai glasses at the slot.
Following the spring season, these sporadic orders have gradually increased, and all kinds of Ai glasses have been placed at the top of the slot. In a moment, Ai glasses became one of the hottest singles in the market.
According to Zhang’s estimate, about half of China’s businesses in North China are doing AI glasses, and the overall sales have more than doubled over the second half of last year. For his colleagues, this is the current wind mouth, the year of Physical AI, the first blast of glasses.
It's just, it's not strange to be so excited. Over the past few years, a consumer electronic product that was once considered a “window” has experienced similar fanaticism, such as the VR's predominance. Now, this familiar script seems to be on AI glasses again.
Rise
The Ai glasses track, the high enthusiasm that has been going on for more than a year, has left people in a state of tremors, as if back to the time of the meta-cosm.
The whole industry is running out of capacity, as if one step later would be lost. Starting last year, the slow-paced Ai glasses industry started running.
First, investors began to realize that the number of telephones was increasing. The FA has begun to look for projects, and industrial capital, which was still on the lookout, has begun to fall. Many of the companies recommended had not produced a product or even a decent PPT.
One investor remembers that one of the entrepreneurs, who used to have a water line in Shenzhen to make a hairdresser, is now preparing to transform into an AI glasses, “This is a completely different type of thing”. He talked to each other for more than half an hour, “not even the first money came in, and he was already thinking about the next round of financing and the volume of post-partum sales”.
And for the first time, many of the entrepreneurs who were already in the Bureau felt like they were out of business. After six years of research and development in the XR industry, Zhang Zhang found that many of the people around him had the guts to go to meet investors with the newly produced Demo, attend various road shows and start telling the story of the next generation of humans, “a bunch of recklessness”.
The wind came so quickly that many of the people in it had not been able to figure out why consumers needed an Ai glasses, and the capital and supply chain had answered the question for them first.
However, the process of making goods by reckless players is complicated. While Chang's brand had only prepared 2,000 by-products for the new product, the booking was more than 10 times higher than expected. In order to be able to deliver the goods on time, he and his boss pursued the supply chain every day to increase their production, and a knock on the door to find a substitute factory willing to open new production lines. But the supply chain is clearly more rational than the branding, “You have to find a way to prove and guarantee the million-volume sales you're talking about, or not.
The other side's attitude is not unreasonable: “It takes hundreds of millions of dollars to lock up the capacity with the supplier, and hundreds of millions, if not billions, on the other side of the factory, and if it is not supported by a long-term order, it is over. But we don't have the capacity to buy the cut-off line, but to convince the bosses over and over again."
That was a painful experience. All we had to do was fix the sales, and now we have to prove ourselves not only by using product and market data, but also by nodding our customers at the table. As Zhang remembers, on one occasion, he and the Vice-President in charge of the market had a hard time meeting the director in charge of the production at a substitute factory to eat and at the wine table the Vice-President had come up and dried up a bottle of wine from the saloon. The last three bottles of the pavilion, as well as a large part of it, entered his and his boss's belly — climbing up the relationship and finally nodding his head, “promising to go up the next day and push the boss”.
But getting the line right is just the first step. A number of AI eyeglass entrepreneurs claim that the initial good taste rate is poor and, in many cases, only a single digits. Moreover, the division of labour is unreasonable and inefficient, and there are those who turn their legs and mirrors back.
In order to catch up, Zhang's colleague simply went to the factory. However, the difference in philosophy and position led to frequent quarrels between him and the workers, and on one occasion he got excited and kicked the door down and ended up in the house with the boss. At that time, the lights were on all night, but everyone was still burning — the futures, the defects, the slow delivery — and became the heavy bag on the back of the Ai glasses industry.
Ma Guowen, the master manager of the XR line of migration communication, felt the same way. In his view, to make an AI eyeglass would require not only optical solutions, showing modules, chip platforms, AI algorithms, voice interactions, integration of systems, content and applications, and, finally, SMT stickers, whole machine production, and factories.
But even after the plant, it is just passing the pass line to ensure the quality of delivery, the most important battle at this time.
Almost every piece and link is likely to have problems, and even if the prototype is finally made, the software could fall off the chain, "The software runs, the card goes, is that a algorithm, the drive or the heat? It's gonna take a few more months to run.
He's had a case. A first-generation firm of smart hardware, whose product had been publicly listed, had been unable to produce a prototype — the first version of the prototype had taken eight months and, despite repeated debugging, the issue of bottom compatibility remained unresolved. Only three days before publish had the final final 60 days to find a remote and two-axis detachment been completed. And this is already fortunate — more brands of financial chains that have not yet reached the final prototype stage.
But this year, things began to change. More and more orders, more suppliers and more businesses. Only a few manufacturers were able to supply core parts and more players began to appear.
As Zhang remembers, the early light machine suppliers were able to count, and the price of a light machine could have reached $3,400, whereas at that time the AI glasses were basically a two-ray machine programme, requiring two light machines and therefore more expensive. As more manufacturers enter, the price of some of the light machines has fallen to around $200 this year, and some of them have even set their prices at $99 to rob their customers.
The front-runner consumer electronics of Lenth, Gol, Sing, etc. began to expand, and OEM/ODM manufacturers, such as the Mover, the Dragon Flag and others, have built dedicated production lines, and traditional glasses factories are looking for new growth spaces. Originally a mature supply chain for TWS, mobile phones and other consumer electronics, rapid migration to AI glasses began.
The most recent data from Omdia show that in the first half of 2026, the global number of AI glasses exported reached 4.2 million, an increase of 127 per cent over the same period, and 520,000 from China. At the same time, there was almost one new Ai glasses publish on average for the first half of the year. This new density is hard not to recall the “Airsea Age” of a national mobile phone 10 years ago. According to incomplete statistics, the number of China ' s AI glasses brands is now close to 400, including large factories such as Ali and Hua.
The AI glasses are officially entering the Quantification Age, and to understand the tide, it is necessary to read the full-scale AI outbreak.
The landing of LLM into hardware is the main direction of the industry since this year. Cell phones, robots, intelligent homes, every home is looking for the next generation of interactive humans. - Let AI know what's happening and respond to what's happening.
In this round of exploration, glasses are generally considered a suitable carrier. The reason probably is that the camera is basically the same as the human eye, with the microphone naturally approaching the side of the mouth, and the speaker directly to the ear. In other words, humans do not need additional operation when they wear them. In contrast, the use of other electronic devices is problematic, as mobile phones need to be unlocked, headphones cannot be filmed, watch screens are too small and VR headers are too heavy and closed.
As a result, the glasses became the first physical AI product to enter the mass consumption market.
Custom Championship
Everyone felt that the tight spell of capacity had finally been removed and that the time had come for the industry to explode.
The introduction of a State subsidy policy, as well as the facilitation of electric power platforms, has added to the fire. During the 618th of this year, the sales of the Kyoto platform alone increased three times over the same period.
It's hot together, and there's a hunting ring. In the past, people in the XR field had been mostly circulating in circles, and many companies were now beginning to dig people out of large factories and, in addition to their high salaries, to offer a strong option as leverage.
Liu Xiao was a recruiter and when he heard about the headhunting, he stopped. The job is provided by a company whose name he has not heard. And he's not really familiar with the industry. Headhunters found him because the recruitment agency named someone with a Chinese or Ali background and was in a hurry to contact themselves at night.
“It is true that the pay is attractive, that it is very high, that it is reported directly to the CEO, and that they have been rushing to respond, and even that the CEO has repeatedly wanted to meet under the line.” However, when Liu smiled at the reports around 618, he realized that something was not right and eventually abandoned the opportunity to increase the total by 40 per cent.
The abandonment was not to say that the end data for the Ai glasses were poor, but rather that the total market was growing at a high rate. What keeps him on guard is the increasingly strange thing behind growth: almost all brands are saying they're first. “This is not a healthy market explosion, more like a fire of chaos.”
Looking back at the 618 war papers, the institutional sources used varied, ranging from IDC to Counterpoint, to other agencies or electric power platforms. There are also differences in calibres, such as volume of deliveries, sales, sales, etc. Moreover, since there are no uniform criteria for the definition of AI glasses now, there are more champions under different terms.
More interesting is the grey business that emerges around “first”. One entrepreneur stated that, at the beginning of the year, a number of data agencies in the industry had volunteered to come to the door and sell customized ranking reports, and that if they were willing to paid, they would be able to customize a set of statistical calibers and place a “subtract track first” title on the brand.
As a former market owner of a head XR brand, Lee Tsing often receives such mail in his mailbox. Almost without exception, the e-mails are branded and customized. The other side's rhetoric is not direct, and almost no one dares to show “building one first” but rather a more professional one — market insight, industry research, authoritative certification, third-party data support.
But at the end of the day, the core claims always converge: do you want a "first"? If the answer is in the affirmative, the scope and calibre of statistics can be re-engineered in line with brand needs.
The grey business was becoming more and more protracted, and some agencies were selling more than reports. They could also continue to provide “first” communication services, such as branding materials and sales materials.
Lee Tsing thinks this "first" is worthless, so she never answered those emails. But soon she was kind of swinging, because she saw a market report, and the "subdivided first" position was replaced by her friend.
However, the central reason for her eventual abandonment was the proportion of others on the list (other than the top five brands).
At present, the “others” share of the Chinese AIS market is as high as 54 per cent. In contrast, the data for mobile phones and tablets was around 17 per cent.
A mature market for consumer goods, the top five or even the top three, should be above 60 and 70 per cent, to be healthy. Fifty-four per cent of others meant that large numbers of white cards occupied the market. "That's the product that John Wigand saw. An industry that relies on supply chain integration does not see a higher upper limit for the time being.” Liu laughs.
This is also often contested by the XREAL founders: “It is not a disgrace to say that today all but XREAL are integrated in the supply chain.” Similarly, XREAL was questioned for not having an AI glasses product for the time being, and many of the lists were not available at the main table.
Even so, “first in battle” remains a compulsory course for every brand. Because the first title is not shown to the user for the track where the money is still burning. It now competes not only for consumer attention, but for the confidence of capital.
Unlike being smart enough to tell different stories, the bottom logic of AI glasses is still hard thinking, and sales are the starting point for all stories. Behind “first” is not only brand-level competition, but also a gesture and proof of capital, each of which is a chip on the finance table.
Count it. Blood loss in the chain.
The industry is hot, there are no signs yet, but not all of us have chosen to continue Buy in, and first of all the first-level market story is difficult to tell.
Yang has recently heard peer-to-peer discussions, with a PPT dating a wave of investors, and now that the entrepreneur has sent the BP, the answer is more and more straightforward: when will it be profitable?
In terms of the amount of financing alone, since the beginning of this year, thunderbirds, films, XREAL, Viture, Rokid, Luang, and a number of industrial chains have completed one or more successive rounds of financing, some of them on a scale of more than $1 billion. According to RUNTO, the figure is not complete: in the first half of 2026, there were at least 64 cases of financing, M&As and strategic investments in the smart eyeglass industry worldwide, with a total disclosed amount of about $9 billion.
However, a change in the financing structure has already revealed the embarrassment of the industry by stripping the digital shell. Of the bulk of the money, 70 per cent was concentrated in upstream optical, micro-showing the supply chain, and the vast majority of the money fell into the end brand, also from industrial capital rather than purely financial investment.
It is not a mistake to switch shares for supply chain resources and capacity, but the collapse of financial investment seems to be an indication that the first-tier market has become impatient.
An FA person in the field of deep-growing consumer electronics revealed that 2025 was the time of investment in AI hardware and that money could be obtained with the label of AI. But this year everyone has come back to the taste that many of the AI glasses are only a shell of AI and still go through the hardware development system, “the battle of the future is not enough now, even if you have $10 billion in cash”.
In his view, “the attention of the future industry will surely turn to an upstream computing chain such as smart, more definitive AI-ray modules and the storage of chips”.
Behind this is the cold pattern of hardware investment — an iterative cycle that consumes electronics and slows commercial validation, far less explosive than the calculus track. It is natural that capital does not continue to smash money until a decent sales response is produced.
Suh Xu, the investment partner of the JTF, believes that the way that Ai glasses follow may not go as well as expected. The length of the track and the short-term market space are not as big as they would have been, which means that it is a 10-year-old track.
The more central pain lies in the market ceiling: even if the head brand is worth hundreds of thousands of tablets a year and doubles it, it is not even a scratch in mature consumer electronics. Even with the convening power of Huacheng and Mi, the first monthly sales volume was only tens of thousands, which temporarily could not support the high valuations and quick returns expected by the first-tier market.
The harsher reality is that most of the end brands are now selling almost one in deficit and reblooding with fresh financing. On the one hand, there are high levels of R & D and industrial inputs in the previous period, while on the other hand there are low levels of M & M.
Zhang has revealed that there is an impossible triangle of performance, weight and continuation of the AI glasses. In terms of weight, the cost of almost every 1g reduction is 12 million, and the optimization is now in 0.1g, and the costs are going to get higher.
Not just hardware, but software, systems and AI. LEE Hwae, the founder of the Thunderbirds, revealed that the number of members of the AI team alone was almost 30.
“In this way, it will be possible to recover the cost of an AI lens with more than 60% of its gross salary.” This data is also the bottom line in the hardware industry, but the reality is far less optimistic. As Zhang found, “Everyone is really pricing at a cost line, and it's possible that AR glasses of a visual type can make money, like the Ai glasses of a graphic type, with more than 20 per cent of Māori, and the key part of the profits will go to dealers.”
Glasses, as a product of intense under-the-line experience, are now the most common practice in the industry, through the opening of experience shops and the distribution of goods to more agents. But even with a 20 per cent profit margin in hand, most agents do not earn money.
Unlike Zhang Zhang Zhang's foreign-trade white card business, Li 象 (alias) is represented by large brands such as Rokid and Thunderbird, which are being sold in parallel online. Even though they can sell 200 or 300 in a good month, it is hard to make money when it comes to investment, gifts, etc.
“We don't sell white machines, it's a one-off deal, but it's really low-cost and rich. When I'm busy, plus the cost of returning the goods, it's not so much dry, but it's almost as low as the output.”
Usually, the return rate for traditional glasses and electricians is 15-30 per cent, and good work is within 10 per cent. In contrast, the return rate of AI glasses is still 60-70% this year, with no significant improvement.
It's a microcosm of the whole of AI's glasses, especially the high rate of return, far above the mature 3C products of mobile phones, smart tablets, etc. “The bottom line is fine, it's all in deep experience. I have a return rate of more than 50 per cent on this line, and sometimes the key is that the brands keep the price on their own.”
Behind the high return rate is not just a matter of product quality or consumer impulse consumption, but a more embarrassing reflection of the current situation of AI glasses: The industry has begun to tell the story of the next generation of terminals using smartphones, but the experience is far from that “Aha Moment”.
"The one who believes that the Ai glasses will be the next super-end, will try to buy them many times." According to one user, he buys it for trial almost every time a new AI glasses are put on the market, “it is the new generation of products that have solved the problems of the previous generation”. But the result is a new disappointment.
The problem with new products in the market is still that the framework is not new, that it is not comfortable for a long time, that it lasts, and that it is a compromise. This is an old problem that was repeatedly mentioned many years ago, and it is still difficult to resolve it in an iterative manner.
Users are willing to spend thousands of dollars on the bill, “but at least the comfort of wearing traditional glasses plus 80% of the experience of smartphones, which is now at best a ‘functional machine’”.
According to Ma, an important reason for the current high rate of return of smart glasses is that the industry is “enhanced by seedlings”. The order has not yet seen a rapid climb, and smart glasses are still in the “Nokia age” like smartphones.
A typical phenomenon is that Mi gave the owner an Ai glasses for a six-cent discount in May this year, but many of them were hung by the owner on idle fish and sold for dozens of dollars.
The money to fill a car with electricity is ultimately the value of the Ai glasses. A man close to Mi told us that the Mi AI glasses project did not meet expectations internally, with a loss of about 200 million, and that the pace of the new product is slowing.
The bottom paradox is that the logic of making money in this industry is not incremental, but a reduction. The best known example is XREAL, which, in order to achieve a positive cash flow, has been doing for the past three years to reduce the cost of sales, which in itself is not what should have been in the Sun-yang industry.
There are, of course, people who make money in the wind, like developers. However, they do not earn money from the user paid, but from participating in development competitions organized by brands, as well as from the B-end Brigade project.
Lisa is one of them. She was the first developer of Applevision OS, but lost hundreds of thousands of dollars, and the software produced encountered considerable resistance in its roll-out, with only some extremist users being interested, and then shifted to the development of the ecology of the country ' s AI glasses software, such as Rokid.
“C must have lost money, but brands are able to sustain themselves by introducing B-side projects such as industry and the Civil Brigade. Now the industry can only wait, at least you have one million users, so that we, the eco developers, can really get out.”
Upstream days are not safe. A leading enterprise, such as JBD, has earned a lot of money through a large number of initial orders, but more money is lost. Lee Hsi-hee revealed that seven out of ten of his friends were at a loss, and that one had opened a new line to make the Ai glasses, but the good rate was low, with a final sale of over 100 million, and the total amount lost tens of millions.
Now she only orders foreign trade because she finds that “there is a strange image in China, that is, that entrepreneurs, whatever they do, become special rolls”.
Today, the AI glasses remind her of the original smartphone era — around 2017, when manufacturers like Ali, Mi and Baidu entered the smartbox market, relying on subsidies to keep prices within a hundred dollars, rapidly magnifying the volume of the goods delivered, but never making a steady profit. After the wind had passed, most brands had been withdrawn, leaving only a few.
Come on, let's go!
When the first market is unsustainable, moving to the second market becomes a common choice for head players. In April of this year, XREAL submitted a request to the Hong Kong office. Before that, Rokid and Thunderbirds were sent out to sprint IPO.
Very few people know that there is still an unsuccessful plan in this market. When IPO news came out in March and April, a head brand moved. According to a well-known trader, there was also a discussion within the brand as to whether to find the media to release the news about to be listed, attract a wave of heat and eventually release the announcement for denial. But this dangerous plan was finally stopped by the boss.
This is probably done in order to temporarily relieve capital pressure. After the spring festival, investors admitted that the next year or two might be the last chance to be on the market. After all, the AIS lenses have not really entered the mass-consumption market, and the size of the user, the business model and the attributes of demand have not yet been fully validated. Once the industry cools, capital fears that it will repeat the course of the universe.
But this “capturing the last window” judgement is not shared by all industry participants. With regard to IPO, the answer given by Seo-hye was more restrained, and he did not think that there was anyone behind it, that investors were not all short-sighted, and that listing was a matter of water to canals.
However, in the eyes of some entrepreneurs, it is more hasty to be seen in this boom. According to Zhang ' s knowledge, XREAL did not even have a full-time IR at the time of the submission of the application, while another IPO company had not had a full-time PR for a long time within two years. Both of them are going to the second-tier market, and there are no full-time teams - That is enough to show that the founders are not really ready, but the arrows are on the string.
One entrepreneur recently heard that a company that had intended to be listed again for about 28 years was now ahead of schedule, “it is not necessarily a personal decision by the founders, and in most cases the investor's position cannot be ignored”. The previous news in the industry was that Misa (founder of Rokid) had been persuaded to accept it as soon as he had had the opportunity to be bought by a large plant.
In a community discussion, when it came to Zuckerberg's “acquisition” of Rokid, Misa replied, “Being sorry, knowing so hard, I promised.”
That obviously means a joke.
But behind the IPO, it seems like a bright way out of the industry, and the most sensitive side of the industry is exposed as well. – Most companies are not yet profitable and investors in the secondary market will see the questionnaire before the first market. Profit and cash flows are just two of the most difficult questions for the entire AI glasses industry at present.
Under pressure, some companies have opted out.
At the beginning of this year, hive technology initiated the cutting off of new lines for Ai glasses, a company that last year had a 10 per cent share in the country and was in an upward trend. From the point of view of the founder, Ha Yong Feng, the path of glasses may be successful, but it may not lead directly to the next generation of AI hardware.
“To stop the damage in time, not to worry about the costs of sinking, he did the right thing.” In the eyes of investors, it became clear that the consensus that the industry would have reached had become more sobered, and now it is a disagreement, “the day when the veil is always torn apart”.
This recognition may have originated in a full-fledged General Assembly this year. At that meeting, one of his founders announced that the company was going to have a full contraction “because the products of the entire industry were not cleared, including themselves”.
But the twists of the industry are bound not to affect everyone. Zhang Zhang Zhang Zhang continues to unpack, ship and process returned goods on a daily basis. He does not care about the “next-generation entrance”, but only about the daily delivery of goods.
This is the survival philosophy of the business people of China and the North, and it is a microcosm of the whole industry. But it is certain that all the eruptions, titles and capital stories will ultimately go back to the simplest question: when the chain will stop bleeding and when the high return rate will solve. Until then, all the “years” and “explosions” had no basis.
And all that Ai glasses are going through, it's not really new.
Over the past few years, VR has experienced a similar trajectories in a group of consumer electronics that has been given high hopes - As everyone begins to discuss the next generation, the market gradually returns to reality.
Ai's glasses don't seem to be able to bypass this familiar road. Of course, it may really be the entry point for the next generation of AI hardware, but it's never just technology. (The characters are all aliases, literature, learning technology)
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.