Logo
Become a Client

国际数据公司(IDC)最新数据显示，2026年第二季度，中国工业PC市场出货量同比增长16.8%，增速创下近五个季度新高。但亮眼数据背后，市场参与者正面临更复杂的周期挑战：本轮增长同时包含 AI 转型带来的真实增量，与短期供需波动催生的前置需求，无法通过简单线性外推判断未来走势。厂商需穿透数据表象调整经营策略，从追求规模增速转向订单质量与产品竞争力的提升。

一、Q2市场复盘：16.8% 高增速的双重增长构成

2026年Q2中国工业PC市场出货118.9万台，同比增长16.8%；上半年累计225.4万台，同比增长11.9%，呈现同环比双高增的态势。

IDC认为，本轮增长由双重性质完全不同的动力共同驱动，也决定了市场增长的结构稳定性不足：

首先是AI 边缘算力带来的真实增量，具备长期可持续性。伴随产业加速进入以智能体为核心的新阶段，边缘侧算力缺口持续扩大；工业 PC 凭借高稳定性优势，可支撑多步骤、长时间运行的本地化 AI 任务，为制造等核心场景提供可靠端侧算力。产业智能化转型的长期趋势，为工业 PC 市场带来了坚实的底层增量。

其次是涨价预期下的需求前置，需要警惕的短期变量。面对核心元器件价格上行与交期不确定性，下游客户集中提前锁货，导致部分订单实质是对 2027 年需求的提前支取。整体来看，2026 年上半年的高增长由多重因素叠加促成，厂商制定 2027 年经营计划时，不宜直接用当前增速做线性外推。

二、周期深层解析：三重因素主导行业增长节奏分化

第二季度的高增长由短期与长期双重动力共同支撑，但拉长全年乃至三年周期来看，行业增长节奏并非线性延续。供需格局、产业链供给与下游资本开支的三重变化，正在重塑工业 PC 市场的周期逻辑。

IDC 预测，2026 年中国工业 PC 市场出货量预计达 448.9 万台，市场规模约 120.0 亿元；2026–2030 年复合增长率 13.1%，2030 年将达 735.5 万台。其中 2026 年下半年预计出货约 223.5 万台，相较上半年环比微降 0.8%，增长动能已出现边际走平的迹象。CPU、PCB、Power IC 供给紧张是增长放缓的核心原因，头部厂商交期已拉长至半年；积压订单虽将递延至 2027 年释放，但订单动能已经边际走弱，头部厂商中国区订单出货比降至 1.21，为全球主要区域最低水平。

具体来看，三重因素共同影响了工业计算机市场的增长节奏：

  1. 上半年高增长掺杂显著的前置需求，透支后续增长动能。存储芯片涨价、交付周期拉长，促使客户提前备货，实质上透支了后续年度的市场需求，2027 年市场需要消化这部分提前释放的订单。背后的深层原因是全球存储行业正处于 AI 超级周期下的结构性供需失衡：三大原厂持续将晶圆产能转向高毛利 HBM，而 HBM 单片晶圆消耗量约为通用 DRAM 的三倍，持续挤压成熟制程产能，本轮涨价周期预计将延续至 2027 年底。。
  2. 供给瓶颈持续时长超出预期，订单大规模向后递延。CPU、PCB、功率 IC 等核心物料紧缺，头部厂商产能饱和、订单交期长达半年。2026–2027 年全球成熟制程处于旧产能已满、新产线尚未投产的瓶颈阶段，新增产能要到 2028 年才会形成有效产出，大量订单因此向后递延。
  3. 制造业资本开支节奏切换，中长期增长动力换档。2026 上半年自动化市场呈现 OEM 领跑、流程行业承压的结构性分化。设备更新资金集中在年内释放后，2027–2028 年缺少同等规模新增资本开支接续，项目型市场增长动力随之回落。

IDC行动建议：三大维度重构经营策略

综合来看，前置需求透支、供给瓶颈延期、资本开支换挡三重因素叠加，意味着行业高增长难以简单线性延续。厂商需要跳出传统的年度规划思路，从经营规划、产品价值、业务结构三个维度针对性调整，以应对周期波动。

1.重构年度经营规划，摒弃线性推演逻辑：针对需求前置带来的周期错配风险，首要动作是打破传统规划惯性。摒弃以年度出货量、积压订单线性外推来年业绩的传统模式，对2026年上半年在手订单进行结构化拆分，分为AI真实刚需、常规设备更新、涨价前置备货三大类型，针对各类订单分别搭建独立的测算假设与经营模型。

避免将本年度高出货、高积压订单直接作为2027年业绩基本盘，充分预留行业周期回落的缓冲空间。预判2027年上半年行业同比增速大概率承压，需提前制定排产、物料采购、库存管控的应急预案，实现主动风控，避免订单下滑后被动减产、调仓的滞后操作

2.依托边缘AI场景，筑牢产品溢价能力：面对产业链成本上行与供给约束的长期压力，企业需要跳出硬件堆砌的内卷，打造差异化价值。高端高性能机型需跳出单纯硬件配置堆砌的竞争逻辑，聚焦边缘AI核心赛道打造差异化价值。通过优化算力配置、深耕细分场景适配、完善配套软件栈，打造设备本地化长流程AI稳定运行的核心能力，形成客户认可的差异化价值支撑，以此实现合理产品溢价。

同时依托落地场景案例、软件配套服务收入，持续验证产品溢价的真实性与可持续性，摆脱低价同质化竞争。

3. 均衡产品结构，强化跨周期对冲能力：为了对冲行业周期波动与单一业务的经营风险，均衡化的产品结构是跨周期发展的核心支撑。避免单一产品线重仓押注的经营风险，构建多元化、抗周期的产品组合体系。依托标准化通用机型的规模优势，通过有序成本传导稳定市场份额，不盲目跟风涨价；充分利用项目型定制订单与AI场景业务、传统设备更新业务的周期错配特性，实现业务互补对冲。

最优经营结构为：AI高性能机型贡献核心毛利增量，标准化机型保障业务规模与稳定现金流，搭配灵活的新品迭代节奏，对冲当前设备更新周期拉长带来的行业增长压力，提升企业整体经营稳定性。

进一步交流

本文数据及观点均源自IDC《中国工业PC市场季度跟踪报告，2026Q2》。如需了解分形态、分行业、分场景的详细数据，或针对特定厂商策略进行定制化分析，敬请联络IDC中国研究团队。请点击此处与我们联系。

2026年以来，主流数据库厂商围绕AI的产品与战略发布明显提速。海外 Oracle 持续迭代 Agentic AI 增强的 Oracle AI Database，Google Cloud发布数据库智能体工具预览；国内 OceanBase、腾讯云、阿里云、华为云等厂商，也相继发布面向 AI/AI Agent 负载的数据库新版本。

这些发布共同指向一个变化：数据库厂商不再只回答“数据库能否支持AI”，而是开始回答“数据库能为AI承担多少数据准备和持续服务工作”。IDC认为，这轮竞争的关键，不只是增加多少AI功能，而是能否将企业数据转化为AI可持续使用、可治理、成本可控的业务上下文。从功能叠加走向可持续的数据供给，是当前AI原生数据库发展最值得关注的产业变化。

厂商密集发布的背后，是需求侧正在发生的结构性变化。当 AI 从演示走向生产，企业面对的不再是“要不要用 AI”的问题，而是“AI 能不能持续、稳定、成本可控地用上企业数据”的问题。数据规模、数据类型、数据消费方式都在改变，数据库作为企业数据的主要承载者，自然被推到了这轮变化的前台。也正因如此，AI 原生数据库才从一个产品概念，变成了一个需要被认真讨论的产业命题。

AI 工作负载重构需求，推动数据库迈向 AI 原生时代

国际数据公司（IDC） 相关调研结果显示，2025 年企业新生成的数据资产中约 90% 为非结构化数据，但大量数据散落在各类存储系统、各类业务系统之中，多模态数据相互割裂，企业手握数据 “富矿”，却缺少将数据转化为 AI 可用燃料的冶炼能力。

《IDC FutureScape: 全球 Agentic AI 2026年预测——中国启示》中提到，到2027年，未优先构建高质量AI就绪数据的企业将在扩展生成式AI和Agentic AI解决方案时面临显著挑战，导致生产力下降15%。

进入到2026年， 产业界已经形成清晰共识：AI 的竞争焦点正在离开大模型本身的能力演示，转向如何让企业存量海量数据真正服务于生产级 AI 业务。造成企业 AI 落地困局，本质是传统架构围绕结构化数据构建，难以应对海量非结构化多模态数据的处理需求。这些数据分散多系统带来副本冗余、版本不一致，极易引发 AI 输出失真。同时数据访问主体向 AI Agent 转变，原有架构缺少适配机器高频访问的安全、隔离与状态管理能力。此外传统治理体系无法覆盖向量、多模态数据，缺少将原始业务数据加工为 AI 可用上下文的完整链路。

为此企业需要数据库具备多模态混合检索、Agent 运行支撑、AI 链路治理以及成熟的企业级能力，支撑 AI 走向生产落地。这一产业现实，也让 AI 原生数据库的价值被推到台前。

向量能力只是入场券，不等于 AI 原生数据库。

增加一个向量检索能力，或者在管理控制台中接入一个对话助手，并不足以说明整个系统已经完成AI原生转型。判断“原生”程度，更有价值的观察标准是：当数据发生变化、模型需要替换、权限需要调整、负载出现波动时，这些能力能否在统一的管理机制下持续运行，还是仍需开发者在多个系统之间反复维护连接与同步关系。

因此，谈到AI原生数据库，不是要求数据库必须从零开发，也不是要求大模型必须运行在数据库进程内部，而是强调AI工作负载是否已经成为数据库架构设计的重要前提，而不再只是附加的使用场景。

AI需要的不是一堆数据，而是可用的业务上下文

向量搜索擅长发现语义相似的内容，但相似度本身无法判断数据是否满足时间、状态、金额等业务条件。例如一个售后服务Agent需要判断某笔订单能否退款，它不仅要找到一份语义相关的售后政策，还要核对订单时间、商品状态、客户身份、政策期限，以及操作权限。

这要求数据库具备标量与向量相结合的混合搜索能力，将结构化字段的精确匹配、范围过滤与向量相似度检索纳入同一查询过程，共同回答“哪些数据既相关，又满足当前业务条件”。对于企业AI而言，结构化查询并未因向量搜索的引入而失去价值，反而成为约束检索结果的重要基础。

因此，评估AI原生数据库，不能只看是否支持向量搜索，还需要看其能否融合标量与向量查询，并进一步承担多模态数据的组织与管理。真正的能力边界，不是“能检索多少种数据”，而是“能否将不同类型的数据组织为适合当前任务的业务上下文”。

Agent成为新用户，带来的不只是更多查询

IDC 预测，到 2031 年中国企业将拥有 3.5 亿个活跃智能体。

当数据库的服务对象拓展至持续执行任务的智能体时，数据访问需求将不再局限于单次查询。Agent 的长周期任务会反复读取数据、留存中间状态、调用业务工具，并依据返回结果持续迭代执行逻辑，这使得记忆与状态管理能力变得至关重要；面向 AI 开发场景，数据库环境的快速创建、隔离测试与回滚能力同样成为刚需。

由此看，AI与Agent并没有让数据库传统能力失去价值。相反，当数据被用于连续执行的业务流程时，事务一致性、资源隔离、故障恢复和可观测性依然非常重要，需要与新的检索、记忆和模型调用能力一起接受检验。

一体化正在升温，但不是把所有数据搬进同一个库

这一轮发布还有一个值得关注的共同点：厂商越来越少只发布一个孤立的检索功能，而是开始提供覆盖数据接入、处理、治理和应用服务的组合。在《IDC FutureScape：全球AI驱动的业务战略2026年预测——中国启示》中显示，到2027年，40%的企业将战略性投资AI融合的数据架构，以避免错误决策和竞争力下降。

面向AI工作负载，“一体化数据管理底座”的趋势愈发明显。但不能将“一体化”简单理解为“一个数据库替代所有系统”。对于企业，更重要的是减少重复的数据搬运、相互矛盾的业务定义和分散的权限管理，而不是追求产品数量最少。同样，支持开放格式或统一接口，也不意味着跨系统的一致性、性能和治理问题已经自动解决。企业仍需验证：数据更新多久可见，权限如何传递，多个引擎能否使用一致的业务定义，以及不同工作负载之间是否会相互影响。

“一体化”的价值应当体现在面相AI工作负载时的复杂度下降，而不是架构图上少了几个方框。

谁将受益，谁又面临压力？

根据《IDC Market Glance: 中国数据库管理系统，3Q26》报告显示，中国数据库管理系统涵盖了包括关系型数据库、多种非关系型数据库、数据仓库与湖仓一体、专用型数据库等等多类细分领域。当前，几乎所有细分领域的产品都在从不同“控制点”向面向AI的数据管理赛道演进。

当前看来，拥有存量业务数据、成熟数据库能力和应用生态的厂商，具备向AI数据管理延伸的有利起点。但这种优势能否转化为新增收入，取决于它们能否降低客户的实际实施成本，而不只是把更多功能放入产品目录。

其他细分领域数据库厂商也并非没有发展空间。对于检索规模、响应延迟、数据类型或部署方式具有特殊要求的场景，专用产品仍可以提供更合适的方案。其竞争力是证明特定工作负载下的效果与成本优势，同时降低与企业现有数据体系对接的难度。但需要警惕的是，当向量生成、混合检索、模型调用和连接服务逐步被平台集成，单靠“支持某一项AI功能”建立的优势可能收窄。

真正被重构的，是数据库向AI提供价值的方式

从2026年以来的代表性发布看，数据库厂商正在同时回答两道题：如何借助AI改善数据库自身，以及如何让数据库更好地支撑AI。两者的交汇，使数据库的能力边界从存储与查询，进一步延伸至检索、上下文、记忆和受控的数据服务。

AI原生数据库的价值，不应由产品名称决定，也不应由演示中一次成功的问答证明。而是应该体现在，在复杂业务和持续变化的数据环境中，面向AI工作负载能稳定提供正确时间、正确范围内的数据，并让这一过程可管理、可追溯、成本可控。

IDC认为，未来的数据库竞争，不只是“谁能把数据管好”，还包括“谁能让AI持续、可靠、成本可控地把数据用好”。

进一步交流

如您希望进一步了解 IDC 对 AI 原生数据库、AI数据基础设施市场的规模预测、厂商格局及企业采用阶段的研究，或需要针对贵公司所在行业与目标市场进行定制化解读，欢迎与我们联系。请点击此处与我们联系。

Nan Wang

Nan Wang - Research Manager, Enterprise System and Software Research, IDC China

Analyst Profile
Nan Wang is a research manager the IDC China Enterprise Research Group. His research focuses on China’s IT security market, and includes the study of the IT security software market and IT security appliance market. Nan is also involved in…

2026年，全球智能眼镜市场迎来新一轮结构性变革。国际数据公司（IDC）最新发布的《全球智能眼镜市场季度跟踪报告》显示，第二季度全球出货量同比增长35.3%，音频及拍摄类眼镜成为增长主力，AR/ER眼镜加速渗透，头显市场则面临调整。国际大厂持续主导行业方向，生态与AI体验成为竞争新焦点；中国创新厂商加快全球化步伐，通过产品多元化和渠道拓展提升海外影响力。与此同时，市场结构正向高阶智能化演进，应用场景不断拓展。尽管中国市场短期承压，但行业活力未减，技术创新、生态协同和全球化布局为未来发展注入新动能。本文将系统梳理全球及中国市场现状、主流厂商动态与未来趋势，探讨智能眼镜产业的新机遇与挑战。

全球市场

IDC最新数据显示，2026年第二季度全球智能眼镜（Smart Eyewear）市场出货量达到354.7万台，同比增长35.3%。其中，音频及音频拍摄眼镜出货量为249.4万台，同比增长54%；AR/ER眼镜出货量为50.5万台，同比增长75.7%；VR/MR头显出货量为54.8万台，同比下降23.3%。

2026年上半年，全球智能眼镜市场累计出货量达到711.4万台，同比增长70.5%。本轮增长主要由眼镜形态产品推动，市场结构正逐步向更高阶和智能化方向演进。下半年，随着新品发布和技术融合，行业格局有望进一步调整，但整体需求表现还需持续观察。

全球市场产品动态

1.国际大厂份额领先，资源投入重心明显转移

国际头部厂商持续引领智能眼镜行业发展方向。Meta凭借音频类智能眼镜稳居市场领先地位，并积极筹备新品发布。苹果首款智能眼镜产品已进入量产倒计时，行业竞争重心正由单纯的硬件出货量转向生态系统和AI体验的深度融合。尽管国际大厂多线布局，但资源投入重心已发生明显转移。

Meta：继续保持全球智能眼镜市场的领先份额，二季度推出更具性价比的新品，进一步巩固其在大众消费市场的基础。Ray-Ban Display等显示类智能眼镜受限于良率和产能，出货量尚未形成规模，细分市场优势尚待建立。头显方面，Quest系列持续推进，但整体出货量预计有限。

Apple：首款智能眼镜产品进入量产倒计时，头显产品线节奏相应调整，下一代头显发布时间或将大幅延后。

2. 中国科技创新厂商加速全球化，产品线与渠道多元化

中国科技创新厂商加速全球化布局，通过丰富产品线、拓展海外渠道和深化本地化运营，在显示类眼镜领域形成差异化竞争力，海外市场份额持续提升。

雷鸟：加速产品迭代，推出主打高端观影体验的GT系列和轻量化V4 AI拍摄眼镜，同时Q3将正式发布iO系列显示类智能眼镜。其今年海外市场拓展加快，线下渠道建设提速，上半年在全球及中国消费级AR&ER眼镜市场保持领先。

Xreal：海外市场出货占全球比例已超过60%。产品线持续丰富，覆盖不同价位段和细分场景，包括入门级XBX A01+、和面向游戏市场的R1。同时Q3将在海外正式发布Project Aura，进一步完善高端显示类智能眼镜布局，全球市场区域覆盖日益完善。

Inmo、Viture等厂商也在海外市场持续发力。Inmo计划Q3推出王嘉尔联名款，拓展时尚消费场景；Viture专注细分市场，持续扩展海外零售渠道，在北美等地表现稳健。

3. 生态搭建与跨界合作提速，应用场景持续拓展

生态搭建与跨界合作加速，智能眼镜应用场景持续拓展，厂商通过AI能力、内容生态和多终端协同不断提升产品价值。

千问：千问在国内带摄像头、以AI功能为主的智能眼镜市场中快速上量，依托大模型与电商、支付生态的深度整合，取得领先。三季度有望与新品千问AI耳机协同，进一步完善AI硬件生态布局。

Rokid：海外出货开始增加，二季度在全球非全彩眼镜市场保持领先。生态层面，Open Day 2026发布Yoda OS完善底层系统，后续计划引入腾讯WorkBuddy AI办公能力，探索眼镜在办公场景的落地。

中国市场

2026年二季度中国智能眼镜市场出货61.1万台，同比下滑8.0%，市场首次出现负增长。下滑主要受到音频眼镜市场影响，但整体行业竞争依然活跃，厂商持续优化产品线和渠道布局，新玩家不断进入，市场活力未减。

中国细分市场概况及市场格局

音频和音频拍摄眼镜市场

2026年二季度，中国音频和音频拍摄眼镜市场总出货量为33.2万台，同比下滑35.4%。市场结构继续升级，音频拍摄眼镜市场份额已超过音频眼镜，占比达到55.3%，同比增长81.4%。

本季度市场下滑，主要是由于头部品牌产品逐步进入生命周期末期，导致整体出货承压，而市场上推出的新产品尚未形成有效接力。同时，音频眼镜产品同质化问题突出，缺乏差异化创新，进一步影响了消费者的换新意愿。尽管促销季带来了一定关注，但对销量的拉动作用有限。在此背景下，部分新厂商对市场持观望态度，入局节奏相对缓慢。未来市场能否恢复持续增长仍取决于创新功能的落地和用户需求的进一步释放。

AR/VR市场

2026年二季度，中国AR/VR市场出货量27.9万台，同比增长85.8%。

AR&ER眼镜市场同比增长123.0%，其中受益于以AI功能为主的轻量级眼镜的带动。单色眼镜的市场份额首次超过全彩眼镜，达到51.2%，以千问、Rokid、Inmo等品牌为代表。价格段覆盖主流消费区间，满足不同用户需求。同时从《中国智能眼镜市场月度跟踪报告》数据来看，讯飞将于7月开始出货，预计将进一步推动市场格局变化。行业整体呈现出技术创新加速、应用场景多元化的积极趋势，AI与AR&ER硬件的深度融合为市场带来更多想象空间。

VR&MR头显市场二季度出货量同比下滑34.4%，下滑有所收窄。随着行业应用的不断拓展，头显在大空间商用、机器人后台数据采集、远程操控等新场景中展现出新的增长机会，有助于消化部分库存。同时Pico将在年底发布头显新品，将对市场格局及未来走向带来较大影响，行业将普遍关注其在产品量级和生态建设方面可能带来的变化。整体来看，随着新产品的推出和行业应用的深化，头显市场有望回暖，迎来结构性增长机遇。

未来展望与新机会

IDC中国市场分析师叶青清认为，尽管有国补政策和促销活动的支持，智能眼镜作为新兴科技产品，用户的购买决策依然偏向谨慎。当前，智能眼镜作为终端生态与AI能力融合的重要交互入口，其潜在价值尚未完全释放。只有当产品体验、内容生态和实际应用场景进一步丰富，产业链参与者和终端消费群体才会持续扩大。

展望下半年，市场将出现以下几个动向：

第一，产品线分化与创新加速。下半年，厂商将加快产品线分化，针对不同用户需求在色彩显示、重量控制、佩戴舒适性等方面推出更多组合方案。轻量化全彩眼镜将成为布局重点，既满足日常佩戴的便携性，也提升视觉体验。部分厂商还将探索AI、影像、健康等多功能融合，推动产品从单一功能向多场景应用转型，进一步激发用户换新意愿。

第二，出海步伐加快，全球化布局深化。随着国内市场竞争加剧、增长空间阶段性受限，越来越多中国厂商加速海外市场拓展。海外细分市场尚未完全饱和，尤其在AR&ER等领域，国产品牌凭借技术积累和性价比优势，已在部分区域建立知名度。轻量级产品的推出有望进一步打开海外市场空间，降低与国际大厂正面竞争压力。除科技创新企业外，部分传统大厂也在积极布局或关注海外机会，全球化将成为行业主旋律。

第三，生态建设与场景落地持续推进。生态体系的完善和跨界合作将成为智能眼镜行业发展的关键。厂商将加大与内容、AI、支付、电商等生态伙伴的协同，推动智能眼镜在办公、娱乐、健康、教育等多元场景的落地。AI大模型与终端硬件的深度融合将提升产品智能化水平，带动新一轮应用创新。

进一步沟通

IDC持续关注全球智能眼镜及可穿戴设备市场的发展动态。我们诚邀行业同仁、投资机构及媒体朋友与IDC中国分析师团队保持沟通，共同探讨市场趋势、技术创新与商业机遇。无论您是希望深入了解数据细节，还是寻求定制化市场洞察，欢迎随时与我们联系。请点击此处与我们联系。

If it feels like buyers are bypassing your branded content, you’d be right.

Until recently, technology company websites were key stops for buyers on the information superhighway. That stopped being true the moment AI tools became buyers’ default starting point.

Your traffic is being rerouted onto the AI expressway: the high-speed road that eschews off-ramps and at-grade crosses. Since it’s built for speedy decision-making, the AI exit is enticing buyers faster than marketing teams can adapt their go-to-market strategies.

As technology companies struggle to keep pace, here’s where the shift is already showing up in sales funnels, starting with how buyers evaluate vendors today.

The buyer journey now focuses on AI

Here’s how fast buyers are merging onto the expressway. During the evaluation stage, 68% of buyers say they use AI always or often to compare products and solution features across vendors. In addition, 70% say they use it always or often to research industry and technology trends.

It’s easy to see why. Ask AI to evaluate a dozen vendors and it can research and analyze all of them at once. A buyer’s favorite AI tools can analyze vendor offerings according to specific preferences and deliver customized rankings – all within a couple of minutes.

What buyers aren’t doing as often is asking your sales team for information, even after they’ve narrowed the list down to two or three vendors.

The credibility gap increases buyer scrutiny

As buyers bypass your website and your sales team exerts less influence, your vendor content is losing to AI tools and analyst sources.

Nearly half, 46%, of today’s buyers say they turn to AI tools like ChatGPT and Gemini to evaluate vendors, while 45% trust industry experts and analysts. Yet only 33% rely on the vendor salesperson as an active evaluation source, and vendor website usage as an evaluation source has dropped from 66% to just 37%, as buyers increasingly treat vendor-published content as marketing rather than something to lean on.

This distrust has created a credibility gap.

Senior leaders now demand greater scrutiny of sources, and they’re creating checkpoints to verify claims before making a buying decision. Ninety percent of enterprise legal buyers are writing AI clauses into guidelines, creating formal triggers to challenge unsupported vendor claims.

Want the numbers on one page? Download the infographic for a quick look at the three stats reshaping how buyers evaluate vendors.

Leading the pack: Why some vendors are pulling ahead

So, buyers have moved. Their AI preferences have restructured the journey.

But vendors? Most haven’t adapted their go-to-market strategy, and they aren’t keeping pace. Their sales and website content is the equivalent of a once-loved, now-forgotten exit on the old information superhighway, still fully staffed while the traffic moved somewhere else.

But not every vendor is stuck on fading superhighways or dusty intersections. Some are adapting to the new AI reality and aligning their go-to-market strategies to match the AI-mediated buyer journey.

A small cohort is pulling ahead. Just 3.1% of organizations have reached the optimized AI maturity stage, up from just 0.4% a year ago. These vendors have moved quickly to close the gaps in strategic decision making, sales enablement, and data and governance.

For vendors struggling to keep up, these organizations offer clear signposts of where to head next. The good news? They don’t have bigger teams or excess resources. Instead, they’re restructuring their GTM strategies to deliver intelligence in the right places along the AI expressway.

3 questions to diagnose your GTM stack

Today’s technology vendors can accelerate their GTM transformations by understanding how these leaders have mapped their operations to the buyer journey. These diagnostic questions about your GTM stack are a good place to start.

  1. Have we invested in AI-infused data architecture to improve strategic decision-making and increase competitiveness?
  2. How far have we scaled production-ready AI tools for lead scoring, proposal generation, and coaching?
  3. What constraints are we experiencing in adapting to the new buyer journey?

Once you know where you stand, benchmark against the leaders and map your plan for meeting buyers on the AI expressway.

Find out who’s leading, why, and what you can do to adapt your go-to-market strategy to today’s AI-mediated buyer journey.

IDC’s full report breaks down the maturity gap by GTM function and includes IDC’s worldwide AI software market share data. Download it here.

Christina Cardoza - Content Marketing Manager - IDC

LinkedIn
Christina Cardoza is a Content Marketing Manager at IDC, where she specializes in brand content and social media strategy. With a background in journalism and editorial leadership, she has a proven ability to transform complex technology topics into clear, actionable insights.

Six months. Twenty-eight stakeholders. That’s the average enterprise sourcing cycle right now, according to IDC’s Effective Agent Cost Management: A Practitioner’s Framework. If that number sounds high, sit with it a second longer: 28 people means 28 opinions on what “good” looks like, 28 calendars to align, and 28 chances for the process to stall while everyone waits on someone else’s read of the market.

Procurement doesn’t lose sourcing cycles to a lack of vendors. It loses them to a lack of a fast, defensible way to narrow a field that keeps growing every quarter. You open a category review with 12 names on a long list. Three months later, you still have nine, because nobody on the team has the bandwidth or the standing to say, with evidence, who belongs and who doesn’t. This is the exact moment IDC Quanta earns its place in the process.

Ask the question, get the field narrowed

IDC Quanta is the interface between your team and IDC’s full research organization: the analyst expertise and market comparisons the sourcing platform has built, grounded in your own environment and history. You don’t file a request and wait for a report. You ask IDC Quanta a plain-language question, the kind you’d ask a colleague who happened to know the entire market: “Help me narrow this vendor field to five names that fit our environment.”

IDC Quanta takes that question, pulls in the relevant analyst expertise, layers in its market comparisons, and returns an answer built around your specific requirements, not a generic category summary. That distinction matters more than it sounds. IDC Quanta’s answer reflects your scale, your existing stack, and your budget reality.

Full roster, not one point of contact

Here’s what most procurement teams have learned to expect from research access: a single assigned analyst and a calendar to work around. IDC Quanta doesn’t route you to one expert and stop there. Ask about cloud cost management, and IDC Quanta surfaces the analysts who cover both the cost economics and the vendor viability questions, in the same session. You’re not waiting for your one contact to loop in a specialist. IDC Quanta already knows who else in IDC’s roster has the answer you need, and it brings them into the conversation.

For a category review with 28 stakeholders pulling in different directions, that breadth changes the conversation. Instead of one analyst’s opinion carrying the room, you bring evidence from the specific experts whose coverage matches the decision in front of you.

From wide field to structured shortlist

The sourcing platform behind IDC Quanta does the heavy lifting on comparison. Ask IDC Quanta to build a shortlist from 12 vendors, and it draws on the platform’s structured market data to produce a comparison built around the criteria that matter for your specific evaluation.

That’s the shift worth naming: IDC Quanta hands procurement an answer already interpreted against their own context.

Momentum is the real deliverable

Twenty-eight stakeholders don’t need fewer meetings. They need a faster path to shared confidence in the same evidence. IDC Quanta gives procurement a way to bring that evidence into the room before momentum stalls, with a clear, defensible reason for each of the five names on the shortlist.

Ask IDC Quanta the question. Let the roster and the platform do what they do best, and get back to running the evaluation.

See IDC Quanta for yourself

Ryan Smith - Content Marketing Director - IDC

Ryan Smith is the Director of Content Marketing at IDC, where he leads brand-level content and social media strategy, aligning research insights with compelling storytelling to engage technology decision-makers. With a background in both IT and marketing, Ryan brings a unique blend of technical understanding and creative strategy to his work. He’s also a seasoned storyteller, speaker, and podcast host who believes the right message, told the right way, can drive both trust and transformation.

Christina Cardoza - Content Marketing Manager - IDC

LinkedIn
Christina Cardoza is a Content Marketing Manager at IDC, where she specializes in brand content and social media strategy. With a background in journalism and editorial leadership, she has a proven ability to transform complex technology topics into clear, actionable insights.

Every technology leader has sat through a vendor pitch built on a survey. A few thousand respondents, a well-designed questionnaire, a chart that says most organizations plan to increase spend on something. That’s the quiet limitation of survey-based market intelligence: it measures intent, not action. It looks like evidence. It’s a snapshot of what people said they’d do, taken at a single moment, months before you’re reading it.

Before you trust a market intelligence source with a board-level investment decision, ask where its data comes from.

IDC Quanta is built on independent, continuously tracked research: the judgment of more than 1,000 experts across over 100 countries, benchmarked against a peer universe of 404,000+ enterprises spanning 52 countries and 41+ technology categories.

For years, the market intelligence pitch was about frameworks: maturity models, best-practice checklists, the how-to layer. That layer no longer differentiates one source from another, since a well-prompted generative AI tool can produce a credible maturity model today. The differentiator is data that doesn’t exist publicly, gathered and interpreted by analysts who track the market for a living.

A dataset is only useful if a technology leader can get an answer out of it without hiring a data science team. Quanta’s contextualization layer pairs your own documents and procurement history with IDC’s research, so a plain-language question gets an answer shaped by your specific situation. Upload an IT budget, an AI cost export, or a vendor contract, and IDC Quanta maps it against that data in the same conversation, with no separate portal or new login required. The more you use it, the more it remembers: your vendors and use cases stay active across quarters instead of resetting with every new report.

This isn’t an argument that frameworks are worthless. It’s that the edge lives in evidence a generative model cannot invent, delivered as a straight answer instead of a hundred-slide deck. That’s the confidence a tech leader needs walking into a board meeting or a renewal, with a number a vendor can’t talk you out of.

See IDC Quanta

Ryan Smith - Content Marketing Director - IDC

Ryan Smith is the Director of Content Marketing at IDC, where he leads brand-level content and social media strategy, aligning research insights with compelling storytelling to engage technology decision-makers. With a background in both IT and marketing, Ryan brings a unique blend of technical understanding and creative strategy to his work. He’s also a seasoned storyteller, speaker, and podcast host who believes the right message, told the right way, can drive both trust and transformation.

IDC’s Insights From the 2026 World Robot Conference in Beijing

The 2026 World Robot Conference (WRC) took place in Beijing this August under the theme “Human-Machine Symbiosis, Integrated Production and Demand.” More than 300 companies exhibited, showing over 2,000 products and launching more than 150 new ones. For the first time, the conference added a dedicated Procurement Day, where 49 state-owned enterprises appeared together as an innovation consortium, laying out real industrial demand across 12 application scenarios. Robotics is becoming more application-focused, and shifts on the demand-side are now driving the industry’s growth at scale.

The numbers back that up. China’s Ministry of Industry and Information Technology reports that revenue at the country’s above-scale robotics enterprises topped RMB 90 billion (USD 13.4 billion) between January and May 2026, up 26.9% year-on-year. IDC’s own data shows global humanoid robot shipments reached roughly 18,000 units in 2025, up nearly 800% year-on-year.

IDC’s research also finds that most enterprises piloting embodied intelligence robots are still stuck at the proof of concept stage. Between validating the technology and deploying it at scale sit several real hurdles: stability, cost, ROI, and the operations and maintenance work that keeps a robot running day to day. This year’s show floor told a more advanced story anyway. Robots demonstrated complete task execution rather than single task actions, and several ran in real business settings rather than labs. Physical AI is now competing in industrialization capability as much as technical capability. Four trends from this year’s conference bring that into focus.

Trend One: The Model Race Is Giving Way to Real-World Learning

Over the past year, large embodied intelligence models, VLA models, and world models have advanced quickly, pushing robots beyond perception and understanding into complex task decisions and execution. Heading into 2026, though, the competition is shifting: whose model is strongest matters less than which vendor can get robots working, and learning, in real scenarios. Models are the capability foundation for Physical AI; real-world data and continuous learning are what will define the next stage’s edge.

At WRC 2026, world models, VLA, simulation platforms, robot “training grounds,” and data collection technologies all drew sustained attention. Combining virtual and real environments is becoming an important path to data acquisition: simulation generates data at scale, real scenario data keeps it grounded, and together they lower the cost of trial and error. Operational data flowing back into the training pipeline forms a flywheel: perceive, train, execute, get feedback keep improving.

China already has more than 40 robot training grounds planned or under construction, spanning more than 14 provinces and municipalities. The most mature ones can produce data output in the millions of records a year. At the same time, steady improvement in dexterous hands, force sensors, and other hardware is giving robots the means to capture richer manipulation data. As models, data, simulation, and hardware start to co-evolve, enterprise competition is starting to hinge less on whose model scores best and more on who can turn operational data into a working training loop.

Trend Two: Demo Capability Is Giving Way to Scaled Delivery

As robots move into real business environments, the standard for judging them is changing too. At this year’s conference, several vendors moved past single-action demos to validate complete task workflows, including depalletizing, sorting, delivery, cleaning, and inspection. They are proving out full processes now, not isolated capabilities. The real dividing line in robotics has shifted from technical demos to stable delivery.

Task capability now means continuous operation, not single operation execution: robots need autonomous perception, task planning, continuous execution, and the ability to recover from exceptions on their own. Deployment is the harder test. Moving past a proofmeans users are raising the bar on stability, deployment efficiency, maintenance cost, and ROI, with automotive, 3C electronics, and logistics and warehousing serving as the key proving grounds. And the metrics buyers actually use have shifted with it: task success rate, continuous run time, cost per task, and payback period, not the technical benchmarks vendors cite in a demo.

IDC’s user research shows more than 20% of users have already moved from market attention into pilot exploration, and the pace of progression from proof of concept to scaled deployment is accelerating. Stable operation, rapid deployment, ongoing maintenance, and commercial ROI are the capabilities that will separate vendors over the next two to three years.

Trend Three: From Single Scenario Breakthroughs to Layered Deployment

Physical AI won’t scale across every scenario at once. Differences in technical maturity, environmental complexity, and commercial ROI are settling applications into a layered path: commercial services are exploring, industrial settings are expanding, and home and consumer use is still building toward the future.

  • Commercial services are in a deep exploration phase. Dining, hospitality, and retail all have clear task demand, but their environments are more complex than industrial settings, so they need stronger autonomous navigation, environmental understanding, and exception handling. Embodied intelligence is accelerating here, carefully.
  • Industrial manufacturing will scale fastest. Automotive, 3C electronics, and logistics and warehousing are relatively structured, with clearer task boundaries and strong demand to cut cost and improve efficiency. IDC data shows China’s industrial embodied-intelligence robot market was worth approximately RMB 5.74 billion in 2025, with industrial robots contributing about RMB 3.62 billion and humanoid robots about RMB 2.12 billion. Right now, this is where technical maturity and commercial value line up best.
  • The home and consumer market holds long-term potential. Home environments are highly unstructured, with complex, long-tail task types that demand more from a robot’s generalization ability, safety, interaction skills, and cost. The potential application scope is broad enough, though, and the long-term market opportunity is significant.

Robot applications will keep working their way from highly structured environments toward moderately and then loosely structured ones, and different robot form factors will keep finding sharper matches to specific scenarios as they go.

Trend Four: Whole Machine Competition Is Giving Way to System Capability

As Physical AI enters its industrialization phase, the robotics industry chain is expanding from a single hardware platform to include computing power, models, data, core components, and applications, all at once. At this year’s conference, companies from across the domestic and international industry chain participated deeply, and the 49 SOEs appeared together as an innovation consortium. Competition is shifting from single-technology supply to industry chain collaboration and joint scenario creation, and the boundary of enterprise competition is expanding into hardware-software collaboration and ecosystem building as a result.

  • Industry chain collaboration continues to deepen: chips, sensors, reducers, servo systems, dexterous hands, and other core components are accelerating collaboration with robot platforms.
  • Models and hardware are converging faster, as large model companies, robot manufacturers, and research institutions collaborate on VLA, world models, and motion control.
  • Industry support systems are accelerating. Dozens of embodied intelligence innovation centers have already launched across China, covering technology development, pilot-scale validation, data training, and scenario deployment.
  • International cooperation is widening the circle further: the conference attracted 30 international supporting institutions across research, engineering, industry, and investment.

The companies best positioned to move from point applications to scaled commercial deployment will be the ones with the strongest ecosystem-organizing capability, not necessarily the strongest single product.

IDC Outlook

Physical AI is moving from technology display to industrial value validation. Robotics companies’ core competitiveness will depend not just on technical sophistication, but on whether they can turn that technology into stable, replicable products and solutions with real commercial value. Physical AI is entering the deep waters of industrialization, and the next stage of competition in robotics will focus more closely on real demand and real value.

The next two to three years will be an important window, as embodied intelligence robots move from technology validation to scaled application. Different scenarios will show a layered deployment pattern: service scenarios will be first to explore application, industrial scenarios will become the key direction for scaled deployment, and the home and consumer market holds long-term development potential. At the same time, Physical AI will push industry competition from single hardware capability toward a systemic capability built on models, data, hardware, and applications together.

Upcoming Webinar: Global Robotics Outlook 2026-2027

WRC 2026 showed where robotics is headed technically. But a new US policy is redrawing who wins. Join IDC’s live webinar, How the US ban is reshaping the robotics market, on Wednesday, October 14, 2026 at 11:00 AM EDT. We’ll cover the 2025-2030 forecast for humanoid, autonomous mobile, and commercial service robots, the competitive landscape, and what the July 2026 US restriction means for your next move. Register now!

Can’t join live? Register anyway — an on-demand recording and translated access will be available afterward.

For Further Discussion

The robotics industry stands at a critical juncture, moving from technology validation toward scaled application. If you’re interested in Physical AI deployment pathways, embodied-intelligence scenario value assessment, or industry-chain competitive dynamics, visit IDC Physical AI Resource Center  for more resources or contact us for an in-depth discussion with IDC analyst team.

Navkendar Singh

Navkendar Singh - Associate Vice President, Client Devices & IPDS, IDC India

Analyst Profile
Navkendar Singh is a Associate Vice President with IDC India, based in Gurgaon. His research domains encompass deep-dive research and insights in and around mobile devices, smart homes, PCs, tablets, wearables, and the printing market in India, Bangladesh, and Sri Lanka.…

2026年，通用Agent成为AI市场最受关注的方向之一。OpenClaw成为现象级产品，海外有Anthropic Cowork、ChatGPT Work等，国内则有WorkBuddy、豆包工作、百度搭子、千问办公等产品，头部厂商纷纷整合产品线，把资源集中到通用Agent上，市场群雄逐鹿。

热度背后是AI应用范式的变革。与去年以Workflow为主的智能体相比，基于Harness架构的通用Agent使用门槛更低、处理复杂任务能力更强，也更容易被普通用户采用。

但市场上的通用Agent究竟能完成多少任务、不同产品各有什么优势、企业又该如何选型，目前还缺少基于统一任务、统一模型和统一测试条件的公开横向实测。

IDC今年启动系统研究，综合企业问卷、用户访谈、产品实测和厂商交流，对主流产品进行深度评测，帮助企业解决选型难题，也为厂商提供市场视角。

一、IDC如何定义“通用Agent”?

IDC将可独立提供服务的智能体分为通用Agent和垂直Agent两类：

  • 通用Agent不限定特定场景或固定任务，目前主要用于企业办公，包括信息检索与研究、文档生成与处理、数据处理与分析、网页与应用生成、沟通协作和日常事务管理，还可加载Skill、通过工具连接企业业务系统完成更多场景更专业的任务；
  • 垂直Agent则预设应用场景，多选择垂直专有模型或接入专业知识库，并配套专用工具和模板，完成专业任务。

Agent Harness是当前这类产品的核心框架，其负责组织上下文、调度工具与Skill、管理任务状态、控制执行环境，完成任务循环。不同产品执行同一任务，其步骤规则、循环处理的差别很大。在同一模型的统一评测下，产品差异很大程度上来自Harness工程能力。

二、企业级应用现状：普遍关注，但当前尚未规模应用

IDC对中国企业用户调研（n=200）显示，早在今年4月，已有合计48.5%的企业开始评估、试点或使用通用Agent。

从分行业看，互联网企业比例最高（83.3%），金融63.3%，电信、能源和媒体分别为55.0%、50.0%和50.0%。IDC在Q3的用户访谈显示，企业目前主要在日常办公场景试用，业务应用仍在建设；虽然尚未规模化落地，但几乎所有受访企业都高度关注，并将其视为未来企业级智能体建设的重要方向。

企业对通用Agent的要求可概括为“好不好用、好不好管”。

  • “好不好用”: 看任务完成效果、执行效率和交互体验，对应任务表现、成本效率、交互体验、多智能体编排、记忆与工作空间等评估维度；
  • “好不好管”: 看能否接入企业系统、安全可控、运行可追踪，对应可观测、安全可控、工具与技能、生态与开放性四个维度。

IDC的调研数据与框架基本一致：53.6%的企业把跨系统完成端到端任务列为评估通过Agent的首要驱动因素。其余为复杂任务自动化、数字员工、统一入口等能力；另一个问题中，52.0%的企业认为Agent规模落地的关键在于ROI的量化问题，使成本治理和效果观测成为选型重点。

在使用场景上，96.9%的企业用于办公自动化，流程自动化、智能客服（均80.8%）、知识管理与企业搜索（77.7%）、数据分析与商业智能（75.6%）紧随其后。

IDC据此设计近百道测试任务的Benchmark，覆盖邮件、日程、会议纪要、企业知识检索、财报分析、表格清洗、文档与PPT生成、视频生成、CRM与审批等真实场景，按复杂度分为常规任务（58.1%）和复杂任务（41.9%）。

三、好不好用：常规任务基本能打，复杂任务仍需谨慎

（以下内容仅展示部分实测评估情况，更多评估请查阅完整报告）

参评产品在统一模型（DeepSeek-v4 Pro）、统一测试环境（产品开箱状态）、统一工具接口和数据基线（统一仿真MCP Server）下完成近百道非公开测试任务。

（注：产品模型均统一配置为Deepseek-V4 Pro，TeleAgent因无法配置模型默认使用旗舰模式）

  • 常规任务：整体表现不错。常规任务目标明确、执行路径确定，主要包括基础文档输出与改写、邮件与日程安排、会议纪要总结和基础数据处理。多数产品能准确理解相对时间、识别联系人、定位信息，并把结果正确写回日历、待办、CRM等系统，一次交付即可使用的任务占比相对较高。
  • 复杂任务：能交付，但还不能直接可用。复杂任务涉及长上下文、长链路和多步骤循环（复杂PPT、复杂表格、浏览器操作、财报抓取与分析、视频生成等），大部分产品能花数十分钟至数小时交付，但结果通常不能直接用。问题集中在：结构与版式不到位；处理要求遗漏，链路越长漏得越多；长上下文内容常被压缩改写；数据错位与识别异常；并且整体执行代价偏高。例如某浏览器操作类任务平均消耗约473.5万Token，复杂的11页PPT生成平均消耗约353.7万（最高超1,000万）。

从“能交付”到“直接用”之间的差距，就是企业级智能体落地的最后一公里。

四、好不好管：管得住，企业才敢大规模应用

而在企业级场景落地智能体，除了Agent完成任务能力之外，企业更关注的是安全机制、权限控制、操作审计、执行稳定性、企业系统连接能力等企业级能力，这也是通用Agent是否能够真正进入企业级场景的前提。

安全可控与权限：当前大部分产品已能识别提示词注入，按风险等级自动放行、请求确认或阻断，高风险操作前二次确认，代码和浏览器在独立沙箱运行；企业级产品还需为Agent建立独立身份和权限体系，配置临时凭证和单独设定的操作范围。

运维与数据看板：企业主要关注三个维度，一是智能体使用普及率；二是成本观测、归因与治理（能否按任务、用户、部门、Agent统计用量并设置预算配额）；三是运行失败、停滞或异常调用时能否及时告警并给出处理建议。

数据与知识治理：Agent读写的是企业自己的文档、知识库和业务数据，哪些内容可进上下文、哪些只能内网处理、产物存哪里、保留多久都要提前确定，并建立智能体相关资产的创建、审核、授权和版本管理能力。

五、当前市场产品平均得分

当前业界普遍能力雷达：好用与好管维度仍需补齐

六、IDC的建议

给厂商的建议

  • 持续提高复杂任务完成能力。常规与复杂任务差距来自规划、长链路执行和多工具协同，都需Harness支撑；相同模型下产品得分仍有明显差异，说明Harness工程能力直接影响任务表现，是下一阶段竞争重点。
  • 把企业级能力覆盖到任务执行全过程。工作空间协同、身份权限、成本统计与预算管控、运行观测和审计留痕需同步建设，接入系统越多，企业对权限、证据和追溯的要求越高。
  • 建立场景交付和合作伙伴生态。厂商难独立完成垂直场景建设，应尽早建立合作伙伴网络；已跑通的任务应主动帮企业构建为Skill，既可在组织内扩散，也减少重复规划的Token消耗。
  • 提供更符合智能体的任务交互方式。复杂任务时长可达数小时，用户需同时发起多个任务或调度多个Agent，传统“一次会话对应一个任务”的交互难以集中管理，应支持并发任务状态监控、异常提醒和随时人工接管。

给企业的建议

  • 观望期也是窗口期。产品快速迭代，观望期同样是积累经验的窗口，企业可优先选择适合的日常办公场景试点，积累能力与组织经验后再深入业务流。
  • 按任务复杂度设计场景。规则明确、重复度高、结果易检查的任务交给Agent或Workflow；边界不清晰、链路较长的复杂任务交给通用Agent，同时安排专业人员检查并最终审核。
  • 建立覆盖任务全过程的治理机制。Agent接入的系统和工具越多，越要统一管理智能体身份权限，并与现有运维、安全和审计体系衔接。
  • 让组织能力跟上Agent能力。Agent会重新划分员工、管理者和系统的责任，企业需同步提升员工AI技能、管理层对能力边界的理解，并调整流程、协作方式和考核机制。企业对AI认知的上限，基本决定了Agent价值的上限。

更多内容请关注IDC报告：

IDC《中国企业级通用Agent产品技术评估，2026 Q3》（Doc# CHC54893926，2026年9月）

如果您希望了解人工智能相关的研究或开展进一步交流，欢迎点击 联系我们！

Zhenya Sun

Zhenya Sun - Research Manager

Analyst Profile
Zhenya Sun is a research manager for the IDC team focused on exploring the application of technology and industrial development of AI and AI agents. He is also responsible for providing clients with consulting services on technologies, products, and markets…

Global shipments of XR headsets and glasses rose 35.3% year over year in the second quarter of 2026, strong growth but a significant moderation from the 130.2% increase IDC tracked in the prior quarter. Much of that slowdown reflects tougher comparisons and a pause between major product cycles rather than any softening in consumer interest. Growth should pick back up as the next wave of product launches arrives later this year.

Audio glasses, the display-less category that includes devices like Ray-Ban Meta, represented the largest portion of shipments, capturing 70.3% share during the quarter, up from 61.9% a year ago. Display glasses represented 14.3% share, up from 11% a year ago, while headsets held the remaining 15.4%. Glasses overall have turned a corner and are quickly being recognized by a mainstream audience. Recent privacy backlash over always-on recording shows not all of that attention is flattering.

Who’s winning right now: Q2 2026 market share

Meta continued to dominate the market with 68.7% share. Its growth this quarter came from pushing into two additional segments: prescription eyewear, with the launch of its Blayzer and Scriber Optics lines, and lower-cost options that skip the fashion branding EssilorLuxottica normally provides. That expansion is only part of the story. Meta’s lead also rests on distribution scale, brand recognition, and AI features built across its product line, advantages that are harder for challengers to replicate than any single hardware launch.

CompanyQ2 2026 Market Share
Meta68.7%
RayNeo3.6%
Alibaba Group2.6%
XREAL2.5%
Xiaomi2.3%
Others20.4%

Source: IDC Worldwide Quarterly Wearable Device Tracker + Worldwide Quarterly Augmented and Virtual Reality Headset Tracker, 2026Q2

RayNeo ranked second with 3.6% share, up 83.3% year over year, on the strength of its low-cost, personal-theater-style devices. Alibaba, a relative newcomer to the category, ranked third at 2.6% share thanks to the popularity of Qwen and the company’s already-established presence in China. XREAL and Xiaomi rounded out the top five, though Xiaomi’s share nearly halved from a year ago as competition in its home market intensified.

What’s coming next: The launches that could reshuffle the board

A lot is going to change in the remaining months of 2026 as next-gen products arrive from multiple vendors, each chasing a similar set of goals: more capable AI experiences, lighter hardware, and use cases that go beyond entertainment into everyday productivity. Those shared priorities, more than any single launch, will shape competition through the rest of the year.

Snap is launching its Specs, which bring genuine AR capabilities packaged into the most consumer-friendly design the category has seen yet, backed by a marketing campaign built around fashion icons.

Samsung and XREAL are both readying Android XR products. XREAL’s Aura pairs the company’s personal-theater-style hardware, already strong for content consumption, with Google’s Android XR platform, putting productivity use cases on equal footing with entertainment. Samsung’s audio glasses, meanwhile, bring Google’s Gemini assistant together with the distribution and design sensibility of Gentle Monster and Warby Parker.

Meta’s upcoming Connect event is widely expected to launch a more premium mixed reality headset with better visuals and a slimmer, lighter form factor that should appeal to enthusiasts. IDC also anticipates new display and audio glasses at the show, with the display glasses aimed at broader availability and a more accessible price point.

Beyond these headline launches, products from Pico, RayNeo, Viture, and a growing list of other brands will help drive further growth in 2026 and into the years ahead.

The road ahead: Forecast 2026–2028

IDC forecasts total unit shipments will grow 26.3% in 2026, then accelerate to 40.5% growth in 2027 and another 26.4% in 2028. Audio glasses will keep leading the category, though display glasses and headsets will continue to ramp alongside them.

For consumers, that growth will be driven by a wider range of products, price points, and channels, plus the spread of AI and friction-reducing use cases like easy photo and video capture. Among businesses, training and design use cases will keep driving adoption, since IDC has tracked ROI gains in early enterprise pilots, and more businesses continue to start them.

Growing pains: Privacy and business-model tension

There are also growing concerns behind the scenes that need to be addressed, as they stand to reshape the adoption curve going forward.

Privacy and shifting consumer attitudes. Many products and services surrounding XR glasses have clear, tangible benefits, though IDC has tracked growing concern over the need for privacy controls. IDC has also observed a shift from early jokes about social acceptance to more serious concerns about the lack of meaningful notice and consent mechanisms while recording, along with AI-enabled data collection, raising the stakes for how vendors respond. Meta, the current leader, has taken action to reduce misuse of its products, but those steps came only after public backlash forced its hand. That leaves room for a competitor willing to build privacy safeguards in before launch, not after a headline forces the issue.

Competing business models. XR glasses are a relatively new category, and tech companies are quickly partnering with fashion companies to leverage their expertise in design, manufacturing, and distribution. There’s tension between these camps, though, as fashion companies often operate on higher margins while tech companies are willing to sacrifice margin in favor of ongoing subscription or advertising-based models, where the hardware can be offered at a lower upfront cost to keep the consumer inside the ecosystem. This is likely one reason Meta launched glasses without EssilorLuxottica’s branding. It also puts smaller players at a disadvantage, since they face pricing pressure on hardware from incumbents like Meta and potentially Google and Samsung, and will likely have to adjust their own business models to offer subscription services alongside their XR products going forward.

What this means for the industry

Smart glasses are genuinely being worn now, not just tested, and that changes the competitive picture. Meta still holds the lead, but the next round of launches from Snap, Samsung, XREAL, and Meta’s own Connect event will show whether that lead holds up. The rest of 2026 will be shaped as much by these launches as by the numbers themselves. Even with this quarter’s slower growth, consumer awareness keeps building and AI is unlocking new use cases — the vendors that move first on privacy may be the ones who convert that awareness into share.

Jitesh Ubrani

Jitesh Ubrani - Director, Consumer Devices Research

Analyst Profile
Jitesh Ubrani is a Director at IDC leading a team of analysts within the Worldwide Consumer Device Trackers group, covering wearables, augmented reality (AR), virtual reality (VR), tablets, phones, PCs, gaming, and smart home devices, with a focus on market…

Every SaaS vendor just had its head chopped off, and Salesforce went first. It voluntarily stepped up to the guillotine and let its own head go. Yes, that sounds a bit savage, but don’t worry. It’s wonderful. The entire SaaS industry is about to go headless, and I couldn’t be more excited.

For thirty years, the interface has masqueraded as the product. That disguise just came off.  The UI was always just the delivery mechanism, and it has quietly become one of the most expensive parts of enterprise software ownership.

Think about what a typical knowledge worker actually does in a day. They log into an ERP system to check a budget, then into a CRM to update an opportunity, then into an expense tool to file a report, then into an HR system to approve a request. Each application has its own navigation logic, its own field names, its own permissions quirks, and its own learning curve. Most of these systems are also loaded with three or four times the functionality any single user needs, which means the interface is optimized for the vendor’s full customer base, not for the person actually sitting in front of it. The result is a tax on productivity that has compounded for years.  Time is lost to switching, time is lost to relearning, and time is lost to hunting for the right screen inside the right app. This was never a minor annoyance. It was one of the largest hidden costs in enterprise software.

The UI was never the product

Oracle was among the first major vendors to say this out loud in a meaningful way. With its agentic applications announced earlier this year (one of the first live market examples of cross-application agents as explained in IDC’s Agentic Evolution of Applications framework), Oracle began dynamically assembling the interface itself, pulling the data, functionality, workflows, and logic a user needed from across whichever applications were relevant, rather than forcing the user to go find them. That was among the first real cracks in the old model. It correctly suggested that the fixed screen was a legacy constraint, not a design requirement.

Salesforce has now taken that idea and extended it a step further.  At Dreamforce 2026, held September 15 through 17 in San Francisco, Salesforce unveiled AIforce, which it is calling a live interface layer. AIforce does what Agentforce already does inside Salesforce (dynamically composing the data, workflows, and business logic a user needs on the fly), but it goes a step further. It extends that same capability outside of Salesforce entirely, into whatever interface a person already happens to be working in. Salesforce launched this with three surfaces at once: Claudeforce, Slackforce, and Agentforce Coworker, meaning the same trusted enterprise data and governance can now show up inside Claude, inside Slack, or inside a dedicated coworker agent, not just inside Salesforce’s own screens.

Meeting users wherever theyare

This is precisely why AIforce matters more than another feature release. The goal is to help customers break free of a shared user interface, so that every individual gets an intelligent, dynamic interface that comes to them, wherever they are, instead of forcing everyone into the same fixed screens. That is not a UI update. It is an admission that the UI was the obstacle all along.

What this means for the rest of the market

Oracle and Salesforce arrived at the same conclusion from opposite ends of the market, and that convergence is the real signal, not either company’s product name. When two competitors this large independently decide that the fixed screen is disposable, the SaaS UI is not being tweaked. It is being retired.

Buyers should not mourn it. The interface was never where the value lived. The value lived in the data, the workflows, and the logic sitting underneath it, and buyers have been paying a productivity tax for thirty years just to reach that value through a screen. That tax is now optional.

For vendors across ERP, HR, finance, SCM, and every other enterprise application category, the imperative is not subtle:

  • Stop pricing and marketing the interface as the differentiator. It never was.
  • Open the data, workflows, and business logic underneath the UI to whatever surface the customer already lives in, whether that is Claude, Slack, Teams, ChatGPT, Outlook, Office, or any other work environment.
  • Build the governance and permissions layer first. A composable interface without composable trust is a liability, not a feature.
  • Assume the next RFP will not ask what your UI looks like. It will ask whether your platform can show up anywhere.

The UI is not being replaced by a better UI. It is being replaced by the absence of one. That is not a loss. It is the correction the market has needed for thirty years.  I never thought I’d see the day when I applaud a beheading, but that day has arrived, and I hope it continues.

Eric Newmark

Eric Newmark - Group Vice President & General Manager of IDC's SaaS, Enterprise Software, CX and Workplace Solutions Division

Analyst Profile
Eric Newmark is Group Vice President & General Manager of IDC’s SaaS, Enterprise Software, CX, and Workplace Solutions Division, which includes several teams of analysts covering SaaS, 18 enterprise application markets, software monetization, business platforms, marketplaces, and services firms focused…