高通称6G商用路线图已启动，目标2029年推出
高通已开始执行其6G商用产品路线图，计划在三年内实现商用，与此前公布的时间表一致。公司预计6G将成为关键基础设施。
原文深度解读
本文报道高通在2026年骁龙峰会（Snapdragon Summit 2026，毛伊岛，9月22日当地时间）上，由技术规划、边缘解决方案与数据中心执行副总裁兼总经理Durga Malladi向韩国记者介绍6G商用路线图。高通已开始执行6G商用产品路线图，计划2028年推出预商用设备、2029年正式商用，并将6G定位为代理式人工智能（agentic AI）生态的关键基础设施。文章涉及高通Dragonfly数据中心产品组合（C1000服务器CPU、AI250与AI300 AI加速器、定制ASIC、HBC内存）、与电信运营商的合作、广域感知（wide-area sensing）测试，以及阿里巴巴Qwen Book端侧AI约30 tokens/秒的示例。与电子元器件供应链的关系在于：6G网络基础设施、数据中心CPU/AI加速器/定制芯片/HBC内存、RAN用低功耗高性能CPU与NPU等需求，可能影响相关芯片、内存与网络设备的供应结构与节奏。
- 高通已从展示6G愿景进入在高性能计算（HPC）产品组合内规划并执行实际商用产品路线图的阶段，Malladi称商用产品组合为Dragonfly，该品牌于6月首次公布。
- Dragonfly包含C1000服务器CPU、AI250与AI300 AI加速器、包括ASIC在内的定制芯片，以及High Bandwidth Compute（HBC）内存；高通正与电信运营商合作并分享该产品策略。
- Malladi表示电信网络需要用于RAN工作负载的低功耗高性能CPU和用于AI计算的NPU，且高通去年进入的数据中心业务的部分商用产品也将应用于电信网络基础设施。
- 高通将6G视为代理式AI生态的核心基础设施：代理式AI设备将实时收集视觉与语音数据并在无直接用户交互时传至云端，端侧AI与云端之间持续推理，随着设备与AI代理数量增加，上行数据流量将显著上升。
- 时间表方面，高通预计2029年正式6G商用，并预计在2028年洛杉矶奥运会前后使用预商用设备进行6G技术演示及新代理式AI服务试验；Malladi称2028年推出预商用设备、2029年正式商用，三年内实现6G商用的承诺不变。
- 高通正在圣迭戈原型测试台测试6G关键技术广域感知，并与主要电信运营商进行感知试验；广域感知将基站和无线网络变为大规模传感器，通过分析无线信号反射与传播路径变化实时识别空间状况、数据流和障碍物，结合AI可形成网络环境的实时数字表示（世界模型），并可能使网络从L0自主（工程师手动监控KPI和调整参数）向类似ADAS的L1、L2自主网络演进。
供应链影响
- 若6G商用按2028年预商用设备、2029年正式商用的节奏推进，相关网络基础设施与终端芯片、RAN用低功耗高性能CPU、NPU以及HBC内存等品类的需求节奏可能随之变化，具体取决于运营商投资与试验进展。
- 高通将数据中心业务的部分商用产品应用于电信网络基础设施，可能使数据中心与电信网络两条供应链在CPU、AI加速器、定制ASIC和内存等环节出现一定程度的交叉与协同需求。
- 代理式AI带来的持续推理与上行流量上升，可能推动网络侧与端侧对AI计算、连接与内存带宽相关元器件的结构性需求变化，但实际影响取决于代理式AI设备普及速度与运营商网络改造进度。
- 广域感知若从原型测试走向规模部署，可能增加基站与无线网络侧传感、计算与数据分析相关硬件的需求，但取决于技术成熟度、运营商试验结果与标准化进程。
- 电信运营商对6G准备态度从谨慎转向考虑代理式AI需求与业务扩展机会，可能影响其后续网络设备与芯片采购规划，但具体采购规模与时点取决于商用服务验证和自身业务策略。
系统解读，仅作为判断线索，不构成备货、出货、涨价或投资建议。
高通技术规划、边缘解决方案与数据中心执行副总裁兼总经理Durga Malladi与韩国记者交谈。（照片：Joint Press Corps，毛伊岛）
高通已开始执行其6G商用产品路线图，并计划按此前公布的时间在三年内实现商用。公司预计6G将成为代理式人工智能（AI）生态的关键基础设施。
高通技术规划、边缘解决方案与数据中心执行副总裁兼总经理Durga Malladi于当地时间9月22日在毛伊岛举行的2026年骁龙峰会上会见了韩国记者。“高通已经超越了仅仅展示6G愿景的阶段，”Malladi说。“我们已经进展到在高性能计算（HPC）产品组合内规划和执行实际商用产品路线图的阶段。”
Malladi提到的商用产品组合是Dragonfly，高通于6月首次公布。该数据中心品牌包括C1000服务器中央处理器（CPU）、AI250和AI300 AI加速器、包括专用集成电路（ASIC）在内的定制芯片，以及High Bandwidth Compute（HBC）内存。
高通正在与电信运营商合作，并向他们分享其Dragonfly产品策略。“电信网络需要用于无线接入网络（RAN）工作负载的低功耗、高性能CPU，以及用于AI计算的神经处理单元（NPU），”Malladi说。“高通去年进入的数据中心业务中的一些商用产品也将应用于电信网络基础设施。”
这意味着电信运营商正在建设能够在6G网络内直接执行和分发AI计算的基础设施。目标是演变为提供AI计算能力的更高价值智能基础设施提供商。
◇ 6G作为代理式AI的核心基础设施
高通将当今的5G网络视为代理式AI时代的起点。骁龙峰会上展示的几款代理式AI设备被展示为在5G环境中顺畅运行。
然而，一个成熟的代理式AI生态预计将需要6G。现有AI功能仅在用户请求时交换数据。代理式AI设备将实时收集视觉和语音数据，即使用户没有直接交互也会将其传输到云端。持续推理将在端侧AI与云端之间进行。
随着设备变得更加先进，且其中运行的AI代理数量增加，上行数据流量将显著上升。高通认为，专用6G通信基础设施对于可靠支持如此大规模的持续推理和连接至关重要。
“6G不仅仅是使峰值数据传输速度比5G提高十倍的技术，”Malladi说。“它是一种使多个AI代理能够在设备上持续运行并自主连接的技术。”他补充说，骁龙峰会上展示的代理式AI设备将为大约三年后演进到使用6G的产品提供技术基础。
阿里巴巴的Qwen Book在骁龙峰会上展出。其端侧AI在本地处理AI工作负载，并以约每秒30个token的速度提供响应。当代理式AI执行整体任务规划和复杂推理时，它也会整合云计算资源。
阿里巴巴是一家主要云服务提供商。美国和欧洲的其他云公司也在出现，并计划开发自己的代理式AI设备。“他们非常重视端侧AI推理以及与云端的持续连接，”Malladi说。
◇ 6G商用定于2029年，兼容设备预计2028年推出
高通预计2029年正式实现6G商用。公司预计在2028年洛杉矶奥运会前后，使用预商用设备进行6G技术演示，并试验新的代理式AI服务。
“正如我们在上一次峰会上所说，我们仍按计划在2028年推出预商用设备，并在2029年正式实现6G商用，”Malladi说。“我们在三年内实现6G商用的承诺依然坚定。”
电信运营商在今年年初之前对6G准备仍持谨慎态度。最近，随着他们考虑代理式AI需求以及扩展业务的机会，这种情绪开始改变。
高通目前正在其圣迭戈原型测试台测试广域感知这一关键6G技术。它还在与主要电信运营商进行感知试验。
广域感知将基站和无线网络本身变成大规模传感器，能够智能地收集和检测周围环境信息。通过分析无线信号反射和传播路径变化，网络可以实时识别基站周围的空间状况、数据流和障碍物。
通过感知收集的大量网络可见性数据可由AI分析，以创建“世界模型”，即网络环境的实时数字表示。该AI模型可作为自动化网络运营的核心数据库，可能改变网络管理的基本方式。
现有5G网络以L0自主水平运行，即工程师直接监控关键绩效指标（KPI）、评估网络可见性并手动调整网络参数。将广域感知数据与AI结合，可能实现类似高级驾驶辅助系统（ADAS）的L1和L2自主网络。RAN AI代理可以实时监控整个通信网络，并向工程师建议网络配置变更或直接实施这些变更。
Durga Malladi, Qualcomm executive vice president and general manager of Technology Planning, Edge Solutions and Data Center, speaks with Korean reporters. (Photo: Joint Press Corps, Maui)
Qualcomm has begun executing its commercial product roadmap for 6G and plans to achieve commercialization within three years as previously announced. The company expects 6G to serve as key infrastructure for the agentic artificial intelligence (AI) ecosystem.
Durga Malladi, Qualcomm executive vice president and general manager of Technology Planning, Edge Solutions and Data Center, met with Korean reporters at Snapdragon Summit 2026 in Maui on Sept. 22 local time. "Qualcomm has moved beyond simply presenting a vision for 6G," Malladi said. "We have progressed to the stage of planning and executing an actual commercial product roadmap within our high-performance computing (HPC) portfolio."
The commercial product portfolio Malladi referred to is Dragonfly, which Qualcomm first unveiled in June. The data center brand includes the C1000 server central processing unit (CPU), AI250 and AI300 AI accelerators, custom silicon, including application-specific integrated circuits (ASICs), and High Bandwidth Compute (HBC) memory.
Qualcomm is working with telecommunications operators and sharing its Dragonfly product strategy with them. "Telecommunications networks require low-power, high-performance CPUs for radio access network (RAN) workloads and neural processing units (NPUs) for AI computing," Malladi said. "Some of the commercial products from the data center business Qualcomm entered last year will also be applied to telecommunications network infrastructure."
This means telecommunications operators are building infrastructure capable of performing and distributing AI computing directly within 6G networks. The goal is to evolve into higher-value intelligent infrastructure providers that offer AI computing capabilities.
◇ 6G as Core Infrastructure for Agentic AI
Qualcomm views today's 5G networks as the starting point for the agentic AI era. Several agentic AI devices unveiled at Snapdragon Summit were shown operating smoothly in a 5G environment.
A full-fledged agentic AI ecosystem, however, is expected to require 6G. Existing AI features exchange data only when requested by users. Agentic AI devices will collect visual and voice data in real time and transmit it to the cloud even without direct user interaction. Continuous inference will take place between on-device AI and the cloud.
As devices become more advanced and the number of AI agents operating within them increases, uplink data traffic will rise significantly. Qualcomm believes dedicated 6G communications infrastructure will be essential to reliably support such large-scale continuous inference and connectivity.
"6G is not simply a technology that increases peak data transmission speeds tenfold over 5G," Malladi said. "It is a technology that enables multiple AI agents to operate continuously on devices and connect autonomously." He added that the agentic AI devices unveiled at Snapdragon Summit will provide the technological foundation for products that evolve to use 6G about three years from now.
Alibaba's Qwen Book was displayed at Snapdragon Summit. Its on-device AI processes AI workloads locally and delivers responses at about 30 tokens per second. When agentic AI performs overall task planning and complex reasoning, it also integrates cloud computing resources.
Alibaba is a major cloud service provider. Other cloud companies in the US and Europe are also emerging with plans to develop their own agentic AI devices. "They place great importance on on-device AI inference and persistent connectivity with the cloud," Malladi said.
◇ 6G Commercialization Set for 2029, Compatible Devices Due in 2028
Qualcomm expects formal 6G commercialization in 2029. The company expects demonstrations of 6G technologies using pre-commercial devices and trials of new agentic AI services around the 2028 Los Angeles Olympics.
"As we said at the previous summit, we remain on track to launch pre-commercial devices in 2028 and officially commercialize 6G in 2029," Malladi said. "Our commitment to commercializing 6G within three years remains firm."
Telecommunications operators remained cautious about preparing for 6G through early this year. That sentiment has recently begun to change as they consider demand for agentic AI and opportunities to expand their businesses.
Qualcomm is currently testing wide-area sensing, a key 6G technology, at its prototype testbed in San Diego. It is also conducting sensing trials with major telecommunications operators.
Wide-area sensing turns base stations and wireless networks themselves into large-scale sensors capable of intelligently collecting and detecting information about their surroundings. By analyzing wireless signal reflections and changes in propagation paths, networks can identify spatial conditions, data flows and obstacles around base stations in real time.
Large volumes of network visibility data collected through sensing can be analyzed by AI to create a "world model," a real-time digital representation of the network environment. The AI model can serve as a core database for automating network operations, potentially changing the fundamental way networks are managed.
Existing 5G networks operate at an L0 autonomy level, where engineers directly monitor key performance indicators (KPIs), assess network visibility and manually adjust network parameters. Combining wide-area sensing data with AI could enable autonomous L1 and L2 networks, similar to advanced driver-assistance systems (ADAS). RAN AI agents could monitor the entire communications network in real time and either recommend network configuration changes to engineers or implement them directly.