模型对话端侧硬件-2026年如何看待端侧落地

Summary of Key Points from the Conference Call Industry Overview - The AI eyewear market is experiencing rapid growth, with Meta and Luxottica's collaboration expected to sell over 7 million units in 2025 and reach 15 million units in 2026, doubling year-on-year [2][3] - The AI smartphone market is anticipated to flourish in 2026, with multiple manufacturers actively developing products that demand higher chip processing capabilities, NPU processing power, and memory capacity, potentially triggering a consumer upgrade wave [2][4][5] - OpenAI is making significant strides in edge hardware, including AI PINs, headphones, glasses, and desktop robots, aiming for an annual shipment volume of 100 million units, leveraging its vast user base to boost hardware sales [2][6] Core Insights and Arguments - 2026 is projected to be a pivotal year for AI edge hardware, driven by advancements in model upgrades and multimodal technology, leading to a shift from traditional voice-interactive devices to multimodal interactive hardware [3] - The introduction of new display-enabled AI eyewear is expected by 2027, capable of photography, video recording, and image presentation, enhancing user experience [3] - The demand for consumer electronics in the first half of 2026 may be affected by fluctuations in memory prices, but a recovery in market demand is expected in the second half as more mature models or agents are launched [7] Potential Beneficiaries in the Market - Domestic manufacturers are primarily focused on storage chips, with companies like Baiwei, Jiangbolong, and Zhaoyi Innovation leading the field. Key component suppliers such as Huanxu Electronics, Longqi Technology, and GoerTek are also expected to benefit from the edge hardware market [2][8] - Traditional consumer electronics companies like Luxshare Precision and Lens Technology are deeply involved in this sector and are likely to profit from the emerging opportunities [8] Additional Important Insights - The trend of model miniaturization is crucial for the next two years, enhancing model evaluation and enabling a wider range of applications, particularly in companion and recording devices [4][9] - The development of native multimodal technology is expected to integrate understanding and generative capabilities, leading to breakthroughs within 1-2 years, which will support the rise of larger and more complex devices like smartphones and NAS [9][10] - Record-type devices are seen as having the most explosive potential in the big model era, serving as both data entry points and sources of data generation, with AR/VR glasses and various smart devices poised for significant growth [10]