华为技术有限公司正大幅提前其下一代人工智慧晶片的发布时程，以期在中国市场取代辉达并参与全球竞争。华为轮值董事长王涛于上海年会上宣布，旗舰款升腾 960DT 晶片将提前至 2027 年第一季推出，另一款 960PR 变体则预计于同年第三季交付。华为更规划于 2028 年推出升腾 970、2029 年推出升腾 980，力求每一代产品运算能力较前代翻倍，并透过 SuperPod 丛集技术与光学技术加速资料传输，以系统整合与网路能力弥补单晶片效能上的差距。
目前华为在中国人工智慧晶片市场已取得领先辉达的市占份额，并吸引了包括幻方量化旗下的 DeepSeek 等本土顶尖实验室进行大规模采购合作，同时亦在马来西亚与埃及等海外市场展开有限测试。然而，受限于制造产能吃紧与零组件供应瓶颈，华为目前难以完全满足庞大的订单需求，近期更将升腾 950DT 晶片报价上调六成。华为副董事长徐直军坦言，当前将优先保障供应中国本土客户，预计中国整体人工智慧硬体制造产能要到 2030 年才可能全面赶上市场需求。
面对美国长期的出口管制与技术封锁，华为积极引领中国科技自立自主的发展浪潮。为突破缺乏先进制程设备的困境，华为透过 LogicFolding 等创新架构技术精进晶片设计，并在升腾 950 系列中首次采用自主设计的记忆体晶片，以此摆脱对美韩先进记忆体供应商的依赖。尽管目前中国前沿实验室在模型训练端仍高度依赖走私或现有的辉达晶片，但华为强调自主造晶片是摆脱受制于人的必经之路，未来预计将有越来越多本土模型转向升腾生态系进行训练。
Huawei Technologies Co. is significantly accelerating the launch schedule of its next-generation AI processors to replace Nvidia Corp. in China and expand its footprint internationally. Rotating Chairman Wang Tao announced at the company's annual summit in Shanghai that the flagship Ascend 960DT chip will be introduced ahead of schedule in the first quarter of 2027, followed by the 960PR variant in the third quarter. Huawei further outlined plans for the Ascend 970 in 2028 and the Ascend 980 in 2029, aiming to double computing power with each successive generation while utilizing cluster-based SuperPod connectivity and advanced optics to compensate for single-chip performance gaps relative to Nvidia.
Domestically, Huawei has surpassed Nvidia in market share and garnered major deployment commitments from leading Chinese AI labs like DeepSeek, while simultaneously testing international demand in select markets such as Malaysia and Egypt. However, the company faces severe manufacturing constraints and component shortages that hinder its ability to fulfill massive volume orders, prompting a 60% price hike on its current Ascend 950DT chips. Deputy Chairman Eric Xu noted that domestic clients remain the top priority, acknowledging that China's overall production capacity for AI hardware and critical components is unlikely to satisfy comprehensive market demand until 2030.
Amid enduring US trade sanctions and export curbs, Huawei is spearheading Beijing's push for technological self-reliance across critical semiconductors. To mitigate hardware and equipment disadvantages, the company is implementing novel design strategies such as LogicFolding and outfitting the Ascend 950 lineup with in-house designed memory to circumvent cutoffs from global suppliers. Although premier Chinese labs still heavily lean on Nvidia hardware for foundational model training, Huawei views domestic chip production as an indispensable path toward technological autonomy and anticipates widespread migration of Chinese models onto the Ascend platform.