华为在近日召开的媒体沟通会上指出，过去半年内，token 消耗量增长了 6 倍，模型调用次数更是增加了约 50 至 100 倍。[1]华为中国政企互联网行业总经理黄健表示，"在算力、存力和运力三者中，运力是目前最突出的瓶颈"。[1]
针对这一瓶颈，华为采取了多项技术改进措施。该公司将 NPO 带宽提升了八倍以上，Token 产生效率提升了五倍，强调应通过网算存协同一体化而非单纯执着于芯片制程来优化系统性能。[1]华为为不同规模的互联网企业提供专属服务体系，并投入数千人在主流开源软件适配工作中。[1]华为预计在 2027 年至 2030 年间，token 经济将全面影响所有作业行为及工作框架。[1]
Huawei has identified network capacity as the most critical constraint in artificial intelligence infrastructure, according to statements made during a media briefing. [1] Over the past six months, token consumption has increased sixfold, while model invocation frequency has jumped by 50 to 100 times. [1] Among computing power, storage capacity, and network capacity, Huang Jian, Huawei's general manager for China's government, enterprise, and internet sectors, emphasized that "network capacity is currently the most prominent bottleneck." [1]
Rather than focusing narrowly on single-chip manufacturing processes, Huawei advocates for an integrated approach combining computing, storage, and networking capabilities. [1] The company has expanded NPO bandwidth by over eightfold and improved token generation efficiency by fivefold. [1] Huawei is providing customized service systems tailored to internet enterprises of varying scales. [1] Looking ahead, Huawei projects that between 2027 and 2030, the token economy will comprehensively influence all operational behaviors and work frameworks. [1] The company has committed thousands of personnel to adapting mainstream open-source software. [1]
