中国大模型产业正经历Token（词元）从技术参数向交易资产的转变。1中国日均Token调用量在两年内实现了惊人增长，从2024年初的1000亿激增至2026年3月的140万亿，增幅达千倍。1这一爆发式增长反映了AI模型应用的快速扩张，也推动了围绕词元的新型商业生态形成。
国家层面对Token交易的支持力度不断加大。1国家数据局在《关于推进行业高质量数据集建设行动的实施方案（征求意见稿）》中首次官方提出"探索词元交易等新型交易模式"。12026年4月，青岛数据集团率先落地全国首个词元计量收益分配体系，标志着Token资产化探索进入实践阶段。1
企业层面的提价与扩张也加剧了这一趋势。1智谱在推出GLM-5时累计提价幅度达83%，同时API调用量激增400%。1芯片租赁市场同样出现快速涨价，英伟达H100一年期租赁价格近半年上涨近40%，Blackwell系列芯片租金暴涨48%。1在AI内容制作领域，120分钟的漫剧成片算力成本从2万元起步，部分团队单部剧消耗甚至达到十多万元。1
China's artificial intelligence sector is witnessing a dramatic transformation in how computational resources are valued and traded, with tokens—the basic units of language model processing—evolving from technical parameters into tradable commercial assets.1 Daily token invocations in China surged from 100 billion in early 2024 to 1.4 quadrillion by March 2026, representing a thousand-fold increase over two years.1
This explosive growth has prompted official recognition of a new market opportunity. China's National Data Bureau included "exploring novel transaction modes for tokens" in its implementation plan for promoting high-quality industry datasets, marking the first official government endorsement of token trading mechanisms.1 Private enterprise has moved quickly to capitalize on this shift: in April 2026, Qingdao Data Group became the first to deploy a nationwide token-based revenue distribution system.1 These developments reflect how large language model providers are simultaneously monetizing through price increases—Zhipu's GLM-5 cumulative price increase reached 83 percent while API call volumes surged 400 percent—and through new token-centric business models.1
The infrastructure supporting this economy is experiencing parallel pressures. NVIDIA H100 GPU annual rental prices climbed nearly 40 percent in six months, while Blackwell series chips saw rental costs jump 48 percent.1 Meanwhile, downstream AI applications are generating substantial computational demand: producing a 120-minute AI-animated series requires computing costs starting at 20,000 yuan, with some productions consuming over 100,000 yuan per episode.1
评论
还没有评论，欢迎留下第一条。