East Data, West Computing, and Descending Labor 东数、西算与下沉的劳动

China should not be drawn as a single labor node in the global AI map. Its AI system has an internal geography: platform cities concentrate models, cloud contracts, capital, and regulatory proximity; western computing hubs absorb data-center demand; local firms, parks, and schools organize annotation, moderation, testing, and training work.

The visible side of Chinese AI is metropolitan. Beijing gathers Baidu, ByteDance, Kuaishou, Zhipu, Moonshot, Speechocean, and Datatang. Hangzhou carries Alibaba Cloud and Qwen. Shenzhen links Tencent, Huawei, logistics platforms, hardware supply chains, and deployment teams. Shanghai concentrates SenseTime, MiniMax, media platforms, fintech, and enterprise AI. These cities name models, products, benchmarks, platforms, and procurement markets.

"East Data, West Computing" adds a second route. The program organizes eight national computing hubs and ten data-center clusters across Beijing-Tianjin-Hebei, the Yangtze River Delta, the Greater Bay Area, Chengdu-Chongqing, Inner Mongolia, Guizhou, Gansu, and Ningxia. It does not describe hidden labor directly, but it makes the cloud local: electricity, land, cooling, operations, security, and maintenance become part of the geography behind AI. "West computing" is therefore not a neutral technical relocation. It is a way of moving energy demand, infrastructure investment, and operational responsibility into places where land, power, and policy incentives can be assembled at another scale.

The third route is labor descending into local systems. This work is broader than "data labeling": drawing boxes around cars and medical images, segmenting roads or tumors, cleaning OCR, transcribing speech, ranking model answers, testing sensitive prompts, reviewing harmful or politically risky content, and monitoring AI customer-service replies. Firms such as Speechocean, Datatang, Testin, Appen/Aupeng, BasicFinder, and Magic Data mediate parts of this service layer.

"Employment transition" is a polite phrase for this rearrangement. Public discussions of AI often follow models, startups, platforms, agents, and elite opportunities in first-tier cities. At the same time, online work such as training, labeling, moderation, testing, and maintenance is moving into second- and third-tier cities, local industrial parks, and training systems. Many of the people pulled into this round of AI development are young graduates, vocational-school students, or first-time digital workers. They contribute time, knowledge, and attention as the trainers, maintainers, and support systems behind AI.

The receiving geography is concrete but uneven. Guiyang and Gui'an make data centers, big-data services, and annotation part of local development. Chongqing, Chengdu, Zhengzhou, Taiyuan, Xi'an, Lanzhou, Urumqi, Datong, Yichang, Nanjing, Ningbo, and Rizhao appear as data-service bases, training sites, or industry-education projects. The result is not one belt, but a patchwork of provincial capitals, second-tier cities, county parks, university towns, and digital-economy zones.

Vocational education is one hinge in that patchwork. AI trainer entered China's official new-occupation vocabulary in 2019, and since then data annotation, digital service, quality inspection, prompt testing, office automation, and AI customer-service training have appeared in public education and industry-education materials. The point is not simply that schools teach new tools. They help convert a national AI strategy into a local labor pipeline.

The political edge is content moderation. In China, moderation is not only platform hygiene or brand safety; it is also liability, political compliance, recommendation-system training, and the daily maintenance of permissible speech. The title's three movements therefore matter: data is sent west, computation is reorganized, and labor descends into local contractors, schools, and platforms. Yet these workers are not only passive entrants into the system. Like those who use AI in large cities to build products and opportunities, they are also imagining a future of their own. Value returns upward as model and platform capability; hope, uncertainty, and risk remain unevenly distributed.

如果把中国只画成全球 AI 劳动地图上的一个点，这张图就太简单了。中国 AI 有自己的内部地理： 平台城市集中模型、云合同、资本和治理近邻；西部算力枢纽承接数据中心需求；地方企业、园区和学校 组织标注、审核、测试和训练劳动。

可见的一面在都市中形成。北京有百度、字节、快手、智谱、月之暗面，也有海天瑞声、数据堂这样的训练数据企业。 杭州连接阿里云和通义千问；深圳连接腾讯、华为、物流平台、硬件供应链和落地团队； 上海聚集商汤、MiniMax、媒体平台、金融科技和企业 AI。这些城市命名模型、产品、benchmark、平台和采购市场。

“东数西算”加入了第二条路径。这个工程组织八个国家算力枢纽和十个数据中心集群，覆盖京津冀、长三角、 粤港澳大湾区、成渝、内蒙古、贵州、甘肃、宁夏等地。它不直接等同于隐形劳动，但它让“云”变得有地方： 电力、土地、冷却、运维、安全和维护，都成为 AI 背后的空间基础设施。因此，“西算”不是中性的技术搬迁， 而是把能源需求、基础设施投资和运营责任，放到土地、电力和政策激励能够被重新组合的地方。

第三条路径是劳动下沉。“数据标注”这个词太窄了：有人给汽车和医学影像画框，有人分割道路和肿瘤， 有人修 OCR、转写语音、排序模型回答、测试敏感提示、审核有害或政治风险内容，也有人监控 AI 客服。 海天瑞声、数据堂、Testin 云测、澳鹏、标贝科技、BasicFinder 等企业，正是在不同位置中介这层服务。

“就业转型”其实是一个很礼貌的词。今天很多人谈 AI，谈的是模型、创业、平台、Agent， 以及一线城市里那些看起来很“高大上”的机会。但与此同时，训练、标注、审核、测试、维护等线上劳动， 正在进入很多二三线城市、地方产业园区和培训体系。真正被卷入这一轮 AI 发展的人，很多是刚毕业的年轻人： 有人来自职业学校，有人刚离开校园，有人第一次进入数字劳动体系。他们贡献自己的时间、知识与精力， 成为 AI 背后的训练者、维护者与支持系统。

承接端的地理很具体，但并不整齐。贵阳和贵安把数据中心、大数据服务和标注做成地方发展身份。 重庆、成都、郑州、太原、西安、兰州、乌鲁木齐、大同、宜昌、南京、宁波、日照等地， 则以数据服务基地、实训项目或产教融合的形式连接 AI 产业。这不是一条完整产业带，而是省会城市、 二线城市、县域园区、大学城和数字经济新区的拼贴。

职业教育是这个拼贴中的关键铰链。2019 年，“人工智能训练师”进入中国新职业话语； 此后，数据标注、数字服务、质检、提示词测试、办公自动化和 AI 客服训练，频繁出现在公开的教育与产教融合材料中。 这不只是学校新增课程，而是把国家 AI 战略转译成地方劳动管道。

内容审核让这张图更尖锐。在中国，审核不只是平台卫生或品牌安全，也关乎责任、政治合规、 推荐系统训练和可见言论的日常维护。所以标题里的三种运动很重要：数据被送往西部，算力被重新组织， 劳动下沉到地方外包商、学校和平台系统；价值则以模型能力和平台能力的形式向上回流。 但这些年轻人并不只是被动进入这个体系。和那些在大城市里使用 AI、创造产品与机会的人一样， 他们也在期待一种属于自己的未来。只是这种未来同时带着希望、不确定性和风险。