报告认为，超预期的AI基础设施需求改善了中资数据中心运营商的增长前景，但随着订单转为长期合约产能，板块正由成长故事转向资本配置故事，投资重点从获取MW产能或合约转向创造有利的资本回报。报告提出ROIC框架，比较三类资本密集度递增的模式：传统powered shell、提供长期GPUaaS合约的集成AI数据中心（AIDC）以及AI代币工厂（token factory）。传统powered shell回报偏低，基准ROIC约5%（杠杆股权IRR约8%），更高回报依赖财务工程、借助REIT回收资本及出海；长期GPUaaS的AIDC需达到收入与单位资本开支门槛以实现15%年化ROIC；轻资产代币工厂ROIC约21%，自有GPU代币工厂约43%。总体上，运营商越接近AI负载变现，ROIC潜力越高，但执行、技术与盈利风险相应上升。
Stronger-than-expected AI infrastructure demand has improved the medium-term growth outlook for China data center operators. However, as order books increasingly convert into long-duration contracted capacity, the sector is becoming less a traditional growth story and more a capital-allocation story: the next stage of China data center investing is less about securing MW capacity or contracts, and more about generating favorable capital returns. The report presents a ROIC framework evaluating and comparing three increasingly capital-intensive data center business models: 1) traditional powered shells, 2) integrated AI data centers (AIDCs) providing GPUaaS under long-term contracts, and 3) AI token factories. Powered shells remain lower-return assets, with a base-case ROIC of c.5% (levered equity IRR c.8%), where higher equity returns depend on financial engineering, capital recycling via private/public REITs, and going global. Integrated AIDCs with long-term GPUaaS contracts require re