CIOS is the Compute Infrastructure Operating System for every MDCX data center. Traditional DCIM watches UPS, cooling, and racks. CIOS watches what pays: utilization, PUE, cost, and revenue.CIOS 是每一座 MDCX 数据中心的算力基础设施操作系统。传统 DCIM 看的是 UPS、制冷与机柜；CIOS 看的是能挣钱的指标：利用率、PUE、成本与收入。
Every MDCX data center ships with its operating system. Hardware blocks deliver capacity; CIOS turns that capacity into operated, metered, revenue-ready infrastructure.每一座 MDCX 数据中心都随附自己的操作系统。硬件模块交付容量；CIOS 把容量变成可运营、可计量、可创收的基础设施。
CIOS is built for modular data center concepts — Tank, CDU, immersion, pod — not generic building BMS. Assets are path-addressed so every sensor and setpoint has a stable identity:CIOS 面向模块化数据中心的概念构建——液舱、CDU、浸没、Pod——而非通用楼宇 BMS。资产按路径寻址，每个传感器与设定点都有稳定身份：
The product ladder is deliberate: first make the site legible, then make it operable under policy, then make usage and capacity exportable for commercial decisions. Autonomous / unattended operation is Roadmap — not claimed as shipped.产品阶梯是有意为之：先让站点可读，再让它在策略约束下可操作，最后让用量与容量可导出以支撑商业决策。无人值守自治运行属于路线图——并未宣称已交付。
Licensing is open-core: Apache-2.0 core with commercial surfaces. One line for the site; details off-site.许可采用开放内核模式：Apache-2.0 内核加商业化模块。站点上只写一行，细节另行提供。
Plugin drivers ingest field protocols into a gateway stream and time-series store. The asset model speaks modular data center language, not generic racks.插件式驱动将现场协议接入网关数据流与时序库。资产模型使用模块化数据中心的语言，而不是笼统的「机柜」。
Modbus TCP and SNMP v2c shipped. Path-addressed points map into the site asset tree.已交付 Modbus TCP 与 SNMP v2c。路径寻址的点位映射进站点资产树。
Pods, CDUs, immersion tanks, and loops as first-class objects — not a flat tag dump.Pod、CDU、浸没液舱与回路都是一等对象——不是一堆扁平点位。
3D scene mapping of your pods is a capability of the lab pipeline — no Omniverse claim.Pod 的 3D 场景映射是实验室流水线的能力——不作 Omniverse 相关承诺。
Operations workflows turn plant state into work that can be tracked, SLAd, and closed — without free-for-all remote control.运维流程把机房状态转化为可跟踪、可考核 SLA、可闭环的工作——而不是放开的远程控制。
Alarm engine opens tickets with lifecycle, SLA scan, and notify (webhooks / email).告警引擎自动开单，具备生命周期、SLA 扫描与通知（webhook / 邮件）。
Schedules, inspections, spares, and maintenance windows that suppress alarms when work is planned.计划、巡检、备件与检修窗口；计划内作业期间自动抑制告警。
CLI cios query / cios set with risk classes — not open write access.命令行 cios query / cios set 带风险分级——不是开放写权限。
CIOS meters what the plant consumes and produces so commercial teams can act. It does not invoice.CIOS 计量机房的消耗与产出，供商务团队决策。它不做开票。
Records, rollups, and export for kWh and GPU-hour style usage — for reconciliation and customer portals.面向 kWh 与 GPU 小时类用量的记录、汇总与导出——用于对账与客户门户。
See where the site still has room before the next block lands.在下一套模块落地前，看清站点还有多少余量。
Ops portal for operators; customer portal for status, SLA, and usage. No billing or ERP claims.运维门户面向运维人员；客户门户提供状态、SLA 与用量。不涉及计费或 ERP 承诺。
Edge-first: a single box can run the site plane offline for ≥7 days. Multi-site aggregation is explicit Roadmap.边缘优先：单台设备即可离线运行站点平面 ≥7 天。多站点聚合明确标注为路线图。
Honest strip — not a promise calendar.如实陈列——不是承诺时间表。
DCGM and per-GPU metrics beyond facility path addressing.在设施路径寻址之上，接入 DCGM 与单卡级指标。
NOC-style aggregation across sites.跨站点的 NOC 式聚合。
Predictive maintenance and capacity forecast assistants.预测性维护与容量预测助手。
Hardware without an operating system is just metal. Pick a block; CIOS ships with it.没有操作系统的硬件只是一堆金属。选定模块，CIOS 随之交付。
Model the economics in TCO, or explore the hardware blocks.用 TCO 测算经济性，或浏览硬件模块。