The blueprint starts with the workload, not a template.蓝图从负载出发，而非从模板出发。
Give the platform a GPU fleet and a growth curve. It returns rack topology, power architecture, a cooling strategy chosen against real thermal limits, fiber design, and a costed bill of materials — every number traceable to a simulation rather than a rule of thumb.向平台输入 GPU 规模与增长曲线，即可获得机柜拓扑、供电架构、基于真实热力极限选定的制冷策略、光纤设计，以及带成本的物料清单——每个数字都可追溯至仿真，而非经验法则。
What you provide您提供
- GPU type and quantityGPU 型号与数量
- Training and inference workload mix训练与推理负载配比
- Site constraints from Pre-planning来自预规划的站址约束
- Growth and phasing assumptions增长与分期假设
What the platform returns平台返回
- Rack count, power density, cluster topology机柜数量、功率密度、集群拓扑
- Power architecture and cooling strategy供电架构与制冷策略
- Network and fiber design网络与光纤设计
- Preliminary BOM and vendor shortlist初步物料清单与厂商候选
Five design decisions, solved against physics.五项设计决策，以物理求解。
AI capacity planning算力容量规划
Translate a GPU fleet into physical infrastructure: how many racks, at what density, in what cluster topology, occupying how much space.把 GPU 规模转化为物理基础设施：多少机柜、何种密度、什么集群拓扑、占用多少空间。
Power planning供电规划
Size the electrical chain against the dynamic behaviour of AI loads, not their nameplate. Busway thermal behaviour under extreme dynamic load is simulated, not estimated.按 AI 负载的动态行为而非铭牌功率来选型电气链路。母线排在极端动态负载下的热行为是仿真得出的，而非估算。
Cooling planning制冷规划
Choose a cooling approach against the thermal envelope the workload actually produces, then quantify what it costs in water and PUE.依据负载真实产生的热力包络选择制冷路线，再量化其在水耗与 PUE 上的代价。
- Air, direct liquid, immersion, or hybrid风冷、直接液冷、浸没式或混合
- Water usage and heat rejection水耗与散热量
- PUE estimate under seasonal conditions季节性工况下的 PUE 估算
Network & fiber planning网络与光纤规划
Produce a preliminary network design sized for the cluster topology rather than retrofitted to it.按集群拓扑生成初步网络设计，而非事后适配。
Preliminary BOM & vendors初步物料清单与厂商
Generate an early-stage bill of materials for budgeting, with vendor options filtered by region, lead time, compatibility, and cost. Vendor-neutral by design.生成用于预算的早期物料清单，并按区域、交期、兼容性与成本筛选厂商方案。架构上厂商中立。
- Power电力 — transformers, UPS, switchgear变压器、UPS、开关柜
- Cooling制冷 — chillers, CDU, CRAH冷机、CDU、CRAH
- Network网络 — fiber, cable, patch panels光纤、线缆、配线架
- Building建筑 — racks, busways, PDUs机柜、母线、PDU
The BOM is intended for early-stage planning and budgeting, not final procurement specification.该物料清单用于早期规划与预算，不作为最终采购规格。
Design becomes a live optimization设计成为实时优化系统
Because the physics engine returns a temperature field and current-density distribution in under a second, the design stage can compare thousands of configurations instead of validating one. When a parameter changes, the whole blueprint updates — and the same model carries forward into the digital twin that will run the facility.由于物理引擎可在一秒内返回温度场与电流密度分布，设计阶段得以对比上千种配置，而非仅验证一种。参数一变，整份蓝图随之更新——同一模型继续沿用为运营该设施的数字孪生。
See the simulation engine了解仿真引擎From infrastructure planning to infrastructure intelligence.从基础设施规划，到基础设施智能。
Bring a workload profile and a candidate site. We will run it through the platform and show you the power path, the thermal envelope, and the interconnection route.带上负载画像与候选站址，我们将在平台中完成推演，呈现供电路径、热力包络与并网路线。