MDCX builds and operates modular data centers for AI compute. Every block is a standard configuration — which is why it ships in ninety days, and why it holds its value when you move it.MDCX 为 AI 算力建造并运营模块化数据中心。每一个模块都是标准配置——这是它能在九十天内出厂的原因，也是它被搬走后仍然保值的原因。
Every AI infrastructure project asks you to commit — to a timeline, a site, and a number — while the things that determine whether you were right are still unknown.每一个 AI 基础设施项目，都要求你先承诺一个时间、一个场地、一个金额——而决定这些承诺对不对的因素，此刻都还是未知。
GPUs turn over in about a year. A conventional build runs 18–36 months. By the time the room is energized, the generation it was designed around has already been replaced — and the density envelope it was sized for is no longer the one you need.GPU 大约一年一代，传统建设需要 18–36 个月。等到机房带电，当初据以设计的那一代芯片已经换过，当初据以定尺的密度包络也不再是你需要的那个。
The fastest sites are the ones where power already exists — beside generation, or where capacity is already allocated. You skip the interconnection queue, which is usually the longest pole in the schedule. What those sites don't have is a local construction supply chain, permitting precedent for what you're bringing, or people on site to run it.最快的场地，是电力已经就位的场地——发电侧，或容量已分配的地方。你跳过的是并网排队，而并网通常是整个进度表上最长的一根杆。这些场地缺的不是电，是当地的土建产业链、针对你要运进来的东西的审批先例，以及现场运维的人。
And what you're allowed to bring differs by authority. Whether lithium batteries may sit inside the compute enclosure is a local determination — not a technical one.而且，允许你运进来什么，各地主管部门口径不同。锂电池能否置于算力箱体内，是属地认定的问题，不是技术问题。
The larger the single commitment, the more assumptions it locks in — demand, density, generation, price. Those assumptions expire long before the asset finishes depreciating.一次性承诺的规模越大，锁进去的假设就越多——需求、密度、代际、价格。这些假设的失效时间，远早于资产折旧结束。
Modularity doesn't remove these constraints. It does one thing: it makes every decision smaller, and later.模块化不会让这些约束消失。它只做一件事：让每一个决定都更小、更晚。
Every MDCX data center is assembled from factory-built Liquid Cooling and Immersion Cooling blocks, run by CIOS — and its spare capacity will earn on DCM. 每一座 MDCX 数据中心由工厂预制的液冷与浸没液冷模块组装，由 CIOS 运行——富余算力将在 DCM 上变现。
Model CapEx, power, utilization, and IRR for the line you just picked — before you lock a block.按你刚选的产品线测算 CapEx、电价、利用率与 IRR——先算清楚再锁定模块。
One pane for the whole site — telemetry, alarms→tickets, usage metering. Included with every block.全站一屏：遥测、告警→工单、用量计量。随每一套模块交付。
Sell spare GPU capacity — spot jobs, plans, and reserved blocks on one marketplace.出售闲置 GPU 算力：现货任务、套餐与预留容量，统一市场撮合。
How we choose these blocks — Why MDCX → · Specs — Docs · Standards · USD 模块选型逻辑 — 为何 MDCX → · 规格 — 文档 · 标准 · USD
We publish the two segments we control, and we don't guess at the one we don't.我们只发布我们控制的两段；不控制的那一段，我们不替你猜。
Expansion blocks: 90 days ex-works.扩容模块：90 天出厂（EXW）。
Every MDCX block is a standard configuration. There is no bespoke engineering cycle, no one-off bill of materials, no re-approval. The unit that ships is the unit that was tested.每一个 MDCX 模块都是标准配置。没有一次性的工程设计周期，没有一次性的物料清单，没有重新审图。出厂的那一台，就是被测试过的那一台。
Configuration happens at the system level, not the box level: how many blocks, how they're arranged, where the power comes from, and where you draw the boundary of the enclosure. The block itself doesn't move.配置发生在系统层，不在箱体层：几块、怎么排、电从哪来、箱体边界划在哪。箱子本身不动。
Power, infrastructure, and compute — run by one operating system, sold as one revenue-ready service. 电力、基础设施与算力 — 由同一操作系统运行，作为可创收服务交付。
Consulting · Design · Deployment · Care+ · Model optimization · Brokerage — the services layer is where per-project configuration happens. How we work →咨询 · 设计 · 部署 · Care+ · 模型优化 · 撮合 —— 服务层是承接单项目配置的地方。我们如何协作 →
Differentiated redundancy — 2N where failure hurts, N+1 where it doesn't. Availability where it matters, without blanket-2N cost. 差异化冗余 — 关键路径 2N，其余 N+1。在需要处保证可用性，避免全盘 2N 成本。
MDCX designs, builds, and operates modular data centers for AI compute — turnkey, from strategic planning through equipment supply to CIOS-driven operations. Factory-built liquid-cooled blocks cut deployment from years to months and are engineered for measured ROI: model yours in the TCO wizard. MDCX 为 AI 算力设计、建造并运营模块化数据中心 — 交钥匙交付，从战略规划、设备供应到 CIOS 驱动运维。工厂预制液冷模块将部署从数年压缩到数月，并以可测算 ROI 为目标：可在 TCO 向导中建模。
Headquartered in Sunnyvale, California, with our APAC hub in Singapore, we work with global partners to serve clients in over 40 countries. 总部设于加州 Sunnyvale，亚太枢纽在新加坡，与全球伙伴协作，服务 40 余个国家的客户。
CIOS already meters usage per block. DCM will let that spare capacity find buyers — spot jobs, plans, and reserved blocks on one market.CIOS 已经按模块计量用量。DCM 将让富余容量找到买家——现货任务、套餐与预留容量，统一撮合。
DCM · Roadmap · Token aggregationDCM · 路线图 · Token aggregationOur interactive TCO wizard walks through CapEx, power costs, utilization, and projected IRR — tailored to your site parameters. 交互式 TCO 向导覆盖 CapEx、电价、利用率与预期 IRR — 按站点参数定制。