Timing
- ~12 mo
- GPU generation
- 18–36 mo
- Conventional build
The chip turns over before the room turns on.
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.
Three things you commit to before the answers exist.
The chip turns over before the room turns on.
The fastest power isn't downtown.
Assumptions expire before the asset does.
And what a site allows inside the box is a local call, not a technical one.
Modularity makes every decision smaller — and later.
Two cooling lines, six standard blocks. Pick by platform, then by boundary.
Telemetry, alarms→tickets, usage metering. Included with every block.
Sell spare GPU capacity — spot, plans, and reserved blocks.
How we choose these blocks — Why MDCX → · Model economics — TCO Analysis → · Specs — Docs · Standards · USD
We publish the two segments we control. We don't guess the third.
Expansion blocks: 90 days ex-works.
Every MDCX block is a standard configuration. No bespoke engineering cycle, no one-off bill of materials, no re-approval. The unit that ships is the unit we tested.
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 enclosure boundary.
Power, infrastructure, and compute — run by one operating system, sold as one revenue-ready service.
Consulting · Design · Deployment · Care+ · Model optimization · Brokerage — the services layer is where per-project configuration happens. How we work →
Differentiated redundancy — 2N where failure hurts, N+1 where it doesn't. Availability where it matters, without blanket-2N cost.
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.
Headquartered in Sunnyvale, California, with our APAC hub in Singapore, we work with global partners to serve clients in over 40 countries.
Meter, compose, and monetize compute rights. The metering ships today; the market is being built.
CIOS records power and utilisation per block, per tank, per rack — the same model the container was built from. Included with every unit, no separate deployment.
Metered capacity becomes something you can package: spot jobs, subscription plans, or blocks reserved ahead of time. One site, several ways to sell the same hours.
Spare capacity finds buyers on one market instead of sitting idle between your own workloads — which is what turns a utilisation gap into revenue.
DCM · Roadmap · Token aggregation How DCM will work →
Our interactive TCO wizard walks through CapEx, power costs, utilization, and projected IRR — tailored to your site parameters.
