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.
Every MDCX data center ships with its operating system. Hardware blocks deliver capacity; CIOS turns that capacity into operated, metered, revenue-ready infrastructure.
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:
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.
Licensing is open-core: Apache-2.0 core with commercial surfaces. One line for the site; details off-site.
Plugin drivers ingest field protocols into a gateway stream and time-series store. The asset model speaks modular data center language, not generic racks.
Modbus TCP and SNMP v2c shipped. Path-addressed points map into the site asset tree.
Pods, CDUs, immersion tanks, and loops as first-class objects — not a flat tag dump.
3D scene mapping of your pods is a capability of the lab pipeline — no Omniverse claim.
Operations workflows turn plant state into work that can be tracked, SLAd, and closed — without free-for-all remote control.
Alarm engine opens tickets with lifecycle, SLA scan, and notify (webhooks / email).
Schedules, inspections, spares, and maintenance windows that suppress alarms when work is planned.
CLI cios query / cios set with risk classes — not open write access.
CIOS meters what the plant consumes and produces so commercial teams can act. It does not invoice.
Records, rollups, and export for kWh and GPU-hour style usage — for reconciliation and customer portals.
See where the site still has room before the next block lands.
Ops portal for operators; customer portal for status, SLA, and usage. No billing or ERP claims.
Edge-first: a single box can run the site plane offline for ≥7 days. Multi-site aggregation is explicit Roadmap.
Honest strip — not a promise calendar.
DCGM and per-GPU metrics beyond facility path addressing.
NOC-style aggregation across sites.
Predictive maintenance and capacity forecast assistants.
Hardware without an operating system is just metal. Pick a block; CIOS ships with it.
Model the economics in TCO, or explore the hardware blocks.