Data Center is about building the AI factory around the rack. Liquid cooling, higher facility water temperatures, 800 VDC and medium-voltage power, and prefab to cut build time. From the hyperscaler, colo and design-build engineers getting 1MW racks into halls, fast.
Who it's for
VP, Data Center EngineeringDirector, Critical FacilitiesHead of Data Center DesignMEP Engineering LeadMechanical and Cooling EngineerConstruction Program Director
Data Center track topics at Digital Infrastructure World 2027
The Data Center track at Digital Infrastructure World 2027, held in collaboration with the Open Compute Project, home of OCP Tech Day. San Jose, February 3-4, 2027. Sessions are being confirmed; these are the topics on the agenda.
01 · Modular & Prefab
Modular and prefab: building at the speed of silicon
GPU generations now arrive every year; data centers still take years to build. Google's answer is late binding: commit as late as possible to what goes in the hall.
Sessions cover factory-built power, cooling and network modules, sized to the silicon vendor's reference cluster (Crusoe described units from 0.5 MW to about 3 MW at the 2026 OCP Global Summit), central plants that serve many modules, build-to-suit when speed is the spec, commissioning from CDU screening to integrated systems testing, and late-binding supply chains that keep a design open to the hardware that ships rather than the hardware that was forecast.
Topics include
Prefabricated modular data centers
Factory-built data center modules
Build-to-suit data centers
Commissioning and IST
Late-binding supply chains
02 · Cooling & Thermal
Cooling and thermal: the CDU becomes infrastructure
Liquid is now the default for AI racks, and the coolant distribution unit has become plant rather than equipment. Google's Project Deschutes in-row CDU, contributed to OCP, is now in its fifth generation.
The block covers multi-megawatt CDUs, warmer facility water (ASHRAE W45 and 45°C supply that lets operators design out chillers), operating temperatures across the loop, and what sustainable cooling looks like at a gigawatt.
At 54V, a 1 MW rack would need up to 200 kg of copper busbar, by NVIDIA's estimate. Hence the move to high-voltage DC: ±400 VDC through the Mt Diablo work by Google, Meta and Microsoft at OCP, and 800 VDC on NVIDIA's roadmap from 2027.
Sessions follow the power chain from the grid to the rack: power sidecars, native low-voltage DC distribution, solid-state transformers fed from medium-voltage AC, HVDC busway, protection and safety, and the migration path from today's AC halls. Google said at the 2026 OCP Global Summit that its spec went from first draft to publication in about six months, against roughly two years for a traditional standard.
Enabling 1 MW racks: floors, aisles and ultra-wide racks
A 1 MW rack is a facilities problem before it is an IT problem: weight, aisle width, floor loading, pipe routing and service clearance. Google expects racks above 500 kW before 2030.
This block takes the 1 MW rack being designed at OCP Tech Day and asks what the hall around it must look like: hyperscale-ready shells, double-wide formats such as Open Rack Wide (ORW), floor loading for liquid-filled racks, and the rack-to-facility interfaces hyperscalers and third-party developers now design together.
Topics include
1 MW racks
Open Rack Wide (ORW)
Floor loading for AI racks
Halls for ultra-wide racks
Rack-to-facility interfaces
05 · Brownfield to AI
Brownfield to AI: liquid in air-cooled halls
Most of the world's data center floor was built for air; most new AI racks need liquid. Meta pairs liquid-cooled racks with liquid-to-air units so they can run in halls built for about 20 kW per rack, and Google's Project Brazos brings liquid cooling to air-cooled brownfield sites.
Sessions cover the retrofit path for colocation, enterprise and edge sites: liquid-to-air sidecars and rear-door heat exchangers, adding facility water to an air-cooled building, power and floor upgrades, and how far an existing hall can be pushed before a new build wins.
Topics include
Brownfield data center retrofit
Liquid-to-air CDU
Rear-door heat exchangers
Project Brazos
Enterprise and edge AI
06 · IT/OT Convergence
IT/OT convergence: one telemetry stack, chip to grid
In an AI factory, the GPU, the CDU and the substation share a fate. Unified IT/OT telemetry is on OCP's Open Data Center for AI agenda for that reason.
The block covers open operations for AI data centers: telemetry from chip to grid, building management and fire systems tied into the IT stack, availability measured in nines rather than tiers, predictive maintenance, and running liquid loops at fleet scale. Google reports about 99.999% availability for its Deschutes CDU fleet since 2020.
Scott WillisPresident and CEODartPoints Data Centers
Brian DeckardTechnical Regulation and Standardization ExpertSiemens