IT Infrastructure for data centers in India

Where the facility meets the workload

White-space infrastructure is procured on a different cycle from the building, often by a different team, and increasingly for densities the building was not designed for. The result is predictable: 42U racks with 16 A single-phase PDUs in a hall that will host 20 kW cabinets within a year, cabling installed by the last contractor on site without a labelling standard, and GPU servers ordered after the allocation window closed.

India's market combines the global OEMs for compute, storage and networking, sold through authorised partners, with strong domestic manufacturing of racks, PDUs, containment and structured cabling. Lead times are short for domestic items and allocation-driven for accelerators, which have become the most constrained item in the industry.

The white-space standard

Write one standard and apply it across halls. Racks: 48U to 52U, 600 or 800 mm wide, 1200 mm deep, with cable management and the option of liquid-cooling manifolds. Power: two feeds per rack from separate distribution paths, 32 A or 63 A three-phase rack PDUs with per-outlet metering and switching, and sufficient amperage headroom for the hall's design density. Containment: hot or cold aisle, with blanking panels mandated. Cabling: pre-terminated fiber trunks (OM4 or OS2), copper where required, a labelling scheme, and documented patching records from day one.

The standard costs little more than the alternative and removes the rip-and-replace that follows a density increase. It also makes operations faster: every technician knows where power, patching and labels are in every row.

Compute, storage and networking

Enterprise servers and storage from the major OEMs run 6 to 12 weeks through authorised Indian partners, configuration dependent. GPU systems run 3 to 9 months and depend on written allocation; an order without a confirmed allocation is a hope, not a delivery date. Network equipment for a modern hall is a spine-leaf fabric with 100G to 400G uplinks, and the long-lead item is often the optics rather than the switches.

Procure hardware price and support terms as separate decisions. Negotiating the hardware hard and leaving the support contract at list price costs more over three years than it saves on day one. Specify 4-hour parts replacement in your city, firmware and security update entitlements, and the OEM's authorised channel so warranty is valid.

Installation and acceptance

Rack and stack, cabling and labelling are a service to be specified and inspected, not an afterthought. Define the installation standard (cable routing, bend radius, labelling, torque on PDU connections, airflow blanking), require power-on testing and a photographic record per rack, and accept by serial number against the bill of materials. Pre-terminated trunks tested at the factory cut installation time per hall from weeks to days and give a test record before the cable is plugged in.

Lifecycle and the secondary market

Plan the refresh cycle at purchase. Enterprise servers refresh at four to five years, network equipment at five to seven, and the residual value of a well-documented, serial-tracked estate is materially higher than one that leaves the building as scrap. Data sanitisation to NIST 800-88 with per-serial certificates is the condition for any resale, and it is cheaper to build into the asset register from the start.

How we procure white space

We write or review the white-space standard, build the bill of materials by hall and phase with delivery dates, run the RFQ across authorised partners and manufacturers with hardware and support priced separately, confirm allocation on accelerators in writing, and specify and inspect installation. Clients contract directly with the chosen partners; we manage the comparison and the acceptance.