The 83-Field Spec Gap in Data-Center Power and Cooling

A 50 MW build specs UPS topology, battery chemistry, generator fuel, refrigerant, and liquid-loop fluid. A standard RFQ asks for five fields. The gap is where wrong-spec risk and a 2027 refrigerant phase-out collide across the Thailand-SE-Asia corridor.

Data-center construction in Thailand has crossed from pipeline into procurement. Thailand’s Board of Investment recorded roughly ฿41.6 billion of newly promoted data-center investment in its latest reported year across about 166 projects, a forward pipeline running 1.8× the 92 data-center factories the Department of Industrial Works lists as operational. The capital is approved. What buyers procure next is not a building; it is a power-and-cooling system, and that system is where most RFQs fall silent.

When MPBxChange Research mapped a 760-row power-and-cooling materials database against the live RFQ surface, the gap was stark. The reference data exposes 83 distinct field categories across 12 product families, UPS, PDU and busway, generators, batteries, CRAC, CRAH and fan walls, in-row and RDHx, chillers, cooling towers, liquid CDUs, immersion, and electrical distribution. The typical data-center RFQ captures five fields: transformer kVA, voltage class, rack density, PUE target, and site country. A hyperscaler buying a 50 MW IT load is specifying topology, chemistry, fuel, refrigerant, and loop fluid; the contract that asks for five fields is silent on the seventy-eight that decide cost, runtime, and compliance.

The system is twelve families, not one line item

Each family carries its own wrong-spec failure mode. The materials database breaks down by depth: 96 battery rows, 91 generator rows, 63 CRAC rows, 58 in-row and RDHx rows, 55 CRAH and fan-wall rows, 50 chiller rows, 40 liquid-CDU rows, 39 UPS rows, 38 immersion rows, and 34 PDU-and-busway rows. A UPS line that omits topology, online double-conversion vs. line-interactive, and ECO-mode efficiency is silent on both runtime and PUE. A battery line that omits chemistry hides a three-to-fifteen-year lifetime delta between VRLA and lithium. A generator line that omits prime-vs-standby kW and fuel type hides both a permit exposure and a continuous-rating mismatch.

The anchor suppliers buyers actually meet

Picking an anchor in each family opens a compatible product line for the rest of the build, and the reference data names them by product-count weight. In continuous power, Schneider Electric, Vertiv, and Eaton anchor UPS, PDU, and busway; generators run on Atlas Copco, Cummins, and FG Wilson (CAT); battery backup concentrates in C&D Technologies, Leoch, and EnerSys. On the thermal side, STULZ and Liebert (Vertiv) anchor CRAC, with Airedale and STULZ strong in CRAH and fan walls; in-row and RDHx skew to Schneider Electric, Huawei, and Liebert; chillers to Huawei, Carrier, and Trane; cooling towers to Evapco, SPX Cooling (Marley), and Baltimore Aircoil. Liquid cooling, the GPU-era growth family, anchors on Motivair (Schneider), CoolIT, and Vertiv for CDUs, and GRC, Submer, and LiquidCool for immersion.

“A hyperscaler buying a 50 MW IT load specs the topology, the UPS chemistry, the cooling rejection strategy, and the liquid-loop fluid, none of which a five-field RFQ asks for.”

The refrigerant phase-out is a 20-year cost trap

The single most expensive silent field is refrigerant. Across the reference databases, legacy refrigerants, R-410A and R-22, outnumber the modern compliant set of R-32, R-454B, and R-1234ze by 23 mentions to 12, with phase-out chemistry concentrated in CRAC units (57 mentions) and chillers (46). The EU F-gas regulation is phasing R-410A out by 2027. A buyer who prices a chiller today against a legacy refrigerant is signing into a 20-year asset whose working fluid is on a regulatory clock, a cost trap that surfaces as forced retrofit, not as a line on the original quote. The same phase-out logic runs through battery chemistry, where legacy NMC, NCA, and LCO mentions (25) are already being overtaken by LFP and solid-state (35), and through two-phase immersion dielectrics where fluid phase-out is an open warranty question.

This is why refrigerant, battery chemistry, and dielectric belong on the RFQ as enumerated fields backed by a phase-out warning, not as free text a buyer discovers after award. When an RFQ touches R-410A, the procurement layer should already be surfacing R-454B and R-32 alternatives from verified suppliers, the same way it would flag a legacy battery chemistry against an LFP path.

Why the corridor changes the sourcing map

The supplier base SE-Asian buyers know today skews to the US and EU: of 20 named DC power-and-cooling suppliers, only about 10% sit in the Thailand corridor, with the top countries being the US, Germany, and France. The interoperability layer tells the same story from a different angle, across 438 OCP-certified products from 129 providers (43 in cooling, 23 in power), the geographic split runs US 58, Taiwan 46, Japan 16. For a Thailand or Vietnam build, that means the power-and-cooling bill of materials is import-led, and corridor proximity (TH/VN/MY/SG) is something to engineer toward, not assume.

The compounding advantage is cross-vertical: the suppliers anchoring DC also anchor adjacent builds. Schneider Electric, Eaton, Siemens, STULZ, Rittal, and 3M all appear in both DC and solar; Delta Electronics, Eaton, Schneider, and Siemens straddle DC and automotive. A supplier already KYC-verified for one vertical does not have to re-prove identity to quote the next, which is precisely how a corridor buyer turns a single trusted relationship into a multi-family build sheet.

What it means for procurement

• Spec all twelve families explicitly, UPS topology and ECO efficiency, battery chemistry and runtime, generator prime kW and fuel, refrigerant grade, chiller part-load COP, liquid-loop fluid, busway amp rating, switchgear voltage class. A silent field is a wrong-spec or system-mismatch risk.
• Treat refrigerant as a dated asset: screen chillers and CRAC against the R-410A/R-22 phase-out and require R-32, R-454B, or R-1234ze where the build must outlive 2027.
• Pin certifications to enumerated standards, UL 891, IEC 61439, ASHRAE TC9.9 envelopes, Uptime Tier, AHRI, NFPA 110, ISO 8528, rather than accepting free-text claims.
• Anchor each family to a named supplier (Schneider/Vertiv/Eaton in power, STULZ/Liebert in cooling, Atlas Copco/Cummins in generators) so the rest of the build inherits a compatible, interoperable line.
• Assume an import-led BOM in the TH/VN/MY/SG corridor and lean on cross-vertical, already-verified suppliers to compress qualification time on the 1.8× BOI pipeline.