THE PRE-ENGINEERED SECONDARY NETWORK
The Coming OEM Battleground from CDU to Cold Plate
Abstract
The secondary fluid network that carries thermally controlled coolant between the coolant distribution unit and the rack-mounted cold plates of AI factory data centers is undergoing a structural migration. Today, between thirty and fifty percent of this scope is fabricated on site by general-contractor mechanical trades. Within twenty-four to forty-eight months that share will collapse to between five and fifteen percent. The remaining work will be performed inside Tier-1 OEM manufacturing facilities, shipped as factory-integrated assemblies, and connected on site only at a small number of pre-engineered transfer interfaces. This paper names the migration, quantifies the redirected capital, identifies the OEM battleground that will be contested through 2028, and prescribes the manufacturing footprint, telemetry stack, leak-detection doctrine, procurement specification language, and commissioning workflow that will determine which Tier-1 OEMs capture the consequent market share. It draws on public M&A activity exceeding fifteen billion dollars of disclosed value during 2023 through 2026, public market sizing from Dell’Oro Group, Omdia, and Data Center Frontier, OCP and ASHRAE technical specifications including the Universal Quick Disconnect and the 2021 fifth-edition thermal guidelines, the NFPA 75 and NFPA 76 fire protection framework, the NVIDIA channel certification pathway, and direct practitioner experience drawn from hyperscale, federal, and colocation infrastructure deployments. The conclusion is direct: the secondary fluid network is no longer a piping problem. It is a manufacturing, integration, governance, and capital allocation problem, and the Tier-1 OEMs that recognize this in 2026 will own the AI factory thermal envelope for the remainder of the decade.
Executive Summary
The CDU-to-cold-plate fluid path inside an AI factory is the most contested unclaimed scope in the data center industry. It has historically been treated as a mechanical-trade subset of the general contractor’s responsibility, specified loosely on plumbing drawings and delivered through a combination of welded stainless steel piping, field-mounted manifolds, and field-installed isolation valves. That model collapses under the conditions of the AI factory. A single uncontrolled fluid event in the white space costs an operator the entire row, multiplying ten-million-dollar GPU racks across an unrecoverable failure surface (Orr, 2025). The hyperscaler community has responded predictably: it is pulling the scope inward, away from the field-assembled trade work, and toward factory-built integrated product. The OEMs that own that product own the AI factory thermal envelope.
This paper documents that migration in five connected arguments. First, the trade scope is moving from thirty to fifty percent prefab today to eighty to ninety percent prefab by 2030, with the most aggressive transition occurring between 2026 and 2028 (DellOro, 2025; DellOro, 2025). Second, the migration is concentrated in what we call Zone 2, the row-level secondary fluid network between the CDU and the rack manifold, because Zone 1 (rack-internal) is already largely factory product and Zone 3 (plant-primary) moves more slowly due to facility-grade engineering inertia. Third, the OEM consolidation pattern is decisive: Vertiv acquired CoolTera in December 2023 (Vertiv, 2023; Data, 2023), Schneider Electric acquired controlling interest in Motivair in October 2024 (Schneider, 2024; Facilities, 2024), nVent expanded into manufactured power-and-cooling integration through the Avail acquisition in 2025 (nVent, 2025), Eaton closed its $9.5 billion acquisition of Boyd Thermal in March 2026 (Eaton, 2026; Bloomberg, 2025), and Ecolab moved on CoolIT Systems in early 2026 (Ecolab, 2026). Fourth, the competitive winners will be defined not by manufacturing alone but by the integration of CDU, manifold, leak detection, telemetry, and cross-vendor control planes into a single AVL-listed product. Fifth, capital allocators should reframe OEM valuation around demonstrated Zone 2 share and US manufacturing density rather than legacy cooling-product revenue.
The paper develops these arguments across seventeen chapters supported by forty-two technical figures, twenty-two analytical tables, and ninety-four cited sources. It then provides a detailed governance prescription for OEMs, operators, and investors. The recommendations are summarized in three statements:
OEMs must build, not merely acquire, vertically integrated Zone 2 manufacturing footprint with native leak detection, isolation valving, and three-variable telemetry. Acquisitions accelerate but do not substitute for the operational discipline of running an integration line that ships listed assemblies at hyperscaler cadence.
Operators must rewrite secondary-loop specifications now. Legacy spec language that delegates Zone 2 scope to the general contractor produces field-assembled fluid networks that fail the AI factory reliability standard. Specifications must mandate factory-assembled scope above defined complexity thresholds and refuse field-welded fluid networks in the white space.
Capital allocators should price Tier-1 OEMs on Zone 2 capability, US manufacturing density, and lifecycle service quality, not on legacy CDU revenue. The valuation paradigm of the AI factory decade is geographic-separation manufacturing capacity translated into integration capability, not unit cooling shipments.
The paper is intended for boardroom, regulatory, investor, and senior engineering review. It is written in the voice and convention of an industry reference document. Practitioner-experience material drawn from direct hyperscaler, federal, and colocation operating practice is framed under the Practitioner-Experience Integration Protocol where it appears.
Full white paper below
