At the kickoff meeting for a liquid cooling project, the IT integrator says the cooling water has to be filtered to 25 microns. The mechanical contractor says the cooling water will be treated to standard HVAC practice. Both are talking about "cooling water," but they mean two different loops. That mismatch usually shows up later as gaps in the spec, water quality problems nobody owns, and finger-pointing during commissioning.
A liquid-cooled data center runs on two loops: the technology cooling system (TCS) and the facility water system (FWS). TCS vs. FWS isn't a choice between two options; most direct-to-chip deployments run both. Knowing where one ends, where the other begins, and who is responsible for each is the first step toward a liquid cooling spec that holds up.
Key takeaways
- The TCS is the sealed secondary loop that serves the cold plates; the FWS is the primary loop that carries heat to the outdoor plant.
- The CDU is the boundary: the two loops exchange heat in its heat exchanger, and their fluids never mix.
- The TCS has far stricter water quality and filtration requirements, often 25 µm or finer.
- TCS supply temperature is roughly FWS supply temperature plus the CDU approach temperature.
- Write the demarcation point and each side's responsibilities into the contract before equipment is ordered.
What Are TCS and FWS?
The technology cooling system (TCS) is the sealed secondary loop that carries coolant from the CDU through manifolds and cold plates to absorb chip heat. The facility water system (FWS) is the building-side primary loop that moves that heat from the CDU to chillers, cooling towers or dry coolers. The loops exchange heat in the CDU but never mix.
Both terms come from ASHRAE TC 9.9's liquid cooling guidelines, and the Open Compute Project (OCP) uses the same vocabulary across its cold plate and liquid cooling documents. OCP's integration guidance spells it out: heat moves from the IT loop to the facility loop through a heat exchanger inside the CDU, and no coolant flows between the TCS and the FWS.
One nuance: ASHRAE treats the liquid path inside the server as a separate datacom equipment cooling system (DECS), while OCP's cold plate requirements describe the TCS as running from the CDU to the rack, through the manifold and IT equipment, and back. In most cold plate projects, "TCS" covers everything on the clean side of the CDU, including the cold plates.
You'll see both loops under other names, so it helps to recognize them in a spec:
- TCS: also called the secondary loop, the IT loop or the TCS loop.
- FWS: also called the primary loop, the facility loop or, when it's served by chillers, the chilled water loop.
Our liquid cooling glossary has short definitions of each term. For what happens inside the CDU (heat exchanger, pumps and controls), see how a CDU works.
TCS vs. FWS at a Glance
| TCS (Technology Cooling System) | FWS (Facility Water System) | |
|---|---|---|
| Scope | CDU secondary side → manifolds → quick disconnects → cold plates → back to CDU | Heat rejection plant (chillers, cooling towers, dry coolers) → facility piping → CDU primary side |
| Also called | Secondary loop, IT loop | Primary loop, facility loop |
| Fluid | Treated deionized water or a coolant such as PG25, one fluid across the whole loop | Facility water or a water-glycol mix, treated for site water chemistry |
| What sets the temperature | Upper limit worked back from chip junction temperature and cold plate thermal resistance; lower limit set by dew point | Heat rejection plant and local climate, tied to the ASHRAE W class |
| Pressure | Sealed loop; every component is rated to the TCS design pressure | Designed like a building HVAC water system, with larger mains serving many CDUs |
| Filtration and water quality | Strictest in the system: fine filtration, with conductivity, pH and inhibitor levels checked on a regular schedule | Standard HVAC water treatment against scale and biological growth |
| Piping and wetted materials | Stainless steel, copper or nickel-plated copper cold plates, compatible polymer hoses; mixed metals tightly controlled | Often carbon steel and other standard HVAC piping materials |
| Usually owned by | IT vendor, liquid cooling integrator, or CDU supplier | Owner's facilities team and the MEP contractor |
| Commissioning | Pressure test, clean and flush, fill, purge air, balance flow | Similar to a conventional HVAC water system |
| If something goes wrong | Chip temperatures and server uptime are hit directly | Cooling capacity for a whole row or hall is at risk |
The widest gap is in water quality. The TCS sits closest to the chips and has the narrowest passages, and OCP's modular TCS guidance calls for filtration of 25 µm or finer in loops with microchannel heat exchangers. For specific water quality limits and how to choose a coolant, see our direct-to-chip coolant guide.
Pressure matters too. OCP's liquid-to-liquid CDU test methodology notes that pressure in the TCS and the equipment cooling loops it feeds can reach 100 psi (690 kPa). Every TCS component, including cold plates, manifolds and quick disconnects, needs to be designed and validated for the loop's design pressure.
Where the Boundary Sits: The CDU and the Demarcation Point
The physical boundary between TCS and FWS is the CDU. A common scope split draws the line at the CDU's primary-side isolation valves: facility piping up to those valves is FWS, everything from the CDU's secondary outlet onward is TCS, and the CDU itself serves both loops. Where exactly the line falls, for example who owns the strainer on the primary inlet, varies by project, which is why it needs to be written down.
Getting that boundary in writing matters more than drawing a clean system diagram, because on many projects the two loops are bought, built and maintained by different teams:
- FWS side: usually the owner's facilities team and the MEP contractor. That covers the heat rejection plant, facility piping, water treatment and everything up to the CDU's primary connections.
- TCS side: usually the IT vendor, liquid cooling integrator or CDU supplier. That covers the CDU secondary side, row and rack manifolds, hoses and quick disconnects, cold plates and the coolant itself.
When the contract doesn't define the boundary, the same arguments come up again and again: who maintains the strainer on the CDU primary side, who pays when poor facility water scales the heat exchanger, and who supplies TCS top-up fluid and which fluid it should be.
A simple fix is to have both sides swap a one-page interface sheet early in the project.
What the FWS side should tell the TCS side:
- Design and maximum supply water temperature (worst case over the year)
- Available flow, available differential pressure, and connection size and location
- Primary-side water quality and filtration
- Plant redundancy and planned maintenance windows
What the TCS side should tell the FWS side:
- Liquid heat load per CDU, plus the phased build-out plan
- Required FWS flow and return water temperature
- CDU approach temperature
- CDU location, footprint and piping connections
The interface has long been treated as a core design issue. ASHRAE's liquid cooling guidelines include specific guidance on interface requirements between the chilled water system and the TCS.
How Temperatures Relate Across the Two Loops
The two loops are linked in series. TCS supply temperature is roughly FWS supply temperature plus the CDU approach temperature (the temperature difference across the heat exchanger). So warmer facility water means a warmer TCS. Working the other way, the highest TCS temperature your chips can accept sets the warmest facility water you can use.
The TCS also has a floor: it can't run below the room dew point. A common practice is to have the CDU track dew point and hold TCS supply at least 2°C (3.6°F) above it to prevent condensation inside the servers.
To work back from chip junction temperature to facility water temperature, and to pick an ASHRAE W class, see our data center liquid cooling design guide. We won't repeat the math here.
Why the Two Loops Are Kept Separate
- Protecting the cold plates. Microchannels and fine fins inside cold plates are sensitive to particles, scale and biofilm. Facility water usually isn't that clean.
- Running different fluids. The TCS can use a purpose-built coolant while the FWS fluid is chosen for the plant and climate, and neither constrains the other. Lenovo's Neptune water-cooling standards make the same point: keeping facility liquid apart from technology liquid is what lets each loop use a different fluid.
- Containing failures. Each TCS holds far less coolant than the facility system, so a leak or contamination event stays within the racks one CDU serves instead of spreading through the building's water system. For more on leak risks and safeguards, see what happens if liquid cooling leaks.
Chip-to-Chiller Architecture: Four Ways to Connect TCS and FWS
Chip-to-chiller architecture is the full heat path from the chip, through the cold plate, TCS, CDU and FWS, to the heat rejection plant that releases it outdoors. Depending on what the site has, there are four main ways to connect the two loops.
1. Liquid-to-liquid (L2L) CDU: the standard approach. The CDU hands TCS heat to the FWS, usually through a plate heat exchanger. It fits sites with existing chilled water or room for a new facility loop, and it's the default wherever facility water is available. The CDU can sit in the rack or in the row; see in-rack vs. in-row CDU for the trade-offs.
2. Liquid-to-air (L2A) CDU: no FWS needed. The CDU rejects TCS heat to room air through an air-cooled heat exchanger, and the existing air conditioning removes it from there. This works for air-cooled halls with no facility water that want to bring in a few liquid-cooled racks. The catch is limited heat capacity and a heavier load on room cooling.
3. IT equipment connected directly to the FWS: no CDU. OCP's guidance on connecting to facility water notes that some liquid cooling solutions tie straight into the FWS while others use a CDU to create a separate TCS. Rear-door heat exchangers typically connect directly. With cold plates, though, it means facility water flows through the fine channels on the chip, and the TCS-level demands on water quality, filtration and material compatibility spread to the whole facility loop. The risk and maintenance burden are high, so it's generally not recommended for cold plate systems.
4. Warm-water cooling without chillers. As chips accept warmer coolant, the FWS can run at higher supply temperatures and reject heat through dry coolers, with chillers rarely needed in the right climate. NVIDIA has said its 45°C (113°F) liquid cooling architecture can run on dry coolers without chillers in favorable climates. The split between TCS and FWS stays the same; the FWS just operates in a higher temperature range, and facility-side water use can drop sharply. See how much water a liquid-cooled data center uses.
For an overview of components and costs, see our guide to data center liquid cooling.
What This Means When You Specify Hardware
From a manufacturer's point of view, the TCS/FWS split shapes the requirements every part has to meet.
TCS components, including cold plates, manifolds and quick disconnects, live in one dedicated coolant for their whole service life and sit closest to the chips. APALTEK builds server cold plates from vacuum-brazed copper, with nickel plating available. Cold plates and manifolds can ship with your specified quick disconnects pre-installed as a single assembly. Every product is helium leak tested before it leaves the factory, and assemblies with pre-installed quick disconnects are leak tested as a whole. See our helium leak testing guide for how that works.
The CDU sits right on the boundary, so its spec has to match both loops. APALTEK designs and builds its own CDUs in both liquid-to-liquid and liquid-to-air configurations. The range covers 5 kW and 150 kW in-rack units and 1.5 MW and 2.5 MW in-row units, with redundant pumps (typically N+1), leak detection and automatic isolation. Browse the full lineup in AIDC thermal management.
When you write the spec, list requirements by loop:
- TCS: coolant type and concentration, design pressure, filtration rating, wetted materials list, supply temperature limits, and flow and pressure drop per cold plate.
- FWS: supply temperature range, available flow and differential pressure, primary-side water quality, connection location and size.
- Boundary: CDU primary and secondary connection types, who owns which strainer or filter, and who handles pressure testing and commissioning.
For flushing, filling and purging before go-live, see our liquid cooling commissioning guide. For ongoing upkeep, see the liquid cooling maintenance checklist.
TCS vs. FWS FAQ
What do TCS and FWS stand for in liquid cooling?
In data center liquid cooling, TCS stands for technology cooling system and FWS stands for facility water system. Both terms come from ASHRAE TC 9.9 and are used in OCP specifications. The same acronyms mean unrelated things in other industries.
What is a TCS loop?
A TCS loop is the technology cooling system: the sealed loop that runs from the CDU secondary side through manifolds, quick disconnects and cold plates and back to the CDU. It cools CPUs and GPUs directly and has the strictest water quality and filtration requirements in a liquid cooling system.
Is the TCS the primary or secondary loop?
The TCS is the secondary loop. The FWS is the primary loop, which carries heat from the CDU to the outdoor heat rejection plant.
Can servers connect directly to facility water without a CDU?
It's possible, but generally not recommended for cold plate systems. Facility water rarely meets the cleanliness that fine cold plate channels need, so scale or particles hit chip cooling directly, and contamination can spread across the whole facility loop. Rear-door heat exchangers often connect directly to facility water; cold plate systems are usually isolated behind a CDU.
What is chip-to-chiller cooling architecture?
It's the complete heat path from the chip through the cold plate, TCS, CDU and FWS to the chillers, cooling towers or dry coolers that reject heat outdoors. Depending on the site, projects choose between liquid-to-liquid CDUs, liquid-to-air CDUs, direct connection to facility water, and warm-water designs without chillers.
Who is responsible for TCS water quality?
Usually whoever owns the TCS side: the IT vendor, liquid cooling integrator or CDU supplier, as defined in the contract. FWS water quality typically falls to the owner's facilities team. Settle this early in the contract, especially who maintains the CDU primary-side strainer and who supplies top-up fluid.
Start with the Boundary, Then Choose the Hardware
Whichever architecture you pick, start by writing down the boundary conditions between the two loops: FWS supply temperature, TCS flow and pressure, and CDU location and connections.
Send that one-page sheet to APALTEK's engineering team. We'll evaluate how CDUs and cold plates should be matched, including the choice between liquid-to-liquid and liquid-to-air CDUs, cold plate flow and pressure drop, and thermal simulation before prototyping. We reply within one business day.