What 'Server Liquid Cooling Wastewater' Actually Means in 2026
A server liquid cooling wastewater treatment supplier in 2026 is typically an industrial water-treatment EPC that delivers a multi-stage train — DAF or lamella clarifier for suspended solids, MBR or multimedia filter for organics, RO plus EDI for deionized-loop polishing, and glycol or fluorocarbon recovery where applicable. Modern systems target 40–90% water reuse, COD discharge below 500 mg/L, and conductivity under 5 µS/cm, with CAPEX of roughly $0.8M–$6M for a 5–50 MW data center site.
The first mistake buyers make is treating this stream the same as legacy cooling-tower blowdown. It is not. Server liquid cooling produces four distinct wastewater streams, each with its own chemistry, and a qualified supplier must address all of them in the process train.
Single-phase cold-plate loops circulate a water–glycol mixture — typically PG25 (25% propylene glycol) or PG50 — through cold plates attached to CPUs, GPUs, and DIMMs. Makeup water is deionized to keep loop conductivity between 50 and 500 µS/cm, but corrosion byproducts (iron 0.1–2 mg/L, copper 0.05–0.5 mg/L) accumulate as the loop ages, and biofilm forms readily at the 18–25 °C operating window. Side-stream filtration and blowdown carry this contaminated glycol mixture to drain or to recovery.
Two-phase immersion cooling submerges servers in a dielectric fluorinated fluid — Novec 7000/7100 (3M fluorinated ketone) or 3M Fluorinert — that boils at 34–61 °C on hot components and condenses on the tank lid. Water enters the bath through top-cover humidification, pump-seal leakage, and condensation during maintenance. A gravity decanter separates the two liquids, but the aqueous phase still carries trace fluorocarbons and must be treated before discharge.
Scale matters. Cooling water accounts for 30–50% of total data center water consumption in 2026 industry benchmarks, and liquid cooling increases per-rack coolant inventory by 5–20× compared with air cooling. A 10 MW site at PUE 1.25 can easily generate 30–80 m³/day of side-stream blowdown. The ASHRAE TC 9.9 2024 update and the Uptime Institute 2025 liquid-cooling survey both confirm that single-phase cold plates and two-phase immersion have crossed the early-adopter threshold and now anchor hyperscale procurement specifications.
A supplier must be qualified to handle four streams simultaneously: glycol-bearing side-stream blowdown, fluorocarbon/water separator effluent, DI loop polish-reject, and humidifier bleed. The full 2026 data center liquid cooling wastewater treatment specs walk through the chemistry in detail.
Process Train Options: Which Treatment Stages Do You Actually Need
Three process trains cover roughly 95% of data center liquid-cooling installations in 2026; choosing the wrong one is the most expensive mistake a buyer can make. The decision is driven by influent chemistry, water cost, discharge route, and whether two-phase immersion is deployed.
Train A — Discharge to municipal sewer: Rotary bar screen → DAF system for coolant-loop TSS and oil removal → MBR system for glycol-bearing wastewater → carbon polish. Targets are COD below 500 mg/L, BOD below 300 mg/L, and glycol below 50 mg/L to align with TRGS 611 sewer limits. This is the standard for hyperscale campuses with unconstrained sewer capacity and water costs below $2/m³.
Train B — High-reuse closed loop: Multimedia filter → industrial RO system for water reuse and DI-loop makeup → EDI → UV. Targets are 75–90% water recovery and permeate conductivity below 0.1 µS/cm, suitable for water-scarce regions (UAE, Singapore, Spain) where reuse mandates exist or makeup water exceeds $5/m³.
Train C — Two-phase immersion: Decanter → activated carbon → aqueous MBR. Recovers above 99% of the fluorinated fluid for reuse, with the water phase treated separately. Only relevant when two-phase immersion is deployed alongside or instead of single-phase cold plates.
| Parameter | DAF | MBR | RO | Vacuum Distillation |
|---|---|---|---|---|
| TSS removal | 85–95% | 99%+ | 99%+ | n/a |
| COD removal | 40–60% | 92–97% | 95–99% | 98%+ |
| Conductivity reduction | minimal | minimal | 95–99% | 99%+ |
| Glycol concentration (product) | n/a | n/a | concentrate 2–5% | 95%+ concentrate |
| Footprint per m³/day | 0.2–0.4 m²/m³/h | 0.5–1.0 m²/m³/day | 1–3 m² per 50 m³/day permeate | 0.3–0.6 m²/m³/day |
Decision rule: Choose Train A if sewer capacity is unconstrained and water cost is below $2/m³. Choose Train B if water cost exceeds $5/m³, the site sits in a water-stressed basin, or a corporate reuse mandate is in force. Choose Train C only when two-phase immersion is part of the architecture. RO plus vacuum distillation in a hybrid configuration can recover glycol to 95%+ concentration for resale or reuse, materially offsetting OPEX.
Sizing the System: Flow Rate, Capacity, and Footprint per MW
Translating IT load into treatment capacity is straightforward once the loop architecture is fixed. The figures below let a buyer produce a defensible budgetary envelope before approaching suppliers.
Single-phase cold plate: Cooling water demand runs 1–2 L per kWh of IT load. For a 10 MW site at PUE 1.25, makeup water is 280–560 m³/day and side-stream blowdown to treatment is typically 30–80 m³/day, depending on cycles of concentration and bleed-off frequency.
Two-phase immersion: Aqueous flow is markedly lower (5–15% of single-phase volumes), but fluorocarbon inventory is substantial at 1,000–3,000 L per MW. Treat the aqueous phase at 20–50 m³/day for a 10 MW site, and size the decanter and carbon polish stage for the full fluorocarbon volume.
Footprint rules of thumb for the major unit operations: MBR requires 0.5–1.0 m² per m³/day of capacity, DAF needs 0.2–0.4 m² per m³/h, and a skid-mounted RO occupies 1–3 m² per 50 m³/day of permeate. Containerized ISO-frame systems are available for 5–20 m³/day modular deployment at edge sites, with the full train shipped in one or two 40 ft units.
Above 20 m³/day, design for 1 working + 1 standby train. A single-train failure on a closed DI loop will push conductivity past 5 µS/cm within hours and risk corrosion-product spikes that are expensive to clean. Redundancy is cheaper than chemistry excursions.
| IT Load | Loop Type | Makeup Water | Treatment Capacity | Recommended Footprint |
|---|---|---|---|---|
| 5 MW | Single-phase | 140–280 m³/day | 15–40 m³/day | 40–80 m² (skid-mounted) |
| 10 MW | Single-phase | 280–560 m³/day | 30–80 m³/day | 60–120 m² |
| 20 MW | Single-phase | 560–1,120 m³/day | 60–160 m³/day | 120–240 m² (with standby) |
| 50 MW | Single-phase | 1,400–2,800 m³/day | 150–400 m³/day | 300–600 m² (multi-train) |
| 10 MW | Two-phase immersion | 10–30 m³/day | 20–50 m³/day (aqueous) | 40–80 m² + decanter bay |
The MBR stage is the throughput bottleneck; sizing the PVDF flat-sheet MBR membrane module for peak daily flow plus 20% turndown margin prevents fouling during low-load night periods.
Vendor Scorecard: How to Compare Server Liquid Cooling Wastewater Suppliers
Three to five shortlisted EPCs is the right number; more than that wastes evaluation cycles. The scorecard below weights six auditable criteria so the procurement manager can defend the selection internally without subjective judgment.
Criterion 1 — Reference projects. Require a minimum of 2 installed systems at above 1 MW liquid-cooled data center sites in the last 36 months. Request a live reference call, not just a project list with logos. Liquid cooling is a young market; suppliers without recent data center work will be learning on your budget.
Criterion 2 — Process engineering depth. The supplier must hold in-house capability for P&ID development, mass-balance modeling, and CFD for the decanter and DAF hydraulics. Outsourced process design is a red flag because coolant-loop chemistry is unforgiving — the influent varies hourly, and the process envelope must be tuned, not copy-pasted from a municipal plant.
Criterion 3 — Equipment scope. Prefer suppliers who manufacture DAF, MBR, RO, chemical dosing (see automatic chemical dosing for pH and coagulant control), and sludge dewatering filter press for coolant-loop residuals in-house. Single-warranty accountability removes the finger-pointing that dominates commissioning when DAF comes from one vendor and MBR from another.
Criterion 4 — Compliance documentation. Documented compliance with the local discharge framework — UAE Federal Decree-Law 12/2026, EU Urban Waste Water Directive 91/271/EEC, US EPA categorical standards — plus glycol-handling certifications (REACH, TRGS 611 alignment). A supplier who treats compliance as a checkbox will cost the operator a permit delay.
Criterion 5 — Service SLA. 48-hour onsite response, 24/7 remote monitoring via SCADA or a digital dashboard for wastewater KPI monitoring, and confirmed spare-parts inventory in the region. Loop chemistry does not respect business hours.
Criterion 6 — Commercial terms. Fixed-price EPC is preferable to cost-plus for budget certainty. Demand a performance guarantee on effluent quality with liquidated damages, and a minimum 24-month warranty covering membranes, instruments, and rotating equipment.
| Criterion | Weight | Pass/Fail Threshold | Evidence to Request |
|---|---|---|---|
| Reference projects (>1 MW, 36 mo) | 20% | ≥2 sites | Reference call + commissioning report |
| Process engineering depth | 20% | In-house P&ID, mass balance, CFD | Sample mass balance for 10 MW case |
| Equipment scope (single warranty) | 15% | DAF + MBR + RO + dosing + dewatering | Manufacturing audit or factory tour |
| Compliance documentation | 15% | Local sewer + glycol certs | Compliance matrix by jurisdiction |
| Service SLA (48h, 24/7, spares) | 15% | All three confirmed in writing | SLA exhibit in contract |
| Commercial (fixed-price, LDs, warranty) | 15% | Fixed EPC, ≥24-mo warranty | Draft EPC with LD clause |
CAPEX and OPEX in 2026: What Should the Quote Look Like
Budget conversations go off the rails when the buyer and supplier are not anchored to the same cost bands. The 2026 figures below are the envelope a credible quote should sit inside; anything 30% below likely means scope is missing, and anything 50% above means the supplier is hedging against an undefined influent.
CAPEX bands 2026: Train A (discharge to sewer) runs $0.8M–$2.5M for 5–20 MW sites and $2.5M–$6M for 20–50 MW sites when RO is added for high-reuse operation. Train C with fluorocarbon recovery is $1.5M–$4M for 5–20 MW.
OPEX bands 2026: Train A sits at $0.12–$0.45 per m³ treated, Train B at $0.25–$0.60 per m³ (RO membrane replacement dominates), and Train C at $0.20–$0.50 per m³ (carbon replacement dominates). These figures exclude electricity, which runs 0.4–0.8 kWh per m³ treated for the full train — meaningful inside a PUE 1.1–1.3 envelope.
Hidden costs catch buyers out: PG25 concentrate make-up at $2,500–$4,000 per m³, fluorinated-fluid top-up at $80–$150 per liter, RO membrane replacement every 3–5 years at $80–$150 per m², and sludge dewatering/hauling at $80–$200 per wet ton. The 2026 wastewater treatment plant OPEX breakdown itemizes these line by line.
ROI trigger: water reuse pays back within 2–4 years when makeup water cost exceeds $4/m³ and the site sits in a water-stressed region per the WRI Aqueduct 2026 dataset. Below $2/m³, the reuse case is weaker and the operator should default to Train A unless a corporate ESG target overrides the financial math.
| Cost Driver | Unit | 2026 Range | Notes |
|---|---|---|---|
| Train A CAPEX (5–20 MW) | USD total | $0.8M–$2.5M | Skid-mounted, no RO |
| Train B CAPEX (20–50 MW) | USD total | $2.5M–$6M | RO + EDI included |
| Train C CAPEX (5–20 MW) | USD total | $1.5M–$4M | Fluorocarbon recovery |
| PG25 make-up | per m³ | $2,500–$4,000 | Recovers via RO + distillation |
| Fluorinated top-up | per liter | $80–$150 | Novec/Fluorinert replacement |
| RO membrane replacement | per m² | $80–$150 | 3–5 year cycle |
| Energy | kWh per m³ | 0.4–0.8 | Include in PUE model |
Commissioning Checklist: 90 Days From PO to Treated Loop
The 90-day window below is the realistic envelope for a 5–20 MW single-phase site; larger Trains B and C stretch to 120–150 days because of RO membrane lead time.
Days 0–30: Site survey, influent characterization (3-week composite sampling for glycol, metals, conductivity, BOD/COD), final P&ID freeze, and long-lead procurement. RO membranes and MBR modules carry 8–12 week lead times from order to delivery; place those POs in week 1.
Days 30–60: Factory acceptance test (FAT) of skids at the supplier's facility, shipping, civil foundation work on site, and loop tie-in preparation. FAT must include a 72-hour continuous run at design flow with simulated influent — anything less surfaces problems during PAT, not FAT.
Days 60–90: Installation, commissioning, and the performance acceptance test (PAT) over 7 consecutive days at design flow. Define PAT criteria up front: effluent COD, TSS, conductivity, glycol concentration, and water-recovery percentage. All values must be measured and signed off before final payment triggers.
Operator training and SCADA handover close the project; a digital twin for wastewater treatment plant handover gives the operations team a 90-day shadow-run before going fully unattended.
Frequently Asked Questions
How much does a server liquid cooling wastewater treatment system cost in 2026?
CAPEX runs $0.8M–$2.5M for 5–20 MW Train A (sewer discharge), $2.5M–$6M for 20–50 MW Train B (RO + EDI for reuse), and $1.5M–$4M for 5–20 MW Train C with fluorocarbon recovery. OPEX sits at $0.12–$0.60 per m³ treated depending on train selection, excluding energy at 0.4–0.8 kWh/m³.
Can the same supplier handle both single-phase and two-phase cooling wastewater?
Yes, but only if the supplier holds in-house capability across DAF, MBR, RO, and decanter/carbon systems. Train A (single-phase) and Train C (two-phase) share the MBR and sludge-handling stages; the decanter and carbon polish are added for Train C. Confirm both scopes in the EPC contract before signing.
What water-recovery rate is realistic for a data center?
40–90% depending on train and influent. Train A (sewer discharge) typically achieves 40–60% recovery through side-stream filtration only. Train B with RO + EDI reaches 75–90% recovery and is the right choice when water cost exceeds $5/m³ or a reuse mandate applies.
Is glycol-treated wastewater hazardous?
Propylene glycol itself is relatively benign, but the stream carries corrosion metals (iron, copper), biofilm, and inhibitors. Discharge is regulated: TRGS 611 in Germany, REACH in the EU, and local sewer ordinances typically cap glycol at 50 mg/L. Documented compliance and chemistry-specific treatment are non-negotiable.
How long does delivery and commissioning take?
90 days from PO to treated loop for a 5–20 MW Train A site, assuming long-lead items (RO membranes, MBR modules) are ordered in week 1. Larger Trains B and C run 120–150 days. The full 2026 data center liquid cooling wastewater treatment specs include a week-by-week commissioning Gantt.
