Why Liquid Cooling Has Made Stainless Steel Pipe a Hyperscale Decision, Not a Commodity One

Five years ago, the data-center pipe room was a quiet corner of the project. Copper Type-L ran the secondary loop, schedule-40 steel with threaded fittings ran the plant water side, and the procurement cycle was measured in weeks. That world is over. A 2026 hyperscale AI training hall is dense, hot, and continuous: 50–100 kW per rack, 365 × 24 operation, with a coolant loop whose failure mode is not a small leak but a flooded GPU room and a multi-million-dollar compute outage. The stainless steel pipe in that loop has become the load-bearing component of the entire AI compute stack.

The shift has three drivers. First, power density: an air-cooled 15 kW rack needs no liquid plumbing; a 70 kW rack does, and the cooling plant at that point becomes more important than the UPS. Second, water quality: most DLC loops use deionized water or a water-glycol mix, and any metal-ion leaching or biofilm growth will shorten the life of the cold plate and the pump seals. Third, regulatory: China's GB 50174-2017 and the new T/CIE 097-2021 "design specification for indirect liquid cooling data center with liquid / air dual channel heat pipe cold plate" both explicitly point to 304 / 316L precision stainless steel pipe as the default first-choice material for the secondary loop.

This guide walks through the four engineering decisions that decide whether a DLC pipe run is right — material grade, standard and dimensional family, surface finish, and jointing method — and how a mill-direct source consolidates the tube, the pipe fittings, the pipe flanges, the gasket stud bolt nut kits and the industrial valves at the CDU into a single quality-controlled package.

Step 1: 304 or 316L — The Material Decision That Drives the Whole BOM

For a DLC secondary loop, only two austenitic grades are usually in play: 304 (UNS S30400 / EN 1.4301) and 316L (UNS S31603 / EN 1.4404). Both are 18Cr-8Ni base austenitics, both are non-magnetic, both weld easily — but 316L adds 2–3 % molybdenum and drops the carbon to 0.03 % max, which together lift chloride pitting resistance and eliminate intergranular corrosion at the weld HAZ. Whether the extra cost of 316L is justified depends on three questions.

• Coolant chemistry. A pure deionized-water loop with resistivity > 1 MΩ·cm and a closed inhibitor package is well within 304's envelope. A water-glycol mix with any chloride contamination, a coastal site pulling sea-air through the intake, or a single-phase immersion dielectric fluid with trace halogens pushes the project into 316L territory.
• Operating temperature. 304 is rated for intermittent contact up to 870 °C, but its sustained-use limit in a chloride-bearing environment is closer to 60–70 °C. 316L raises the practical ceiling to about 80–90 °C before pitting becomes a concern.
• Cleaning and passivation regime. If the loop will see an annual chemical flush (e.g. citric acid for iron removal, or a peroxide flush for biofilm control), 316L tolerates the aggressive clean far better than 304. 304 will survive routine inhibitor flushes, but multiple aggressive cleans will eventually show tea-staining and pitting.

Step 2: Pick the Right Standard Family — ASTM A312, A249, A270, EN 10312, JIS G3459, GB/T 12771

Once the grade is fixed, the next decision is the standard. DLC loops are built to several overlapping families, and the choice of standard is partly a regional preference, partly a wall-thickness argument, and partly a finish-and-tolerance argument. The four families that cover the majority of 2026 DLC secondary loops are summarized below.

A practical note: for the secondary rack loop, the dominant choice in 2026 is welded A249 (or its EN 10312 / GB/T 12771 equivalent) in the 1/2″ to 2″ range, with the primary outdoor loop in welded A312 in the 3″ to 12″ range. A project that mixes both should ｄｅｃｌａｒｅ both standards on the purchase order, with the mill responsible for cross-referencing dimensions and test methods so the field fab has no ambiguity at the transition fitting.

Step 3: Surface Finish Codes — 2B, BA, EP and the Ra Values That Matter

The surface finish of the tube's inner wall is the single most underestimated variable in a DLC loop. The roughness controls the biofilm growth rate, the pressure ｄｒｏｐ, the ease of chemical cleaning, and the amount of particulate that can shed into the cold plate. The three finishes a data-center buyer should know are:

• 2B (mill finish). Cold-rolled, annealed, pickled and lightly skin-passed. Typical Ra 0.3–0.6 μm. Adequate for primary outdoor loops and for secondary loops where the coolant is well-controlled and chemically inhibited. The default for cost-sensitive inland builds.
• BA (bright annealed). Solution-annealed in a hydrogen or vacuum furnace, no post-anneal pickle. Typical Ra 0.2–0.4 μm. A noticeably smoother, brighter surface. The standard finish for most high-quality rack loops and CDU headers.
• EP (electropolished). After BA, an electrolytic process strips 10–30 μm of surface material, leaving a chrome-oxide-rich, micro-smooth surface. Typical Ra ≤ 0.25 μm, with the best production runs at Ra ≤ 0.18 μm. Required for high-purity DCLC loops, for any loop feeding direct-to-chip cold plates with sub-100 μm micro-channels, and for any single-phase immersion fluid circuit where particulate shedding must be near zero.

Step 4: Jointing — Why Orbital Welding is the Default for the DLC Loop

The joint is the weak point of any stainless pipe run, and the DLC loop is unforgiving. A standard screwed or compression joint on a 304 secondary loop at 6 bar and 30 °C is fine for a process-water system. The same joint on a DLC loop with 18 °C supply / 35 °C return and a 50,000-hour mean-time-between-failure target is a liability. The four jointing families a DLC project should consider, in increasing order of capital cost but also of joint quality and reproducibility, are:

• Tri-Clover sanitary clamp (DIN 32676 / ISO 2852). A EPDM or FKM gasket between two ferrules, clamped by a wing-nut or bolt clamp. Used in EP-grade DCLC, food-grade, and pharma-grade loops. Slow to install, expensive per joint, but easy to break for cleaning. Reserved for EP loops.
• Compression fittings (Swagelok-type). Dual-ferrule bite-type mechanical joint. Common in 1/2″ and 3/4″ instrument air and sample lines. Not preferred for full-size DLC rack loops because every joint is a potential leak point under thermal cycling.
• Press-fit (Viega ProPress Stainless, Raccorderie Inoxpres). Mechanical press with a profiled elastomeric O-ring. Faster than orbital welding, but the O-ring material must be carefully matched to the coolant (EPDM is usually safe for water / glycol; FKM for higher temperature or aromatic fluids). Press tools are operator-dependent and need a calibrated jaw set.
• Orbital TIG welding (autogenously). A water-cooled welding head orbits the tube and produces a repeatable, full-penetration, gas-purged weld. The default for permanent rack-loop joints. Requires a WPS (Welding Procedure Specification) qualified to ASME IX or EN ISO 15614, and a PQR (Procedure Qualification Record) on the same tube grade and size.

For a hyperscale build, the realistic split is roughly 90 % orbital-welded joints and 10 % Tri-Clover joints (at equipment inlets, sampling ports, drain valves). Press-fit and compression appear in instrument and sample lines, not in the main rack loop. The pipe fittings used at the welds are long-pattern 90° elbows, equal tees, and concentric reducers to ASME B16.9, in the matching 304 / 316L grade. Pre-fabricated spool pieces (the rack-loop assemblies that arrive on site ready to lift into place) are orbital-welded at the mill and arrive with WPS documentation, weld maps, and a 100 % radiographic or dye-penetrant test report.

Step 5: The Bundled Package — Tube, Fittings, Flanges, Gaskets, Valves

A DLC rack loop is not just tube. The full mechanical BOM at the CDU and at the rack manifold includes:

• Stainless tube spool pieces — orbital-welded, pre-fabricated, dimensionally checked against the BIM model. The bulk of the tonnage.
• Pipe fittings — long-pattern 90° / 45° elbows, equal tees, concentric and eccentric reducers, instrument tees, all to ASME B16.9, in matching 304 or 316L.
• Pipe flanges — ASME B16.5 Class 150 weld-neck flanges at the equipment connections (CDU, plate-frame HX, expansion tank). Slip-on flanges for the air-side headers. Steel flanges are not used inside the DLC loop; the entire wetted path is stainless.
• Gasket stud bolt nut kits — spiral-wound graphite + stainless or PTFE-filled gaskets for the flanged joints, matched to ASME B16.20. Stud bolts in B8M / B8M Cl.2 (316 grade) per ASTM A193, with 2H heavy hex nuts per ASTM A194.
• Industrial valves — stainless-body ball valves for isolation, globe valves for sampling, check valves at the pump discharge, and pressure-safety relief valves at the expansion tank. Sizes are typically DN15 to DN50 for the secondary rack loop, larger at the CDU primary side.

The reason a mill-direct source matters here is that all five families need to be qualified to the same standard of documentation: MTC per EN 10204 3.1 (mill certificate) for the bulk items, MTC per EN 10204 3.2 (independent third-party witnessed) for the high-pressure primary side, dimensional reports, hydrostatic test reports, surface-finish certificates with measured Ra, and weld maps with weld-procedure references. A project that sources each family from a different trader ends up with five different MTC formats, three different traceability systems, and a QA engineer who spends the first three months of the project reconciling paper.

Step 6: Pre-Commissioning — Pickling, Passivation and the First DI-Water Flush

A 304 or 316L tube that has been cut, welded and stored on site has a surface iron contamination and a heat-tint layer at every weld. If the loop is filled with DI water in that state, the iron contamination and the heat-tint layer are the first places biofilm will grow. The standard clean-and-passivate sequence before first fill is:

1. Mechanical clean. Blow or pig the line with foam pigs to remove weld spatter, swarf and shop dirt. Especially important for rack-loop spools that have been on site for weeks.
2. Degrease. Alkaline degreaser (typically NaOH 5–10 % at 60–70 °C) circulated for 2–4 hours, then flushed with DI water to neutral pH.
3. Pickling. A nitric / hydrofluoric acid mix (e.g. 10 % HNO₃ + 1 % HF at 50 °C for 30–60 min) removes the heat tint from the weld HAZ and exposes fresh chromium-rich surface. Rinse thoroughly to neutral pH. For EP systems, the pickling is less aggressive to preserve the EP layer.
4. Passivation. 20–30 % nitric acid (or citric acid for a more environmentally acceptable alternative) at 50–60 °C for 30–60 min grows the protective chromium-oxide layer back to full thickness. Rinse to neutral pH and resistivity > 1 MΩ·cm.
5. DI-water flush and fill. The final flush should achieve resistivity at the design value, with the inhibitor pre-mixed into the fill water. The fill water itself should be pre-filtered to 5 µm absolute and ideally 1 µm for EP systems.

A common procurement pitfall is buying the right tube and right fittings, then leaving the pre-commissioning chemical clean to a sub-contractor who has no experience with stainless passivation. The clean is at least as important as the tube grade, and it should be specified and witnessed as part of the project QA plan, not left to field improvisation.

Five Procurement Pitfalls That Bite Hyperscale DLC Projects

• Specifying 304 for a coastal or immersion site to save 8–12 % on tube cost. The cost of replacing a 1,000-rack loop is 50–100 times the saving. 316L is the cheaper option over the life of the asset.
• Buying to A249 without confirming the weld is autogenously TIG, not laser. Both are valid under A249, but the HAZ microstructure is different and the post-weld pickling requirement is different. A mill that sells "A249 welded" without specifying the process is a mill to query.
• Specifying Ra without a test method and a sample location. Ra is meaningless unless the test direction, cut-off, and instrument are locked down. A 0.4 µm contact reading can be 0.7 µm on an optical profilometer over the same weld HAZ.
• Leaving the pre-commissioning passivation to a sub-contractor. Passivation is a metallurgical process, not a generic chemical clean. The contractor should provide a documented procedure (per ASTM A380 or A967), a witness plan, and a final passivation verification report (e.g. copper-sulfate test or resistivity measurement).
• Sourcing the tube, fittings, flanges, gaskets and valves from five different vendors. The MTC paperwork does not match, the heat numbers do not cross-reference, and the QA engineer spends three months reconciling it. A mill-direct package with one MTC format, one traceability system and one set of test reports is faster, cheaper and lower risk.

FAQ — Stainless Steel Pipe for Data Center Liquid Cooling

For an inland site with a closed DI-water loop, well-controlled inhibitor chemistry, and no chloride exposure, 304 is a perfectly defensible choice. The premium for 316L is real (roughly 8–12 % on tube and 15–20 % on fittings), and on a 5,000-tonne rack loop the saving from going 304 is material. Use 316L where the chloride risk is non-zero: coastal sites, immersion-cooling loops, and any loop with a planned chemical-clean regime. Most 2026 hyperscale builds default to 316L for the rack loop and 304 for the secondary outdoor loop.

A312 is the general-purpose standard for austenitic stainless pipe in sizes up to 30″. A249 is specifically for welded austenitic tube up to 5″. For a 1/2″ to 2″ rack loop, A249 is the more common call; A312 is more typical for the 3″ to 12″ CDU headers. Both standards share most of their chemistry and mechanical requirements, so a mill that holds both can cross-supply without re-testing.

BA is adequate for the majority of secondary loops. EP (electropolished, Ra ≤ 0.25 µm) is needed where the cold plate has sub-100 µm micro-channels that would clog with biofilm, where the loop feeds a single-phase immersion dielectric that must remain particulate-free, and where the operator wants to extend chemical-clean intervals. The cost premium for EP is meaningful (20–40 % on tube), and it should be targeted at the sections that need it, not blanket-applied to the whole run.

Press-fit is faster than orbital welding by a factor of three to five, and the quality of the joint is excellent when the tool is calibrated and the operator is trained. The trade-off is that every press joint is a service item: the O-ring has a finite life and a known temperature rating. For a hyperscale rack loop with a 15–20 year design life and minimal maintenance windows, the consensus in 2026 is still orbital welding for the permanent joints and Tri-Clover at the equipment inlets.

EN 10204 3.1 for the bulk tube, fittings, flanges, gaskets and valves. EN 10204 3.2 (independent third-party witnessed) for the high-pressure primary side and for any sour or chloride service. A documented weld map for the spool pieces, a hydrostatic test report, a surface-finish certificate with measured Ra, a chemical composition report, and a mechanical-properties report (tensile, hardness, grain size for austenitic). Anything less and the QA engineer will be reverse-engineering documentation for the first three months of the project.