In 2026 the average U.S. data center chiller lead time sits at roughly 35 weeks, with water-cooled units spanning 12–60 weeks and air-cooled units clustered near the low end of that band [S3].
Dry coolers occupy a parallel but shorter path: across the U.S. In the U.S. data center coolers market, dry coolers maintain an 18.11% share within the cooling towers and dry coolers category (2025) [S1]. Procurement engineers weighing a chiller order against a dry cooler order are therefore choosing between two different scheduling regimes, not two flavors of the same equipment.
What the 2026 chiller lead-time curve actually looks like
The 2026 Global Data Center Outlook places the mean U.S. chiller lead time at approximately 35 weeks; the 12–60 week spread is reported specifically for water-cooled chillers, reflecting centrifugal and screw-compressor build slots at Carrier AquaEdge, Trane Series-E/CGAM, and Johnson Controls YK-Screw platforms [S3]. Air-cooled units, including Carrier AquaForce 30XF and Trane RTAC-class machines, generally quote faster because they eliminate the condenser-water loop and ship as a single package, but the published mean is still a long-lead mechanical item in 2026 builds [S3].
The schedule penalty compounds: submittals, factory witness tests, shipping, rigging, flushing, and controls integration all sit between purchase order and plant turnover. A 35-week chiller quote therefore translates into a 45–55 week mechanical milestone once the surrounding scope is added, which is why 2026 gigawatt-class AI site schedules in published hyperscaler reference designs treat the chiller as a critical-path item rather than a buy-on-demand commodity [S3]. For a comparable 10 MW thermal load, the refrigeration-ton equivalent is approximately 2,843 tons before N+1 redundancy and design margin, using 1 refrigeration ton = 3.517 kW [S3].
Dry cooler lead times: why the curve is shorter and flatter
Dry coolers are fin-and-coil heat exchangers with fans, controls, and a structural frame; the bill of materials contains copper, aluminum, galvanized steel, and EC fans, all of which draw from shorter and more parallel supply chains than large compressor assemblies. Delta T Systems and Munters both publish dry-cooler product lines aimed at data center heat-rejection, with Munters emphasizing integration with air, liquid, and hybrid cooling architectures [S2][S5].
In 2026 the practical dry-cooler lead-time band sits at roughly 8–16 weeks for standard skid sizes, with the wider end driven by non-standard fin materials (copper-finned or epoxy-coated coils for corrosive ambient air), custom fan counts, and large single-coil units above ~2 MW heat-rejection per module [S4][S5]. The shorter curve is the structural reason dry coolers are increasingly used to back-stop long chiller deliveries: a project can land a dry-cooler array first, run on dry cooling or hybrid mode while the chiller slot matures, and then hand off to full mechanical cooling once the water-cooled chiller is commissioned [S2].
Where dry coolers fit, and where they do not
Dry coolers only reject heat to ambient air, so their capacity tracks the dry-bulb temperature and the design approach to ambient. In most U.S. and northern European climates, dry coolers cover roughly 60–85% of annual hours in a hybrid chiller/dry-cooler configuration; the residual high-humidity and high-ambient envelope is handed to a chiller or adiabatic pre-cooling stage [S2][S4]. For water-constrained sites, the trade is a larger outdoor footprint and higher fan power for lower water consumption, which is why hyperscale AI builds in arid regions lean toward dry cooling while temperate and tropical regions stay on water-cooled chillers with cooling towers [S4].
The boundary cases matter. A dry cooler alone cannot meet a stable leaving-water temperature low enough for direct liquid cooling of high-density AI racks (typically 35–45°C LWT to the rack manifold) during summer peak; a chiller or a chilled-water buffer is required for that last leg [S3]. Conversely, on a 5–10 MW edge or colocation build, a packaged air-cooled chiller often wins on schedule and footprint because the chiller and the heat-rejection device ship as one factory-tested package, eliminating the second vendor from the critical path [S3].
Decision matrix: chiller vs dry cooler vs hybrid for 2026 data center builds
For procurement scoping, the three options compare against four criteria: lead time, water use, peak-ambient capability, and per-MW installed cost. The values below are drawn from the 2026 reference material and should be treated as order-of-magnitude for budgeting [S1][S3][S4][S5].
Water-cooled chiller + cooling tower: 12–60 week lead time, high water use, full peak-ambient capability, mid-to-high per-MW cost once the condenser loop, pumps, and tower are added. Air-cooled chiller: typically lower end of the chiller lead-time band, zero cooling-tower water, full peak-ambient capability, larger roof or yard footprint, mid per-MW cost. Dry cooler (standalone or hybrid): 8–16 week lead time, near-zero water use, derated at high ambient, low per-MW cost for the heat-rejection stage but does not provide active refrigeration [S1][S3][S4][S5].
The hybrid chiller/dry-cooler architecture is the structural answer most 2026 reference designs converge on: the dry cooler carries the base load for 60–85% of hours, the chiller tops up during summer peak, and both pieces of equipment can be ordered on different lead-time tracks to compress the overall schedule [S2]. This is also the architecture that shows up in published AI-facility reference designs from Johnson Controls and in the Munters dry-cooler product positioning [S5].
Failure modes and constraints that decide the order
Three failure modes repeatedly show up in 2026 data center cooling procurements. First, the chiller arrives but the condenser-water side is not ready: pumps, tower, dry cooler, and piping on a 35–60 week tail, which the chiller manufacturer will not own. Second, the dry cooler arrives on time but the glycol mix, flow rates, or ambient-derating assumptions were not locked at PO, forcing a re-spec and pushing the commissioning window. Third, hybrid controls are treated as a chiller add-on rather than a plant-level scope item, so the two systems fight each other in shoulder seasons [S3][S4].
The constraint layer is regulatory and standards-driven: refrigerant phase-down rules (AIM Act schedule in the U.S., F-Gas in the EU) push 2026 orders toward low-GWP refrigerants such as R-1234ze and R-134a variants, and several large centrifugal platforms are still in re-certification for new low-GWP options, which is a non-trivial contributor to the chiller-side lead-time tail [S3]. Ambient-derating curves and noise limits are specified per site and should be locked at the chiller and dry cooler PO, not at commissioning, because both affect fan and compressor selection in ways that are expensive to retrofit [S4].
Sourcing signals worth tracking through the rest of 2026
Three signals are worth watching in the remaining 2026 quarters. The first is whether the chiller lead-time band tightens from the current 12–60 week spread toward the lower end as 2026 factory slots at Carrier, Trane, and Johnson Controls clear; published quarterly capacity guidance from those three OEMs is the cleanest leading indicator [S3]. The second is dry-cooler pricing under copper and EC-fan supply pressure: published Munters and Delta T Systems quoting behavior for Q4 2026 will show whether the 8–16 week band holds or slips [S2][S5]. The third is hyperscaler reference designs for 2027 builds, which already point to a higher share of hybrid chiller/dry-cooler plants for AI campuses in water-constrained metros [S1][S4].
Component reference pages worth checking: data logger.
See also our earlier report, LPDDR4 supply for industrial and embedded boards: what is actually available in October.