The Engineering Core

800 VDC Chiller Explained: Where Trane's 'Up to 2%' Efficiency Gain Comes From


Hook: Most 800 VDC headlines have been about racks and power gear. On Sept 30, 2026, the cooling plant joined in: Trane Technologies, working with Eaton and Danfoss, modified an existing high-efficiency chiller to run from an 800 V DC feed and delivered more than 1,000 tons (3.5 MW) of cooling in a lab demonstration, with "the potential for up to 2% improvement in system efficiency" over AC (Trane press release). Two percent sounds small. The interesting question is where it physically comes from, and why the biggest chillers may not follow.

Key takeaway: A variable-speed chiller already runs its compressor from an internal DC link. Feeding 800 VDC lets it skip the drive's AC front end (rectifier plus harmonic hardware). The rectification does not vanish, though. It moves upstream to whatever builds the 800 V bus. So "up to 2%" is a boundary-dependent number, and Trane has not published that boundary.

What happened

  • Lab proof-of-concept, not a product: an existing high-efficiency chiller, modified for an 800 V DC input. Trane calls it the industry's first 800 V DC cooling architecture. No model, ship date or price was given.
  • Capacity: over 1,000 tons of refrigeration, which is about 3.5 MW thermal (1 ton = 3.517 kW).
  • Claimed upside: up to 2% higher system efficiency. In a typical 200 MW data center, Trane says that could free up to 1.8 MW for compute, enough for fifteen 120 kW racks or three 600 kW racks.
  • Why the cooling plant matters: in the joint Trane–Eaton Reference Design #505, a 100 MW compute block "can impose 20 MW of motor load," and that design still "anchors on 480 V AC today" while DC protection and components mature (Reference Design #505, PDF).

Engineering mechanism

1) The chiller already has a DC bus, so 800 VDC skips its front end

A variable-speed compressor drive is three blocks: AC input → rectifier (diode bridge or active front end) → DC-link capacitors → inverter → motor. Trane's release says AC cooling "relies on multiple stages of power conversion to operate compressor drives, pumps and fans," with losses at each step. Feed the DC link directly and the input rectifier, its line reactors and any harmonic-mitigation hardware drop out of the chiller. Your EV already works this way: the electric A/C compressor runs from the high-voltage battery's DC bus through its own inverter (see our breakdown of how much range EV air conditioning really costs). The catch for buildings: someone still has to turn grid AC into 800 V DC, whether an upstream rectifier, a sidecar or a solid-state substation. Whether that loss sits inside or outside the "2%" boundary decides how real the gain is.

2) 800 V on a 480 V-class DC link is a redesign, not a plug swap

Standard drive arithmetic: a six-pulse bridge on 480 VAC gives an average DC link of about 1.35 × 480 ≈ 648 V. An 800 V feed is about 23% higher, so DC-link capacitors, switching-device voltage margin, the precharge circuit (no rectifier means you must soft-charge a big capacitor bank from a stiff DC source) and DC-rated disconnects all need checking. The upside is on the motor side. With space-vector PWM, the maximum line-to-line RMS output is about Vdc/√2, so roughly 458 V from a 648 V link versus about 566 V from 800 V. That is extra voltage headroom for the compressor motor at high speed. Which partner handled what (Danfoss drive hardware? Eaton DC protection and distribution?) was not disclosed. That is item #1 on the to-verify list below.

3) The current formula: 800 VDC does not cut amps much, it cuts conductors and stages

Reference Design #505 spells out the sizing formula: I = P / (√3 × V × PF) for three-phase AC and I = P / V for DC. At 1 MW with PF 0.95 that gives 1,267 A per phase at 480 VAC versus 1,250 A per pole at 800 VDC, nearly the same current. The gain is wire count: two conductors instead of three, which Eaton puts at "up to 34%" copper cross-section and weight savings. A quick check with equal conductor size gives 2 × 1,250² / (3 × 1,267²) ≈ 0.65, about 35% less feeder I²R loss (illustrative only: it ignores neutral, grounding conductor and skin effect). Power factor is part of why AC carries more current for the same real power, which we unpack in Efficiency vs. Power Factor: The Real Reason AI Data Centers Need SSTs.

4) Why the biggest chillers stay on medium-voltage AC for now

In Trane's 1 GW reference design, each ~100 MW cooling block uses 10 CenTraVac chillers of about 3,000 tons. Higher-capacity units require medium voltage, and unloading is handled by staging and inlet guide vanes with reduced-voltage starters instead of MV VFDs. For a 1,500 HP pump, 4.16 kV cuts line current from 1,420 A to 164 A. No low-voltage DC bus competes with that on copper. The 800 V demo chiller is roughly a third of that size and drive-based. My read: 800 VDC cooling enters first where a drive already exists and the load sits near the DC whitespace, meaning CDUs, pumps, fans and mid-size variable-speed chillers. The 3,000-ton MV machines out in the yard come later, if at all.

To verify:

  • Partner roles: Trane did not say which Danfoss drive or Eaton hardware was used, or who built the 800 V source.
  • "Up to 2%" boundary: it is not stated whether this is chiller-only or plant-level, includes upstream AC→DC conversion, or was measured at full or part load.
  • 1.8 MW per 200 MW: that is 0.9% of the facility. If the 2% applied only to cooling power, the base would be ~90 MW, well above the "20 MW motor load per 100 MW block" in RD #505 (~40 MW for 200 MW gives ~0.8 MW). The basis is unpublished, so treat 1.8 MW as an upper-bound estimate.
  • Timeline: no commercial chiller model, certification path or ship date was announced.

Practical takeaways

This is a real step: a multi-megawatt mechanical load running on the same voltage class as next-generation racks. But it is a lab result, and the number to argue about is the boundary, not the 2%. If you are specifying an 800 VDC hall or reviewing a vendor deck, ask these five questions:

  1. Where is the meter? Is the efficiency gain counted grid-to-shaft, or only from the DC terminals inward?
  2. What is the DC-link and device rating at 800 V plus transients? What precharge method is used?
  3. How are DC faults cleared? DC has no current zero, so ask about DC-rated breakers or SSCBs and the grounding scheme.
  4. Who owns harmonics now? Removing the chiller's AC front end shifts IEEE 519 compliance onto the shared rectifier or SST.
  5. Ride-through: RD #505 already puts CDUs on dedicated UPS because liquid cooling is now mission-critical. Can a shared DC bus with storage do that more simply?

Related on this beat: The Engineering Core archive (800 VDC protection, SSTs, SiC/GaN power stages) and @TheEngineeringCore-v. Traditional Chinese deep dives: vkinngworld.

Sources: