Industry data center

By Ian Fordyce, Head of R&D, Conflux Technology

If you’ve evaluated an additively manufactured cold plate, you’ve probably sat in this meeting. The thermal data looks good. Then someone from supplier quality asks whether the vendor can make tens of thousands of them, each one the same, at a price that survives a cost review. Can additive manufacturing scale that far, when a build takes days and every part needs hand finishing?

Ten years ago the honest answer was no. We’ve spent much of the decade since on the reasons why, first on motorsport and aerospace heat exchangers and, since 2024, on automotive production programmes. Most of those reasons turned out to be engineering problems, and engineering problems can be fixed.

Why additive manufacturing build time isn’t fixed

The objection usually runs like this: a laser powder bed fusion machine takes days to build a plate of parts, so volume would need an unrealistic number of machines. The arithmetic is fine. The input is wrong.

Build time isn’t a property of the printer. It’s set mostly by the part and how it’s built: how tall it is, because every layer costs time; how many parts share a build plate; how much support material the machine lays down only for someone to cut it away later; and how well the laser parameters suit the geometry. A part designed for machining and then printed pays for all of it.

We learned this in Conflux’s very early days. Our first water-to-charge-air coolers were brick-shaped and slow to build. Moving them to new laser parameters and design-for-additive principles took the build from something like 10 days to around a day, and improved build quality along the way. Four changes since then have done most of the work, on any heat exchanger we make.

Four key changes that moved additive manufacturing to volume production

1.    Nest more parts on every build plate

Each build carries fixed time for loading, inerting and warm-up, plus the full height of the tallest part. Spreading that over more parts is the cheapest gain available. On our automotive barrel intercoolers, we sized the core so that five parts fit on each build plate. Six would have forced a core too long for the customer’s package, so we stopped at five. We also kept the parts low, since build height sets the number of layers, and shaped the ports to support themselves instead of needing sacrificial supports.

Five additively manufactured barrel intercoolers on one laser powder bed fusion build plate during powder removal.

Five barrel intercoolers share one build plate.

2.    Use design for additive manufacturing to remove machining

Our earlier heat exchangers needed machining on the air and coolant interfaces and on the mounting points. Every machined feature adds a fixture, a set-up and a second process with its own lead time. On the same programme we designed the hose connections and mounts to be used as printed, so the part needs no machining at all.

3.    Automate post-processing

In early additive work, most of the labour sat after the print: powder removal, cleaning, coating, marking and inspection, mostly by hand. We moved powder removal and cleaning to automated or semi-automated processes, and inspection to semi-automated. For the steps that stay manual, fixtures matter more than you’d expect. A simple reusable jig cut one coating preparation step from about 15 min to under 1 min per part.

4.    Prove repeatability across machines

Volume needs more than one machine, so a qualification team will ask whether parts from different machines behave the same. We flow-tested parts from different positions on the same build plate and from two separate machines. The variation wasn’t large enough to affect performance in any meaningful way, so in production a part can come off either machine. Each part is serialised and traceable to its build, then pressure-tested and inspected under our AS9100D and ISO 9001:2015 quality systems.

Six measures that show whether an additive process can scale

Our results come from our history of manufacturing automotive heat exchangers, but the measures behind them apply to any additive programme, because they describe the process rather than one good part.

  1. Parts per build plate. Every build carries a fixed overhead, so the more parts that share a plate, the less of that time each part has to carry. Ask a supplier how many parts fit on one plate; on our barrel intercooler, we fit five.
  2. Build time per plate. Part height and laser parameters decide how long each build takes, and both improve as a design matures. Ask how a supplier’s build times have changed over the life of a part; on our first coolers, they fell from roughly 10 days to about a day.
  3. Supports to remove. Every support the machine prints is material someone has to cut away and finish by hand after the build. Ask how much of the part needs supports; our ports are designed to support themselves, so they need none.
  4. Machining after the print. Each machined feature adds a second process, with its own fixtures, set-up and lead time. Ask how many features need machining after printing; our barrel intercooler needs none, including the hose connections.
  5. Hands-on time per part. Post-processing is where most of the labour, and so most of the cost, tends to sit. Ask for the hands-on minutes per part after printing; our programme target is 30 minutes or less.
  6. Consistency across machines. Capacity only scales if a second machine produces the same part as the first. Ask for flow or thermal data comparing parts across machines and plate positions; ours showed no meaningful difference across two machines.

Over the decade from 2015 to 2025, improving these measures delivered about three times the heat exchanger performance at about a third of the price. These same methods sit behind the production heat exchangers we supply for the Pagani Utopia and the Donkervoort P24 RS.

Where additive cold plates still cost more, and what doesn’t scale yet

I’d rather you hear the limits from me than find them in an audit.

A simple plate on a moderate-power chip doesn’t need additive. If a skived or machined copper plate meets your thermal resistance and pressure drop budget, it will usually cost less per part, and you should buy it. Additive earns its place where internal geometry does the work: high heat flux, pronounced hot spots, a tight pressure drop budget or an awkward envelope.

Each build still takes hours. Throughput comes from running machines in parallel, so capacity is added in steps, and each step carries capital and lead time. Ask any supplier, including us, for dated capacity figures, and check which are installed and which are projected.

Repeatability results apply to the parts and machines that were tested. A new cold plate design still needs its own first-article and consistency checks. A supplier’s track record should make those checks faster, not optional.

Finally, “no joints to leak” is an advantage over brazed and skived assemblies, not over a plate machined from a single block. Compare like with like.

What should you ask an additive cold plate supplier?

Ask for production evidence, not prototype results: the six measures above, with numbers from parts in production. They tell you more about scale than any capacity claim.

Checklist of six questions to ask an additive cold plate supplier: parts per build plate, build time per plate, supports to remove, machining after the print, hands-on time per part, and consistency across machines.

Six questions to ask any additive cold plate supplier.

For a fuller supplier audit, add these:

  • Is every part serialised and traceable to its build, and which quality system covers it?
  • How many powder sources and machine service arrangements back the supply?
  • What capacity is in place today, what is projected, and by when?

These are the questions we’re building the Conflux Production System to answer, starting with cold plates for data centre programmes. If you’re qualifying an additive supplier, send us your audit list and we’ll answer it with our production data.

Send us your supplier qualification questions → Contact Us