Cost, throughput and scaling to production
📍 Where we are: Part V · Industrial Reality — Chapter 15 of 19. The previous chapters covered qualification, data and sustainability; this one asks the plain commercial question: can additive manufacturing make enough parts, cheaply enough, to matter?
Think of a bakery with one very precise oven. The oven can make a beautiful, complicated loaf that no factory line could shape. But it bakes one tray an hour, someone must hand-trim every loaf afterwards, and each loaf must be inspected. Buying a second oven helps only if the trimmers and inspectors keep up. Scaling additive manufacturing is the same: the printer is one station in a long chain, and the slowest, most manual station sets the pace and the price.
What this chapter covers
We look at what actually limits the output and the unit cost of metal additive manufacturing (AM), mostly laser powder bed fusion (L-PBF), and at where polymer AM has already reached true mass production. Evidence is labelled as demonstrated at lab scale, demonstrated in production, or contested. Market figures are quoted only with their source and date.
Problem
Conventional processes such as casting, stamping and machining win on cost per part once volume is high, because their tooling cost is spread over many parts. AM has no tooling, so it was expected to win at low volume and to scale gracefully. In practice the unit cost of an L-PBF part falls only slowly with volume. Cost studies have long shown the shape of this: Hopkinson and Dickens found layer manufacturing cheaper than injection moulding for some geometries up to production in the thousands [1], and Baumers and colleagues' 2016 model of two commercial metal systems put specific costs at £2.39 (electron beam melting) and £6.18 (direct metal laser sintering) per cm³ deposited, concluding that deposition rates were insufficient for high-volume use although economies of scale were achievable [2]. A model of a GE engine bracket, by contrast, found AM cheaper than forging at 2,000 to 12,000 brackets per year once redesign and lifetime fuel savings were counted [3]. A 2014 NIST review of AM costs already framed the question as one of cost drivers (machine, material, labour, build time) and noted that automation could change when distributed production pays off [4]. A decade on, the central problem is unchanged: the same slow, serial physics that gives AM its design freedom also caps how many parts one machine can make per year.
The aggregate numbers are not small, but they describe an industry still shaped by equipment sales and services rather than by high-volume part output. Wohlers Associates reported global AM revenue of 24.2 billion US dollars for 2025, up 10.9% year on year, with printing services growing 15.5% and system sales only 3.6% (press release, 17 February 2026) [5]. AMPOWER estimated the 2024 metal AM market (equipment, material and parts) at 3.4 billion euros, a decline of 2.9% from the previous year [6]. The two sources define markets differently, so the figures are not comparable with each other; the point is that growth is moderate and uneven.
State of the art
Cost models
Demonstrated at lab scale (models validated on case studies, not on whole factories). Modern cost models treat a build as a queue of activities. Karaş and Shokrani built an activity-based model of L-PBF with discrete-event simulation and found that post-processing significantly influences overall cost, and that part geometry and production rate shift the key drivers [7]. An older reference point is the cost-estimation study of laser-sintering production at low to medium volumes by Ruffo and colleagues [8]. Parenti and colleagues compared metal extrusion with single- and multi-laser L-PBF using a multi-criteria decision method, which shows that the "best" process depends on the production scenario, not on the machine alone [9].
More lasers
Demonstrated in production, with limits. The main commercial answer to slow build rates is more lasers per chamber. A review of L-PBF progress in Europe notes that laser counts per chamber stagnated between 2013 and 2020, with effort going into higher laser power instead, and that the 2020 SLM NXG XII 600 introduced twelve simultaneous lasers [10]. More lasers do not scale output linearly. Wang and colleagues note that stitching the scan paths of several lasers makes thermal history harder to control and demands precise calibration, and that recoating time between layers is unavoidable in Cartesian beds, which caps the gain from extra or faster lasers [11].
Lab-scale concept. The same paper analyses a rotary (toroidal-bed) multi-laser architecture. Its modelled per-laser rates are tens to hundreds of cubic centimetres per hour (for example 18.17 cm³/h per laser for an existing two-laser rotary system at 1000 mm/s), and the authors argue output could rise by up to four orders of magnitude if laser counts, optics and scanning improve, assumptions they flag explicitly [11]. This is a productivity analysis, not a demonstrated factory.
Post-processing
Contested. Everyone agrees post-processing (depowdering, support removal, heat treatment, machining, surface finishing, inspection) is a major cost; nobody agrees on how large. The activity-based model above found it significantly influences overall cost [7], but published shares depend on scope, process and part mix, and many steps are priced per plate rather than per part, so the per-part share depends on how many parts share a plate. I found no peer-reviewed, comparable share and give no percentage; do not average figures quoted in trade sources.
Where it is truly in series production
Demonstrated in production. GE's Auburn, Alabama plant built the 30,000th additively manufactured LEAP fuel nozzle tip in October 2018, with more than 40 printers; the tip replaced roughly 20 welded and brazed pieces with one and is about 25% lighter [12]. The enabling logic is design consolidation, not raw print speed. In polymer AM, Align Technology's annual report describes making clear aligners by pressure-forming sheets over stereolithography-printed molds, one series per patient, and names the Invisalign Palatal Expander as its first direct 3D-printed device (company filing, fiscal 2025) [13]. Align closed its acquisition of Cubicure in January 2024 to scale direct printing [14]. Direct-printed clear aligners at scale remain an announced goal, not a verified production fact in the sources reviewed.
The cost of a metal AM part accumulates along a chain; the modelled specific cost of deposited metal is shown for two systems, and the clearest series-production cases are GE's LEAP fuel nozzle tip and Align's aligner production.
Original diagram by the authors, created with AI assistance.
Why it is hard
- Melt-rate physics. Build time is set by how much metal one laser can melt per hour at an acceptable quality. Raising power or speed pushes toward keyholing, lack of fusion and spatter (see Melt pool physics), so speed trades directly against defects.
- Utilization and batch economics. Machine cost is spread over productive hours, so load factor matters; per-plate steps reward full plates and punish small series. This is an inference from the activity-based models above [7] and from per-plate pricing practice, not an independent measurement.
- Labour and manual steps. Depowdering, support removal and finishing are still largely hand-guided; automating them is difficult because each geometry differs, and low volumes do not repay the investment (an argument in industry commentary, not a measured result).
- Qualification cost per part number. Each new part and machine combination must be qualified (see Qualification and certification); that fixed cost favours few part numbers at high volume, which is the opposite of AM's customization strength.
- Financial fragility. Cost pressure is visible at company level: Desktop Metal, a binder-jetting pioneer, filed for Chapter 11 on 28 July 2025, according to its parent company, Nano Dimension [15]. One company's failure does not prove the technology cannot scale, but it shows that scaling capital is hard to repay.
Open questions
- Can multi-laser stitching be made transparent enough that a twelve-laser machine yields close to twelve times the output with unchanged properties? Evidence is mixed and mostly proprietary.
- Will rotary or other new architectures leave the analysis stage? The assumptions about laser cost and scanning speed are untested at scale [11].
- What is the true, comparable post-processing share across part families, and which steps are realistically automatable?
- Can cost models be validated against real, audited factory data rather than case studies? Most published models are not.
Industry impact
The evidence supports a narrow, specific statement: AM reaches serial production where design consolidation or mass customization outweighs per-part cost, as in the LEAP nozzle tip and aligner production [12][13]. Elsewhere, buyers should model cost per part for their own geometry and volume instead of trusting a headline build rate. Growth in services faster than machine sales [5] is consistent with, but does not prove, a shift toward outsourced production; the wider market-structure implications of such cost structures are analysed by Weller and colleagues [16].
Why it matters
Technical progress that ignores cost does not reach production. Build rate, post-processing and qualification together decide which parts are ever printed in quantity.
In the real world
GE's Auburn plant and Align's high-volume operation show what scale looks like; Desktop Metal's insolvency shows the risk of building equipment businesses ahead of the economics.
Key terms
- Build rate — the volume of material a machine solidifies per hour, usually in cubic centimetres per hour.
- Multi-laser system — a powder bed machine in which several lasers melt different regions of the same layer simultaneously.
- Stitching — joining the regions scanned by different lasers so the seam has no defect.
- Recoating time — the non-melting time spent spreading a fresh powder layer, which caps gains from faster lasers.
- Activity-based costing — a cost method that adds up the cost of each activity (build, depowdering, machining, inspection) instead of using one overhead rate.
- Utilization (load factor) — the share of time a machine is productively building.
- Post-processing — all steps after the build: depowdering, support removal, heat treatment, machining, finishing and inspection.
- Design consolidation — merging an assembly of many parts into one printed part.
- Series production — manufacture of many identical parts on a repeating schedule, as opposed to prototypes or one-offs.
Where this leads
Having looked at the economics, the last chapter weighs everything in this book against it: The honest verdict: what industry still cannot do.