3D Printing Hardware Prototypes: When It Works and When It Lies to You

Printing is a commodity: a bureau or a $400 desktop machine will run your file, and the file matters more than the machine. The question that costs founders months is which parts of a printed prototype you are allowed to believe. A printed enclosure answers some questions honestly, answers others in a way that flatters the design, and answers a third set with confident nonsense that surfaces after the injection tool is cut.

The Real Question Is Not Whether It Can Be Printed

Almost anything can be printed, which is why "can we print it" is a dead end. The useful question is what claim the part has to support. "The board, the battery, and the speaker fit in the volume I promised" is answered by an overnight FDM part for $20. "The seams look clean and the button feels right in a stranger's hand" needs a resin part, sanded and painted, at ten times the cost. "The latch survives 5,000 cycles" is not answered by a printed part in any process at any price. Write the claim down first.

FDM, Resin, and SLS/MJF: What Each Process Actually Buys You

Three processes cover almost everything a hardware team needs, and picking the wrong one wastes a week.

• FDM (PLA, PETG, ABS): 0.1-0.3 mm layers, about ±0.5 mm on a handheld part, visible layer lines, and roughly half the in-plane strength across the layers on a well-tuned machine, as little as a fifth on a badly cooled one. Orientation is a design decision, not a print setting. Right for volume checks and jigs. About $10-60 per enclosure half.
• SLA/resin: 25-100 micron layers, ±0.1-0.2 mm on small features, and a surface that photographs like a molded part after light sanding. Most photopolymers are brittle, creep under load, soften somewhere between 45 and 70 °C, and yellow in sunlight. Right for cosmetic reviews and tight fit checks. About $40-150 per half, one to three days.
• SLS / MJF (powder-bed nylon, usually PA12): ±0.3 mm, near-isotropic strength, and a real engineering thermoplastic that bends instead of snapping. The surface is matte and grainy, and the parts are porous enough to wick water unless sealed. Right for functional parts, field-test units, and small runs. About $60-250 per part, three to seven days.

Those are Bay Area bureau prices in 2026 for a hand-sized part. The print is rarely the expensive line: six iterations might cost $600 in parts, and the CAD work behind them is the real budget, as with hardware prototyping costs. If you are still choosing between resin families and filled nylons, our review of printing materials goes deeper on the chemistry.

What a Printed Part Tells You Honestly

Printed parts are very good at geometry and human questions. Does the assembly go together in an order a technician can follow, or does the last screw need a third hand? Is there room to route the battery cable clear of the antenna? Can someone with weak hands open the lid? A $20 print answers all three a week before a machined part would.

Where Printed Parts Lie to You

The trouble starts when a printed part is asked to predict how a molded part behaves. Five places it misleads you:

• Snap-fits. A resin cantilever is stiff with almost no elongation: it clicks in once, feels excellent, and cracks on the fourth cycle. An FDM snap shears along a layer line. PA12 misleads in the opposite direction: more compliant (roughly 1.5-1.8 GPa flexural modulus against 2.2-2.5 GPa for ABS) and far more ductile than the ABS you will ship, so a hook tuned to feel right in nylon needs noticeably more force in ABS and is likelier to crack than to flex when it gets it.
• Living hinges. A polypropylene living hinge works because flow through the thin section aligns the polymer chains during molding, and no printer does that. A molded PP hinge survives hundreds of thousands of cycles; a printed one fails in tens. If your product folds, that cannot be validated before tooling.
• Wall thickness and draft. A printer will happily build a 4 mm solid slab, a rib as thick as the wall it stands on, and a vertical face with zero draft. A mold answers that geometry with sink marks, voids, warp, and a part that will not eject. Molded plastic wants a uniform 2-3 mm wall, ribs at 40-60% of it, and a minimum of 0.5-1 degree of draft, rising to 1-2 degrees or more as the texture on a face gets deeper. The full set is in enclosure design rules.
• Thermal behavior. An FDM part at 20% infill is mostly trapped air, so it insulates better than the 2 mm molded wall replacing it. The usual error therefore runs pessimistic: your bench test shows a regulator 15 °C hotter than it will ever be in the molded housing. The error inverts the moment the production design brings in a thinner wall, a metal boss, or a pad to the chassis. Either way the printed number is not the production number.
• Sealing and cosmetics. FDM parts tend to weep along layer boundaries however good the gasket groove is, and powder parts are porous, so an IP claim needs parts from the production process. Printed parts also have no parting line, gate vestige, weld lines, or sink over the bosses. Molded parts have all four, in predictable places.

Print, CNC, Soft Tooling, or Steel: Where Each One Stops Making Sense

CNC machining works in the real bulk material and holds about ±0.125 mm on machined plastic like ABS or POM, ±0.05 mm in aluminum, which makes it right for a sealing surface or a heat-spreading housing: $150-600 per piece, about a week. Soft tooling, an aluminum injection mold, runs $5,000-25,000 and 2-4 weeks, and returns thousands of shots in the production resin with real draft, real sink, and real snap behavior. Five to ten thousand is normal in an unfilled material, fewer with glass fill or an abrasive colorant. Hardened steel is $20,000-80,000 and 6-10 weeks to T1 samples.

Where the crossover sits depends far more on part complexity than on volume, and anyone quoting a single unit number is guessing. A simple two-piece shell with no side actions can justify soft tooling in the low hundreds of units. Steel earns its place somewhere north of 10,000, or earlier if the resin is glass-filled and eats aluminum, if cycle time drives your unit cost, or if you need a tool that will still be running in five years. A part with four side actions, a textured A-surface, and an overmolded seal may not pay a steel tool back until well past that. Price the tool, divide by the units you believe in over 18 months, compare against the per-part difference, and do it before the EVT build rather than after.

Printing Around the Electronics

A printed enclosure earns its cost when it is checked against a populated board, not a bare PCB outline. Connector shroud heights, the electrolytic cap nobody modeled, the crystal taller than its datasheet drawing: those kill a fit check. Print the housing, drop in the real Rev A board, plug in the real cable. A USB-C cutout sized to the receptacle instead of to the overmold on a chunky third-party cable is one of the most common mechanical bugs we find; it survives to tooling because everyone tests with the slim cable from the box.

Plastic also detunes antennas. A 2.4 GHz chip antenna tuned in free air shifts as soon as a wall sits 3 mm away, and how far depends on the material: ABS and PC sit near a dielectric constant of 2.7-3.0, while nylon runs higher, around 3.0-3.6, and an SLS part is porous enough that its effective constant lands below solid nylon, so an SLS housing is representative of neither molded nylon nor the PC you will ship. Define the keep-out as a 3D volume, keep metal and metallic paint out of it, and tune the matching network on production-process parts. It is the same fight for millimeters as a wearable antenna.

A Printed Part Is Not a Manufacturing-Ready Design

An STL file is a bag of triangles: no tolerances, no material, no draft, no datums, no surfaces a toolmaker can re-machine. Hand one to a molder and the quote comes back for something other than your product. What a manufacturer needs:

• Native CAD plus STEP (AP242), with a parametric history someone else can edit without rebuilding the model.
• 2D drawings with datums and GD&T on the dimensions that matter — PCB standoffs, sealing faces, connector cutouts — plus a general tolerance class for the rest (ISO 20457 covers plastics).
• A material spec naming grade, colorant, and flame rating, not just "ABS". A few printed materials do carry a UL94 V-0 listing — ULTEM 9085, some FR nylons, some FR photopolymers — but the listing attaches to a specific material-and-process combination at a stated minimum wall thickness, so a V-0 resin printed on the wrong machine at the wrong wall is not a V-0 part.
• Draft everywhere, a proposed parting line, and agreed gate locations, since both leave visible marks on the shipped product.
• Cosmetic zones: which surfaces are A-grade, which texture (a Mold-Tech number, not "matte"), and which defects are acceptable where.

Where We Draw the Line

inteRFace R&D does not run a print farm and will not sell you a print. If you have a finished file and need parts, a bureau or a desktop machine is cheaper and faster, and we will name one. What is not a commodity is the rest: deciding what a prototype has to prove, doing the industrial design and the mechanical engineering as one job so the printed version predicts the molded one, choosing between print, CNC, and soft tooling against a real cost model, and writing the package a molder can quote. We are on-site at first-article and pilot to hold the CM to that package, but we do not cut tools, mold parts, staff a resident process engineer at your factory, or run a compliance lab.

Frequently Asked Questions

Can I ship a 3D printed enclosure to real customers?

Yes, at low volume and with conditions. SLS/MJF nylon is the usual choice, and a few hundred units is a sensible ceiling before per-part cost becomes a permanent tax on margin. Cosmetic variation between builds is wider than with molded parts, and moving to a molded material later changes the material of record, which can reopen certification testing.

How accurate is a printed part against my PCB?

Resin lands within about ±0.1-0.2 mm on small features, MJF around ±0.3 mm, FDM around ±0.5 mm. Design the clearance rather than the interference: nominal gaps of 0.3 mm around connectors and 0.2 mm on slip fits survive across processes, while a press fit tuned on a resin part will be either loose or impossible in nylon.

What should the last print before tooling be?

Not a pretty one. It should be the manufacturing-ready model built as a print: uniform wall, draft on every vertical face, parting line and gate locations already chosen, bosses and snaps sized for the production resin rather than for whatever is in the printer. Print that and the toolmaker's DFM report comes back as a short list of confirmations instead of a redesign. What the print still cannot answer — snap fatigue, seal integrity, flammability — you buy off on soft-tool parts, not another print.

If You Are Not Sure Which Prototype You Need

Send the CAD you have, even a rough sketch and a board outline, with one sentence on what the next prototype has to prove. We will tell you which process fits, roughly what it costs, and what it will not answer. Send it to an engineer at https://interfacernd.com/#contact.