CNC Machining vs 3D Printing: Which to Choose and When

CNC machining and 3D printing are the two default answers for a custom plastic or metal part, and they split the decision cleanly. CNC machining cuts a part from solid stock, holding ±0.025 to 0.13mm in isotropic material that carries full load in every direction, which is why it owns production parts, tight fits, and anything in metal. 3D printing builds the same part layer by layer with no tooling, so it owns complex internal geometry, one-off parts, and anything that must exist tomorrow morning. Neither process is better; they trade precision and material strength against geometry freedom and setup cost, and this page works that trade with concrete numbers. If you want the one-row-per-process view across all nine common processes first, the manufacturing process comparison matrix summarizes it, and this page goes deep on the single decision between cutting and printing.

The core trade: subtractive precision against additive freedom

A CNC mill starts with a billet, plate, or rod of a real production material and removes everything that is not the part. The properties of the finished part are the properties of the stock it came from, held evenly in every direction, and the machine controls dimensions to a tolerance band you can specify. The cost of that control is setup: programming, fixturing, and often multiple operations, paid before the first chip flies. The constraint is geometry, because a rotating cutter must physically reach every surface.

A 3D printer inverts every one of those terms. It starts with nothing and adds material, so geometry that no cutter could reach, such as enclosed channels, lattice fills, and nested assemblies, builds as easily as a cube. There is no setup to speak of: a sliced file starts printing, which is why one part is cheap and fast. What you give up is precision and material, because the part is built from a limited palette of filaments, resins, or powders, deposited a layer at a time, and the layer process leaves its fingerprints in the dimensions, the surface, and the strength.

What each side gives up

The machined part cannot have an enclosed cavity, because the tool needs a path in and out, and internal corners carry a radius of at least 0.2 to 0.5mm, the smallest practical endmill radius. The printed part cannot match the machined part’s fit or load capacity: FDM parts run 20 to 30 percent weaker across their layers than within them, and even the best resin part is a photopolymer, not an engineering stock. Every real part decision is a ranking of those four factors: tolerance, strength, geometry, and quantity.

When to choose CNC machining

Choose CNC machining when the part must fit, carry load, or repeat. The clearest case is a part with mating features: a housing with bearing bores, a manifold face with O-ring grooves, a bracket with reamed dowel holes. Those features need dimensional control in the ±0.025 to 0.13mm class, which no printer holds, and they need it in a material whose properties do not depend on build orientation.

Metal parts push the same way. If the part is aluminum, steel, stainless, or titanium, machining is the default: the CNC machining hub covers the process families, and CNC milling covers the most common one. A machined 6061-T6 part carries roughly 290 to 310MPa tensile strength with yield near 276MPa, and it carries the same numbers loaded in X, Y, or Z. Printed polymers are polymers: useful, increasingly capable, and still plastics, several times below that strength level.

Volume is the third trigger. Machining pays a setup cost once per batch, then each added part costs only cycle time and material, so the per-part price falls steeply as quantity grows. A part that quotes badly at quantity one often quotes well at quantity fifty, which is the exact mirror of the printing cost curve described below.

Tight fits and repeating batches

The pattern to watch for is a drawing with toleranced features and a quantity above a handful. That combination, fit plus repetition, is CNC territory almost regardless of geometry, because the process was built to hold a dimension the same way ten thousand times. Printing was not; each printed part rebuilds its own accuracy from scratch, layer by layer.

When to choose 3D printing

Choose 3D printing when geometry, urgency, or quantity-one economics dominates. The strongest case is geometry no cutter can reach: a part with an internal serpentine channel for coolant or air, a lattice structure that trades weight for stiffness, a conformal duct that wraps around other components. A printer builds these without blinking, while machining cannot build some of them at all and needs the rest split into pieces and fastened together.

The second case is the single part needed quickly. Printing skips programming, fixturing, and machine queue, so a part can go from file to build in minutes of preparation. For concept models, fit checks, jigs, and fixtures, that immediacy is the whole point, and the looser tolerance is irrelevant to the job the part does.

The third case is small, complex plastic parts at low volume, where the material palette is enough and the geometry is fussy. Between the printing processes, FDM is the low-cost generalist, SLA wins on fine detail and smooth finish, and SLS and MJF win on functional strength without supports; the FDM vs SLA vs SLS comparison settles that inner choice.

Where printing stops

Printing stops where the part must hold a tight fit, carry serious or directional load, survive heat or chemicals beyond the polymer palette, or repeat identically across a long run. A printed part that must mate with a machined part usually needs its mating features opened up: printed-part clearance guidance runs about 0.3 to 0.7mm per wall for assembled parts. And a metal requirement does not simply convert FDM to “metal printing”; metal additive is a separate process family, covered on the 3D printing hub, with its own economics.

Worked examples

A few parts make the boundary concrete. Take a sensor housing with two reamed dowel holes at ±0.05mm and a sealing face: machine it, in aluminum or PEEK, because the fit requirements and the isotropic material decide the process before cost is even discussed. Take the same housing’s coolant adapter, a curved duct with a 6mm internal channel that snake-glides around the sensor body: print it in SLA for the fine detail or MJF nylon for toughness, because machining that internal channel is impossible and splitting the part adds joints and leaks.

Take a robot gripper finger: printed MJF nylon is the standard answer, because the part is geometrically intricate, carries moderate load in a known direction, and exists in quantities of one to twenty. Now stiffen the requirement, same geometry, but the finger must survive a 40kg crush load and a 120°C washdown: machine it from aluminum, and the complexity that made printing attractive becomes pocketing and 5-axis work that machining handles at a price.

Volume moves the same part across the line. A cable guide printed in FDM at quantity one is nearly free of setup cost; at quantity five hundred, the same guide is a candidate for machining from extruded stock, an injection-molded part if demand holds, or a sheet-metal form if the shape allows. The crossover is discussed with numbers in the cost section below.

Tolerance and surface finish compared

CNC milling holds ±0.025 to 0.13mm depending on tolerance class, with ISO 2768 defining the general bands (class f, fine, for metals runs ±0.05mm on a 0.5 to 3mm dimension; class m, medium, ±0.10mm). Among printers, SLA leads at ±0.05 to 0.15mm on small parts, MJF follows at ±0.2 to 0.3mm up to 100mm, FDM spans ±0.1 to 0.5mm by material, and SLS runs ±0.3mm under 50mm and ±0.5mm above. Two facts matter more than the raw bands. First, printed tolerances widen with part size faster than machined ones, because error accumulates over hundreds of layers. Second, a machined surface can be ground and lapped below Ra 0.8µm, while a printed surface starts at its as-built finish and improves only by secondary processing.

The finish ladder

As-built finishes stack in a stable order: SLA resin at Ra 0.5 to 2µm is the smoothest printed result, near molded quality; SLS and MJF sit at Ra 3 to 9µm with a grainy matte texture that dyes well but needs media finishing for cosmetics; FDM shows layer lines at Ra 4 to 12µm that usually need sanding or vapor smoothing for any visible surface. As-machined CNC lands at Ra 3.2µm, fine machining reaches Ra 1.6µm, and grinding goes below Ra 0.8µm. In practice, finish decides the process for visible and sealing surfaces: a clear lens mold is machined and polished, an appearance model is SLA, and an internal bracket can be anything.

Strength, isotropy, and material range

Strength differences come from two places: the material itself and how the process assembles it. Machining cuts from wrought or molded stock, so a machined part is isotropic, and the designer reads alloy properties straight from the datasheet, such as 6061-T6 at roughly 310MPa tensile and 276MPa yield, or 7075-T6 at 572 and 503MPa. Printing builds the material at the same time as the part, which changes the rules. FDM parts are anisotropic, 20 to 30 percent weaker across the layers, so orientation becomes a strength decision. SLS and MJF nylon parts are near-isotropic and genuinely functional, which is why they anchor printed end-use parts. SLA resins are stiff for their weight but relatively brittle, fine for form and fit, weaker for snap fits and impact.

Material range is wider on the machining side in metals, wider on the printing side in shape. A machine shop’s palette is the materials catalogue: aluminum alloys, carbon and stainless steels, brass, titanium, plus machinable plastics from acetal to PEEK. A printer’s palette is its feedstock: FDM covers PLA, ABS, PETG, PC, nylon, and TPU filaments; SLA covers resin families tuned for toughness, temperature, or castability; SLS and MJF are essentially nylon-family (PA12, PA11, TPU). Metal additive exists as DMLS and similar processes, building inside roughly 150 to 400mm build volumes, but it is a specialist route with its own cost structure, not a drop-in substitute for machining a metal part.

Unit cost by volume: where the lines cross

The cost structures differ in shape, not just height. Machining cost behaves like a fixed setup plus a linear per-part term: cost ≈ setup + N × (cycle time + material). Printing behaves almost purely linearly: cost ≈ N × (build time + material), with little or no setup. At N = 1, the machined part carries the whole setup alone, so printing wins. As N grows, the setup share per part shrinks toward zero while the printed part keeps paying full build time every copy, so the machined line crosses below and keeps falling.

The crossover quantity is therefore arithmetic, not a mystery: it sits where per-part savings from machining have repaid the setup. If a job’s setup cost equals the per-part premium of printing for 20 parts, the lines cross near 20 units; if it equals 200 parts’ worth of premium, they cross near 200. No honest page can print one universal crossover number, because part size, cut volume, complexity, material, tolerance, and finish all move both lines, and the site’s cost comparison by process page keeps to tiers and drivers for exactly that reason.

What moves the crossover

Four drivers do most of the moving. Part size: printing cost scales with volume and height, machining with removed material and cycle time, so large simple parts cross to machining early. Complexity: every machined feature adds cycle time, while complex print geometry is often free, so intricate parts stay printable longer. Material: printing a metal-adjacent requirement is impossible in polymer, and metal additive is far costlier than either, which forces machining. And finishing: a printed part that needs machining anyway for its mating faces carries both processes’ costs, so a hybrid part only makes sense when the printed geometry is truly unmachineable.

Lead time and setup

Lead time follows the same structure as cost. Printing front-loads nothing: slice, build, post-process, done, so a single part arrives fast and each added part adds its own build time. Machining front-loads programming, fixturing, and queue, then runs a short cycle per part, so its first part is slower and its hundredth is quick. Powder-bed printing splits the difference at batch scale, because one SLS or MJF build nests many parts in a single run. The site’s lead-time benchmarks page compares these structures across processes without day counts, since real lead time depends on the shop’s queue and is confirmed at quote.

Geometry freedom and part size

The geometry ledger runs clearly in printing’s favor. A printer builds enclosed internal channels, thin-walled organic shapes, lattices, and pre-assembled mechanisms, constrained only by minimum feature size (SLA reaches 0.5mm holes and 0.2mm walls; SLS needs about 0.8 to 1.2mm walls; MJF 0.3 to 0.5mm) and by support rules, with FDM needing supports under 45-degree overhangs while SLS and MJF powder self-supports past about 30 degrees. Machining is constrained by tool reach, corner radii of 0.2 to 0.5mm, and wall behavior below about 1.5mm under cutting forces, though 4- and 5-axis machines recover much of the geometric ground.

Part size runs the other way. A printer’s part is bounded by its build box, in the few-hundred-millimeter class for SLS and MJF, smaller for desktop SLA, with large-format FDM extending further. A machining center’s limit is its travels and table, and large gantry machines cut fabrications measured in meters. Small parts fit both processes; large parts effectively choose machining for you.

How to choose

Rank the four factors and the process names itself. Need a tight fit, a metal, or isotropic strength at any quantity: machine it. Need impossible geometry, one part, or ten intricate plastic parts: print it. Need flat sheet parts instead: go to laser cutting or sheet metal fabrication. Need thousands of the same plastic part: injection molding overtakes both once tooling amortizes, and internal profiles in hardened steel belong to EDM. The matrix covers those neighbors in one table, and the hubs, CNC machining and 3D printing, carry the detail behind each family.