Six years ago, I took a job as an applications engineer for a metal additive manufacturing company. I expected to spend my days doing technical validation, material testing, that sort of thing. Instead, I've spent a surprising amount of time telling people not to buy what we sell.
Not because the technology doesn't work. It does—sometimes spectacularly. But I've personally made (and documented) 23 significant mistakes, totaling roughly $180,000 in wasted budget. Not one was a machine failure. Every single one was a mismatch between the part and the process.
The pattern is always the same: someone gets interested in metal 3D printing, sees that Velo3D is a documented SpaceX and Anduril supplier, and assumes the "most advanced" system is the answer to everything. Then the quote shows up, and the math gets uncomfortable.
There's no universal answer here—anyone claiming otherwise hasn't validated enough parts. But after six years, I've landed on four scenarios that cover the vast majority of metal part decisions:
- Scenario 1: Your part has geometry that can't be machined → metal additive manufacturing (the Velo3D Sapphire line is built for this)
- Scenario 2: Your part is machinable and you need more than a few → CNC beats AM on cost, almost every time
- Scenario 3: You're prototyping without extreme mechanical loads → SLS vs FDM 3D printing, and you probably don't need to buy a machine right away
- Scenario 4: You're solving a completely different problem → like the guy who called us about a "metal roofing cutting tool" and actually needed $40 aviation snips
Scenario 1: The part that can't be made any other way
Metal AM's real value isn't "it's modern." It's the ability to create geometry that no cutting tool can reach. Internal cooling channels with complex curves, lattice structures, consolidated assemblies that replace 15 welded parts—that's the world where the Velo3D Sapphire 3D printer earns its keep.
The specific advantage of Velo3D's Sapphire systems is printing without supports. That matters more than most buyers realize. On conventional metal AM systems, overhangs need support structures—thin metal scaffolds that keep the part from sagging during the build. Those supports must be removed after printing. If the geometry is deep inside the part, removal is impossible. The result is a scrapped part or a compromised design.
This becomes critical when you're working with materials like Inconel 718 or titanium Ti-6Al-4V. These aren't cheap. A single failed build in aerospace-grade powder can cost $8,000 to $15,000 just in material and machine time. If your design requires supports in inaccessible areas, the risk profile changes dramatically. According to Velo3D's published product specs (velo3d.com), the Sapphire XC offers a 600mm diameter by 550mm build volume—big enough for serious flight hardware.
SpaceX and Anduril both use Velo3D for flight and defense hardware. That's publicly documented, and it tells you more than the spec sheet will. Those customers put parts through qualification processes that make most engineering teams cry. Passing that bar is a real signal.
Here's a concrete example from my history. In Q2 2023, a customer came to us having spent $38,000 at another AM shop on a part with internal channels that kept collapsing during support removal. We ran it on the Sapphire—no supports, second build was production-ready. Well, technically it was the third. The first was what the customer brought us. That's the exact scenario where metal AM makes sense: the part can't be machined, and it can't be printed on traditional systems without supports that destroy it.
Scenario 2: When CNC machining is the right call
This is the one that embarrasses me a little. In 2022, we produced a bracket on the Sapphire. Simple part—3×4 inches, four threaded holes, 200 units. The customer wanted "the most advanced manufacturing technology," and we were happy to oblige. The parts came out flawless.
The cost analysis said we'd make margin. My gut said this part should never be on a $500,000 metal AM system. I went with the analysis. That was the $8,280 mistake.
For context: a CNC machining shop in Essex, MA quoted the same part at $1,120. Delivered in two weeks. Our price was nine times higher, and our lead time was worse (2022 pricing; verify current rates if you're budgeting now—this gap hasn't gotten smaller).
That's not an unusual gap. Standard machining is a mature, brutally efficient industry. AM can't compete on price for simple geometry, and it shouldn't try. If your search history looks anything like "cnc machining essex ma" and your part has standard dimensions with accessible features, you're probably in the right place already.
In 2021, I had to make the same call under time pressure. A defense customer needed a part in 48 hours. The geometry was borderline—could've gone either way, CNC or AM. I had two hours to decide, and I chose AM because the machine was right there in the lab. In hindsight, I should've asked for the rush CNC quote. AM worked, but the customer paid $2,400 more than necessary. Haste is expensive.
Scenario 3: Prototyping — SLS vs FDM
People love to argue about this one. SLS is "better" than FDM in most technical dimensions—smoother surfaces, better layer adhesion, no visible layer lines, no support marks. But "better" doesn't mean "right for your project."
FDM is completely fine if you're checking fit, clearance, or form. A part printed in ABS or PETG on a desktop machine can tell you whether a housing will snap together. It'll look rough, and it won't survive abuse, but it doesn't need to.
SLS is the right move when the prototype needs to survive handling or testing. Fused nylon powder gives you actual mechanical properties. Parts can be flexed, dropped, and sometimes even installed for a test run. The price per part is higher, but the information value is better.
My rule: if the prototype answers a question about dimensions or fit, FDM is enough. If you need to learn how the part behaves under real stress, SLS is worth the upgrade.
But—and this is the unpopular take—don't buy either machine unless you're running multiple iterations per week. Use a service bureau first. I've seen too many companies buy a $30,000 SLS system for a single project and then never turn it on again. It becomes an expensive shelf display. We had a customer in 2023 who did exactly that with a $50,000 system, and they still contracted us for metal afterward because the SLS machine they owned was suddenly "a depreciating asset they needed to justify." That's not a great reason to use a machine.
Scenario 4: When you just need a cutting tool
Occasionally, the right answer isn't additive manufacturing or subtractive manufacturing. It's a hand tool.
In 2019, a caller asked about a "metal roofing cutting tool." I spent five minutes trying to understand what that meant in an industrial context. It turned out he needed to trim aluminum panels for a barn roof. He'd found us because he searched for "metal cutting" and assumed all metal fabrication required industrial machinery.
I told him to buy aviation snips. Forty dollars, not forty thousand. He was relieved. That call stuck with me because it was the clearest example of a principle I now apply to every inquiry:
Match the tool to the problem, not the problem to the tool.
How to decide which scenario you're in
If you're still with me, you're probably wondering which bucket you fall into. Here's the exact checklist I use with our team—the one that took 18 months and 16 mistakes to build:
- Draw your part. If every feature is machinable with standard tooling, go CNC (Scenario 2). You don't need AM for that.
- Look for internal features. If there's a channel, tight radius, or geometry that physically blocks a cutting tool, that's Scenario 1.
- Know your annual volume. Under 50 parts per year? Use a service bureau. Under 10? Definitely. Buying machinery for 10 parts is how capital budgets get destroyed.
- Check the mechanical loads. No real stress? FDM or SLS is fine (Scenario 3), assuming you've confirmed the material for the environment.
- Re-read your search terms. If you typed "metal roofing cutting tool," stop and buy some aviation snips. You're in Scenario 4, and you're welcome.
This checklist works for us because we've got enough build volume to keep our machines fully loaded. If you're a smaller operation with one proof-of-concept per quarter, the "buy a printer" math doesn't work—service bureaus exist for precisely that reason.
I'm not a supply chain specialist, so I can't speak to vendor selection or global sourcing strategy. What I can tell you from a technical validation perspective is that this checklist has caught 47 potential errors in the past 18 months—each of which would've cost money and credibility.
The worst decision you can make isn't choosing AM over CNC. It's choosing without a framework. And if you've read this far, you're already ahead of where I was six years ago.