The simple calculation that breaks quickly
The first time you try to calculate the cost of a 3D print, it usually looks simple. The slicer says the part uses 82 grams of PLA. The spool cost $22 per kilogram. So the part costs about $1.80 in plastic.
For a single hobby print, that may be close enough. If you are printing one bracket for yourself, the exact number does not matter much. The trouble starts when that same calculation is used for quotes, repeat orders, Etsy listings, small batch production, or deciding whether a job is worth running at all.
At that point the filament number is not the cost of the part. It is only the easiest number to see.
Why most cost calculations are wrong
Most bad cost calculations are not wrong because the math is complicated. They are wrong because they choose the wrong boundary.
A common approach is to take filament used, multiply it by filament price, add a small percentage, and call it done. Another version adds electricity, which feels more complete, but still misses most of the real workflow. The printer did not only consume plastic. It occupied a machine for several hours, used wear parts, required setup time, and may have blocked another job from starting.
Even electricity is easy to misunderstand. A printer drawing 120 watts for five hours is not usually the expensive part of the job. The expensive part is often the five hours themselves, especially if the machine could have been printing something with a better margin.
This is why two prints that use the same amount of filament can have very different real costs. A simple vase-mode part and a tolerance-sensitive functional part might both use 100 grams, but one may need careful setup, slower speeds, support cleanup, inspection, and a higher chance of being reprinted.
What actually contributes to cost
A more useful calculation starts by separating the things that are easy to measure from the things that are easy to forget.
Material is still part of it. Filament cost matters, and it should be based on the actual spool price, not a rough default that never changes. If PETG costs more than PLA, or a customer wants a specific color that you do not normally stock, that should be reflected.
Machine time matters too. A printer that runs for six hours is unavailable for six hours. That time has a cost even if nobody is standing next to it. Belts, nozzles, build plates, fans, bearings, and hotends wear slowly, but they do wear. Maintenance is not a dramatic expense on every print, but over hundreds of hours it becomes real.
Power should be included, but with the right expectations. It usually will not dominate the cost unless energy is expensive or the print is very long. Still, leaving it out makes the calculation less consistent, and consistency is what makes the numbers useful later.
Then there is human time: loading filament, slicing, queueing the job, removing the part, cleaning supports, checking dimensions, packing, messaging the customer, and dealing with the job when it does not go smoothly. A five-minute task repeated across many prints becomes a real operational cost.
The hidden problem: failed prints
Failed prints are where many pricing models quietly fall apart.
If a part succeeds the first time, filament-only pricing looks harmless. If the same part fails once halfway through, the real cost changes immediately. You used material, machine time, electricity, and attention, but you still have zero sellable parts. If the replacement print succeeds, the cost of the successful part is not just the second print. It also carries the cost of the failed attempt.
Suppose a part uses $2.00 of filament and takes four hours. If it succeeds every time, maybe the price works. But if one out of every five attempts fails, the cost per successful part is higher than the clean slicer estimate. The failed jobs have to go somewhere in the math. Pretending they do not exist usually means the margin is being paid for by your time.
This matters most with parts that look profitable on paper but are unreliable in practice: tall prints with small bed contact, large flat parts that warp, multi-color prints, abrasive materials, or parts with supports that sometimes fuse too aggressively. The slicer estimate describes the print you hoped would happen. Your cost model has to describe what actually happened.
The bigger issue: multiple printers
With one printer, you can sometimes keep the whole picture in your head. You remember that the black PETG job failed last night, or that the small printer is slower but more reliable for a certain part. Once there are several printers running, memory stops being a cost system.
The same file may be printed on different machines with different speeds, nozzle sizes, power draw, reliability, and maintenance history. One printer may finish the part in three hours. Another may take four and a half. A third may be faster but fail more often because the bed surface is worn or the enclosure is not stable enough for the material.
If you sell ten copies of the same part and they were produced across three machines, the cost should be aggregated across the whole batch. Looking only at the successful prints from the fastest printer gives a flattering number, but not a useful one. The customer receives one batch of parts. The business absorbs the cost of all the attempts needed to make that batch.
This is also where uneven machine usage becomes visible. A printer that is technically cheaper per hour may not be cheaper if it causes more failed jobs or needs more operator attention. A slightly slower machine can be the better production choice if it produces more successful parts with less intervention.
A more realistic way to think about cost
The useful metric is cost per successful part. Not cost per sliced file, not cost per attempt, and not cost per gram of filament. Cost per successful part is the number that answers the practical question: what did it actually cost to produce something you can deliver?
To get there, start with each print attempt. Track material used, print duration, printer used, and whether the attempt succeeded or failed. Add reasonable rates for machine time, power, maintenance, and any labor that belongs to the workflow. Then group attempts by the part or job they belong to.
For example, if a customer orders twelve parts and fourteen attempts were needed because two failed, the cost belongs to twelve successful parts, not fourteen theoretical ones. The failed attempts are part of the production cost. Dividing the full batch cost by the twelve deliverable parts gives a number you can actually use for pricing.
This does not need to be perfect down to the cent to be useful. A consistent estimate that includes failures and machine time is far better than a precise filament calculation that ignores half the workflow. The goal is not accounting theater. The goal is to stop making decisions from numbers that are too optimistic.
I ended up building a small tool, My3DMonitor, to handle this automatically for my own multi-printer workflow, but the logic can be applied manually as well. The important part is deciding what counts as the job and making sure failed attempts are not silently discarded.
Final thoughts
The real cost of a 3D print is rarely hidden in one dramatic expense. It is usually spread across small things: a little filament, a few hours of machine time, some electricity, a worn nozzle, five minutes of cleanup, and the occasional failed print that nobody wants to include in the quote.
Filament-only pricing is attractive because it is quick and visible. It can work for casual prints where the stakes are low. But for repeat work, customer jobs, or multiple printers, it becomes too narrow. It tells you what the plastic cost, not what the part cost.
Once you include failures, printer differences, and the time required to produce successful parts, pricing becomes less guessy. Some jobs that looked profitable will look weaker. Some machines will turn out to be better for certain parts than expected. Most importantly, you can stop treating failed prints as exceptions and start treating them as part of the real production system.