This question usually gets treated as a cost problem, but the published data points to a different answer. A Protolabs survey of 767 engineers and designers found that respondents ranked time as the top obstacle at every stage of development, mentioned more often than cost, and 82% are continually looking for ways to move faster. On top of that, carbon fiber's price-volume curve is much flatter than metal's: going from 1 to 100 pieces cuts the unit price by only 45%, versus 63% for aluminum. Put the two together and the conclusion is this: ordering in volume to spread cost does little for carbon fiber. Lead time is what you should really be counting.
The One-Minute Version
- Carbon fiber's price-volume curve is flat. $53.42 for 1 piece, $29.60 for 100: only 45% lower. On the same platform, aluminum drops 63% and ABS 60%. The reason is that material makes up a large share of the cost.
- Lead time differs by an order of magnitude. Outsourced waterjet takes 2–4 days and CNC job shops 3–5 days; cutting in-house takes a few hours.
- Taiwanese job shops don't publish prices at all. Every one of them quotes only after you upload a drawing, so every quote request adds another round of waiting.
- There is no minimum order barrier: Taiwanese shops state no MOQ, SendCutSend says you can order 1 part or 10,000, and Protolabs states there are no upfront NRE costs. So "how many do I have to order" is not the real obstacle to outsourcing.
- The real cost is time. At 4 revisions a month, outsourcing means 96–192 working days of waiting per year.
- In-house hidden costs must be counted: dust collection, P100 respiratory protection, tool wear, CAM programming time, and the floor space the machine takes up.
Start with the Published Price Data
We checked the websites of several Taiwanese carbon fiber machining shops. The finding: Taiwanese job shops generally do not publish price lists; all of them quote only after you upload a drawing. One shop in Taichung promises a quote reply within 24 hours and explicitly states no minimum order quantity, but does not publish production lead times.
The price data below therefore comes entirely from US online platforms that publish their prices. Its purpose is to show the cost structure and the shape of the price-volume curve, not to be converted into New Taiwan dollars. For actual costs, request quotes from Taiwanese shops.
Price-Volume Slopes for Three Materials
Unit price from 1 piece to 100 pieces; the carbon fiber line is noticeably flatter
| Term | SendCutSend | Protolabs |
|---|---|---|
| Minimum quantity | No minimum; 1 part or 10,000 parts | 1 to 200 parts |
| Setup fee / NRE | Not mentioned on pricing page | States no upfront NRE costs |
| Starting price | Custom parts from $1 | From about $65 |
| Lead time | Sheet cutting 2–4 days; as fast as within 24 hours | CNC Standard 3 days; Economy 7–18 days |
| Volume discounts | Discounts start at 2 parts, up to 80% | Breaks not published |
| Carbon fiber process | Waterjet only; no laser or CNC routing | — |
| Notably, minimum quantity is not the real barrier to outsourcing: both platforms accept single-piece orders, and Taiwanese shops also state no MOQ. The real difference is lead time. | ||
Turn Lead Time into Cost
Since engineers themselves say time is the top obstacle, let's quantify it. Assume a drone project in R&D revises its frame 4 times a month:
Cumulative Waiting Time Across R&D Iterations
Waiting days accumulated over one year at 4 revisions per month
Because the waiting is serial, not parallel. You can't revise the next version while waiting for this one to be cut, because what you change next depends on what you learn from assembling and testing this one. That is the nature of R&D iteration: each round waits on feedback from the previous one.
The Protolabs survey backs this up: the average product development cycle runs 22 months, with 3–4 months of early-stage prototyping and 3–4 months of late-stage prototyping. In other words, prototyping takes up a third of the whole development cycle.
An Honest List of In-House Hidden Costs
Counting only the machine price underestimates the true cost of bringing cutting in-house. These items have to be considered as well:
| Item | Type | Notes |
|---|---|---|
| Machine | One-time | Depends on model and options |
| Dust collection system | One-time, not optional | Carbon fiber dust is conductive and can exceed regulatory exposure limits |
| P100 respiratory protection | Consumable, not optional | Ordinary dust masks offer limited protection against particles under 5 µm |
| Vacuum table or fixtures | One-time | Practically required for thin sheet |
| Tool wear | Ongoing | CVD diamond-coated about USD $181–$216 each; uncoated $6–$55 |
| Spoilboard | Ongoing | MDF, replaced and resurfaced periodically |
| CAM programming time | Ongoing labor | The most commonly underestimated item |
| Learning curve | Up-front labor | Dialing in parameters, trial and error on workholding |
| Space | Ongoing | A strength of desktop machines, but still needs a dedicated machining area |
| Sheet inventory | Ongoing | But offcuts can be reused; when you outsource, offcuts stay at the job shop |
| Tool prices are actual listed prices from US distributors, accessed September 3, 2026; we found no public source for Taiwan distributor prices in NT$. CAM programming time varies enormously with part complexity, from ten-odd minutes for a simple profile to over an hour for a complex nest. | ||
When to Cut In-House, When to Outsource
| Situation | Recommendation | Key reason |
|---|---|---|
| R&D phase, 2 or more revisions per month | In-house | Waiting time is on the critical path |
| Material validation (ASTM test coupons) at the same time | In-house | Coupons and parts share the same process and material batch |
| Parts fit within 360×300 or 510×760 | In-house feasible | Anything beyond the travel must be outsourced |
| Repeated fine-tuning of the same dimension | In-house | Changing one number and recutting takes tens of minutes |
| 100+ pieces at once, design frozen | Outsource | Outsourcing has lower labor cost per unit in production |
| Parts longer than 760mm or needing 5-axis | Outsource | Beyond what a desktop machine can do |
| No separate space for a machining area | Outsource | Conductive dust cannot share space with an office |
| Only a few jobs a year | Outsource | Idle equipment costs more than the lead-time benefit |
In-house and outsourcing are not either/or. A very common setup is in-house during R&D, outsourced in production: iterate quickly on a desktop machine, then hand the final version to a job shop for production once the design is frozen. You get iteration speed without investing in bigger equipment for production.
This combination has a side benefit too: the drawing files you hand the job shop have already been cut and proven machinable, so you won't find out at the production stage that some inside radius can't be made.
How to Calculate Your Own Break-Even Point
Rather than borrowing someone else's numbers, run the math with your own. You need only five inputs:
| Item | How to get it | Purpose | |
|---|---|---|---|
| 1 | Revisions per month | Go through project records and count actual revisions over the past six months | The most important variable |
| 2 | Parts per revision | A frame revision is typically 6–12 parts | Converts to outsourced part count |
| 3 | Actual quotes from Taiwanese job shops | Must be requested from shops; not available online | Cost on the outsourcing side |
| 4 | Actual outsourced lead time | Ask when requesting quotes; most Taiwanese shops don't publish it | Converts to cost of waiting |
| 5 | Engineer time cost | Per Taiwan's DGBAS 2024 (ROC year 113) data, average monthly regular earnings for all employees were NT$46,450 and total earnings NT$60,984 (Taiwan figures) | Converts waiting into a dollar amount |
| Items 3 and 4 are the key ones, and they can only be obtained by requesting quotes. That also illustrates a real friction in outsourcing: even comparing options means waiting through a round of quotes first. | |||
The commonly repeated claim that "a product goes through 3–5 revisions on average" has no authoritative statistics behind it. Most sources are consulting firm blogs with no methodology or sample described.
Here is what empirical research with an actual sample says: a study analyzing 88 competition teams and 7,956 CAD iteration records found significant differences between high- and low-performing teams in the number, size, and pacing of iterations. But those are differences between teams, not an average you can apply.
So use the actual revision count from your own projects, not someone else's average.
FAQ
Is desktop CNC part quality really comparable to a job shop?
On tolerances, yes. iFlight's own published frame manufacturing tolerances are ±0.2mm on thickness and ±0.1mm on hole diameter, SendCutSend's carbon fiber waterjet is ±0.23mm, and a desktop CNC's ±0.05mm gives 2–4.6x margin over both. The real gap is in capacity and large-format capability, not single-part accuracy.
I have no CNC experience at all. Is the learning curve too steep?
Carbon fiber sheet work is one of the simpler CNC applications: 2.5D machining, a single material, and well-defined parameter windows (tool makers publish cutting data sheets). What really takes some figuring out is workholding and dust collection, not cutting theory. Start with simple profiles in 3mm sheet, get workholding and Z zeroing solid, then move on to thin sheet and complex parts.
Taiwanese job shops don't publish prices. How do I compare them?
When requesting quotes, ask five things up front: unit price, quantity breaks, standard lead time, rush lead time and surcharge, and whether there is a setup fee. Send the same drawing file to at least three shops so you have a basis for comparison. Note that this process itself takes several days, which is one of the hidden benefits of cutting in-house.
Does cutting in-house pay off in production?
Usually not, unless your volume happens to fall within what a desktop machine can handle. Even though carbon fiber's price-volume curve is flat, a job shop's labor cost per unit at high volume is still clearly lower than a desktop machine's (one machine can cut only one sheet at a time). A more sensible setup is in-house for R&D and outsourced for production.
If I only make small non-structural parts, do I still need a machine?
It depends on frequency. If you only cut parts a few times a year, outsourcing makes more sense: the cost of idle equipment (space, maintenance, forgetting what you learned) outweighs the lead-time benefit. But if your project is ongoing, even with only small parts, being able to revise and cut on the same day still has value. One more reminder: carbon fiber dust is conductive and can exceed regulatory exposure limits, so cutting in-house requires proper dust collection and P100 respiratory protection. Do not skip this.
Sources
- Protolabs, Product Development Trends 2024 (survey of 767 engineers: time is the top obstacle at every stage, 82% look for ways to move faster, average development cycle 22 months). Accessed September 3, 2026. https://www.protolabs.com/resources/guides-and-trend-reports/product-development-trends/
- SendCutSend, Sheet Cutting Pricing Examples (carbon fiber $53.42 for 1 / $29.60 for 100; aluminum and ABS for comparison). Accessed September 3, 2026. https://sendcutsend.com/services/sheet-cutting/
- SendCutSend, Pricing (no minimum quantity, custom parts from $1, discounts from 2 parts). Accessed September 3, 2026. https://sendcutsend.com/pricing/
- SendCutSend, Processing Times (sheet cutting 2–4 days, as fast as within 24 hours). Accessed September 3, 2026. https://sendcutsend.com/sendcutsend-processing-times/
- Protolabs, FAQs (1 to 200 parts, no upfront NRE costs, from about $65). Accessed September 3, 2026. https://www.protolabs.com/resources/faqs/
- Protolabs, CNC Machining Lead Times (Standard 3 days, Economy 7–18 days). Accessed September 3, 2026. https://www.protolabs.com/help-center/lead-times/
- Chaur-An Technology (Taichung, Taiwan), CNC machining service page (quotes require a drawing, 24-hour quote reply, no minimum order quantity stated). Accessed September 3, 2026. https://www.chaur-an.com.tw/cnc/
- Directorate-General of Budget, Accounting and Statistics (DGBAS), Executive Yuan, Taiwan: 2024 (ROC year 113) employee earnings statistics (average monthly regular earnings NT$46,450, total earnings NT$60,984 per employee). Accessed September 3, 2026. https://www.dgbas.gov.tw/News_Content.aspx?n=3602&s=234606
- Vrolijk, Deng & Olechowski (2023), Connecting design iterations to performance in engineering design, Proceedings of the Design Society 3:1067-1076 (88 teams, 7,956 CAD iteration records). https://doi.org/10.1017/pds.2023.107
- iFlight official frame spec page (manufacturing tolerances: thickness ±0.2mm, hole diameter ±0.1mm). Accessed September 3, 2026. https://shop.iflight.com/Chimera9-ECO-Frame-Kit-Pro2080
In R&D, the Payoff Isn't Saving Money. It's Turning Waiting into Hours.
Carbon-CNC is a desktop CNC cutting machine for composites, designed and built in Taiwan. It cuts a single part as easily as a batch, lets you recut a revision the same day, and holds ±0.05mm, tighter than job-shop waterjet. To discuss the iteration pace and machine choice for your project, contact us.
Get a Quote