Where the money goes when AI designs physical things
AI makes drafting a physical product cheap. Checking it is still scarce, making small parts is cheap but priced all over the map, and the last time a hardware cost collapsed, the losses landed at making and delivery. The evidence for each, the routes ranked by where the losses landed, and a check to run on your own idea. Every figure attributed.
Horizontal bars grow from a zero line, one per row. RapidDirect (before about $112 estimated landed cost): $65.55; Xometry: $118.10; SendCutSend: $150.54; Hubs: $151.58; eMachineShop: $166.81. Five of the nine platforms quoted in the one independent test.
Contents
The answer
A list of 5 rows: Making simple parts is already cheap and fast; AI design output runs more often than it is right; Earlier cost collapses did not reward the people who entered the cheap stage; The 3-5x shop claim is untested; The labour premise has weak support so far. A green tick fills each box in turn.
On 4 October 2026, four posts said AI would open physical design to everyone and move the bottleneck to the factory. We took that as given and asked where a person with software skills, little capital and no factory should stand.
Five research passes were each told to find evidence against the premises.
- Drafting got cheap. Checking didn't. The best AI passed 8.15% of real circuit-board projects once simulation was included.[6] The strong results pair a model with a deterministic checker or an expert's notes.[7][8]
- Making small parts is cheap, and priced all over the map. Boards are advertised from $2 for five.[9] One aluminium bracket's highest quote was more than 2.5 times its lowest, $65.55 to $166.81, across nine platforms.[12]
- The last time a hardware cost collapsed, the losses landed at making and delivery. Of 381 successful Kickstarter design and technology projects, 24.9% delivered on time.[14] Quirky and Shapeways, built to make other people's products at scale, went bankrupt.[15][20]
- The 3-5x small-shop claim has no case behind it. On Census figures, 5x would put a 10-19 person machine shop above every size class in its own industry.[13]
- No flow of office workers into factories has been measured. The clearest AI job signal is a 16% relative decline for 22-25 year olds in exposed jobs.[1]
The tables behind each line, with every figure attributed:
- The evidence behind each vibe manufacturing answer
The tables behind the vibe manufacturing answer: design benchmarks, small-order prices, past cost collapses, small-shop output and labour data.
Where to stand given where the losses landed
The agent, drawn as a hopper, pours drafted circuit boards onto a sieve labelled checks. Most pile up on the sieve; a person labelled You stands at a laptop wired into it. Two boards fall through into a small open box labelled first order, cheap to try, and its flaps close. To the right, a large crate marked with a dollar sign and labelled anything that needs capital rolls along the ground and jams against a gate labelled making, short of a second gate labelled delivery.
The money in factories is going to large players: Hadrian raised $1.37B at a $7.87B valuation in August 2026.[16] The stages that need capital are where the people who entered early lost.
Not yet proven
A list of 6 rows: LLM-generated CAD or boards taken through fabrication, counted; A measured first-time-customer failure rate; A 10-19 person shop at 3x or more at constant headcount; A measured flow of office workers into manufacturing; Sales-cycle length for software sold to machine shops; Failure rates for Amazon FBA, print-on-demand or Shopify brands. The boxes stay empty.
- That a 10-19 person shop can reach 3x or more at the same headcount. No case was found.
- That AI-designed parts survive fabrication. No study took LLM-generated CAD or boards through a factory and counted.
- How often first-time customers' orders go wrong. Only practitioner opinion exists: incomplete specs, over-tight tolerances, minimum-order mismatch.
- That displaced office workers are moving into manufacturing, data centres or robotics.
What this says about how we work
A list of 4 rows: Census August 2026 levels; HWE-Bench 8.15% pass rate; MUSE abstract; Every other figure. A green tick fills the boxes for the rows that apply and the others stay empty.
When a field is loud, we take its claims as given and send the research to find evidence against them.
Then we do the arithmetic nobody posted, rank the routes by where the losses landed, and mark which figures we rechecked against the source: here, the Census August 2026 levels and the HWE-Bench pass rate. That's the reading we'd give your idea before you spend on it.
Run it on your own idea
A person labelled You stands at a bench beside a notepad titled your log. Four lines write themselves: check, timer chip, not for 3 V; check, via almost on a pad; limits, minimums, lead time, tariff; order, what went wrong, followed by a blank. A small box of 5 green boards sits on the bench. A copy of the log shrinks into a book and files into the one empty slot on a shelf of sources, under the note no source measures how often first orders fail.
- Has a working unit been made? If not, don't take anyone's money for it yet.
- What checks the design? Name the design-rule check, the simulation and the vendor's manufacturability check you'll run before ordering.
- What does the smallest order cost, landed? Add tariffs and shipping to the advertised price.
- Quote it three times. One bracket varied 2.5x across platforms.
- Write down what each check and the first order caught. Nobody has published that number, so yours will be the first.
The plain-language version is the post. If you're weighing a physical product and want a second reading, I'd like to hear about it: mackenzie@mackenziebowes.com or @_mackenziebowes.
Sources
- Brynjolfsson, Chandar & Chen (Stanford Digital Economy Lab, 2025). Canaries in the Coal Mine. ADP payroll data. digitaleconomy.
stanford. edu/ publications/ canaries- in- the- coal- mine/ - BLS Occupational Outlook Handbook, electricians. www.
bls. gov/ ooh/ construction- and- extraction/ electricians. htm - BLS Occupational Outlook Handbook, industrial machinery mechanics, maintenance workers and millwrights. www.
bls. gov/ ooh/ installation- maintenance- and- repair/ industrial- machinery- mechanics- and- maintenance- workers- and- millwrights. htm - BLS Occupational Outlook Handbook, metal and plastic machine workers. www.
bls. gov/ ooh/ production/ metal- and- plastic- machine- workers. htm - BLS Occupational Outlook Handbook, software developers. www.
bls. gov/ ooh/ computer- and- information- technology/ software- developers. htm - HWE-Bench (arXiv 2603.18102, 2026). 300 board-level projects checked with static rules and LTspice simulation. arxiv.
org/ abs/ 2603. 18102 - SchGen (arXiv 2605.30345, 2026). arxiv.
org/ html/ 2605. 30345v1 - IoT-SkillsBench (arXiv 2603.19583, 2026). arxiv.
org/ html/ 2603. 19583 - JLCPCB, PCB assembly service page. Vendor prices. jlcpcb.
com/ pcb- assembly - SendCutSend, home page. Vendor prices. sendcutsend.
com/ - Xometry, second-quarter 2026 results (4 Aug 2026). www.
globenewswire. com/ news- release/ 2026/ 08/ 04/ 3338170/ 0/ en/ xometry- reports- record- second- quarter- 2026- results. html - Otron Atlas (1 Mar 2026). Is Xometry expensive? One STEP file quoted on nine platforms. www.
otronatlas. com/ price- comparison/ is- xometry- expensive/ - US Census Bureau, Statistics of US Businesses 2022, detailed firm-size tables (NAICS 332710). Receipts are gross shipments. www2.
census. gov/ programs- surveys/ susb/ tables/ 2022/ us_state_naics_detailedsizes_2022. txt - Mollick (2014). The dynamics of crowdfunding: an exploratory study. Journal of Business Venturing. leeds-
faculty. colorado. edu/ bhagat/ Dynamics- of- Crowdfunding. pdf - Quirky (company). Wikipedia. en.
wikipedia. org/ wiki/ Quirky_(company) - Tech Startups (6 Aug 2026). Hadrian raises $1.37B. techstartups.
com/ 2026/ 08/ 06/ defense- tech- startup- hadrian- raises- 1- 37b- at- nearly- 8b- valuation- to- scale- ai- powered- factories/ - The Robot Report. Foxglove raises a $40M Series B. www.
therobotreport. com/ foxglove- raises- 40m- scale- data- platform- roboticists/ - TechCrunch (17 Jun 2026). Robot training-data collection at XDOF. techcrunch.
com/ 2026/ 06/ 17/ collecting- robot- training- data- is- dirty- unglamorous- work- some- ai- labs- are- already- paying- xdof- to- do- it/ - Lasso Supply Chain Software (Q1 2026). State of digital manufacturing. A vendor survey. lassosupplychain.
com/ research/ q1- 2026- state- of- digital- manufacturing- report- small- midsized- companies- in- the- united- states/ - CG Channel (2024). Shapeways files for bankruptcy. www.
cgchannel. com/ 2024/ 07/ 3d- printing- firm- shapeways- files- for- bankruptcy/