AI can design your board. Can you source the parts?
We generated hardware designs and BOMs with AI, then checked them with Parter’s live supply chain data to identify sourcing risks and better component choices.
AI can now get you to a working prototype fast. OpenAI’s “Accelerating engineering for the physical world” describes exactly that, followed by rounds of review for manufacturing limits, tolerances, and assembly. But without live supply chain data, AI picks parts that are obsolete, out of stock, or months away. That data belongs in the loop from the first component choice.
At Parter, we run experiments to understand how access to our supply chain and manufacturing data changes AI’s design decisions. We asked several frontier models to design a wide range of boards and write their BOMs, then review their own choices and check production readiness. Then we ran every BOM through Parter. Most of them had at least one part that would cause trouble in production.
What the AI’s own checks missed
Choosing a component means more than matching specs. Is it still in production? Can you buy the exact ordering option? How long will it take to get more stock? Is there a lower-cost alternative that meets the design’s requirements?
Christian Keil makes the same point in “Superhuman Hardware Design Is The Easy Part”: better AI design doesn’t remove the work of sourcing and manufacturing.
In the iterations we reviewed, extra validation reduced some flagged issues. But the BOMs still carried risks. Parter’s BOM optimization surfaced alternatives and the tradeoffs behind them, giving us concrete changes to review.
Some of what we found surprised us. The AI had checked its own work, and the parts still didn’t hold up. Here are three of them.
Matching key specs, a lower listed price
Sometimes nothing is wrong with a part. It just costs more than it needs to.
One BOM included KEMET C1210C106K5RAC7800 at $0.679 per unit. Parter suggested KYOCERA AVX 12105C106KAT2A at $0.251, with matching capacitance, tolerance, voltage rating, dielectric, and case size.
Both were listed as active, in stock, and on an 18-week factory lead time. At the displayed prices, the alternative was $0.428 less per unit, making it a cost-saving candidate with no listed lead-time penalty.
Data from Parter as of November 8, 2026. Prices shown are USD per unit at 5,000 units, rounded to three decimals. Differences use those displayed prices. Suggested alternatives require engineering validation for the design.
An obsolete capacitor, swapped for an active one
In another experiment, the agent selected Murata GRM188R71H103KA01J. Parter flagged it as obsolete and non-stock, with a listed factory lead time of 42 weeks.
Parter suggested GRM188R72A103KA01D: active, in stock, and listed at 17 weeks, a 25-week reduction. It keeps the same 10nF capacitance, X7R dielectric, and 0603 case size, with a 100V rating instead of 50V.
The exact ordering option matters here, too. The original part’s 7-inch reel version, GRM188R71H103KA01D, is also obsolete. Changing reel size alone wouldn’t resolve the lifecycle issue.
In stock today, 61 weeks factory lead time
Samsung CL31B106KAHNNNE was active and in stock. Its listed factory lead time was 61 weeks.
Parter suggested Murata GRM31CR71E106KA12L, with matching key specs and a 19-week factory lead time. That’s 42 weeks shorter.
The alternative cost slightly more, but had a shorter factory lead time, making it worth considering for future builds.
Earlier is easier
Each finding gives the next design iteration something concrete: an active replacement, a shorter lead time, or a lower-cost candidate. The opportunity is to bring that context into component selection itself.
The earlier you catch a risky part, the easier it is to fix. Engineers still set the requirements and make the call. They just deserve the full picture while the design is still easy to change.
The same principle applies to AI design tools. Engineers set requirements, weigh tradeoffs, and approve changes. Giving their tools access to Parter’s data earlier helps them make those decisions with fewer gaps.
That’s why we built Parter MCP. It brings our live supply chain data into your AI workflow, so your agents can check parts while they design.
Designing hardware with AI? We’re opening early access.
