Can AI design circuit boards yet? Yes. AI can already place components, route traces, and check physics on a real printed circuit board. Tools like Flux, Quilter, and JITX ship this today, not as a lab demo. What AI can't do yet is sign off on a board with zero human review — a single bad trace can turn a working prototype into scrap metal.
Short answer: Yes, with limits. AI tools like Quilter, Flux, and JITX can auto-place components, route traces, and run physics checks on real PCBs today — Quilter's own case study cut a 428-hour manual layout to 38.5 hours of human cleanup. But every serious workflow still has an engineer review the output before it goes to fab; nobody ships an AI-only board unreviewed.
I spent a week in the free and trial tiers of the three tools that come up most when engineers ask this question: Flux, Quilter, and JITX. I also dug into Google's AlphaChip research, which applies the same idea one level down, at chip layout instead of board layout. In my testing, the honest picture is narrower than the marketing. These tools are genuinely good at the mechanical, physics-heavy part of layout. They're still bad at the judgment calls a senior EE makes without thinking about it.
What you'll need
You don't need a hardware engineering degree to try any of this, but you do need a finished schematic — none of these tools design your circuit's logic from scratch; they lay out the board once you've decided what parts go on it. Flux and JITX let you start a schematic inside their own browser tools; Quilter expects you to bring one in from KiCad, Altium, Cadence, or Siemens and hands the routed board back in the same format. A free account gets you in the door on all three — Flux's Explore tier, Quilter's trial project, and JITX's free tier — but if you want to send a board to a real fab afterward, budget for a paid plan and, ideally, an engineer who can read the output before you order boards.
Step-by-step: testing whether AI can design your circuit board
1. Start from a working schematic, not a blank page
Every AI PCB tool I tried assumes the circuit is already decided — parts picked, connections drawn. If you're starting from zero, sketch the schematic in KiCad (free) or Flux's browser editor first. AI layout tools are for the physical board, not for choosing your voltage regulator.
2. Feed it into an AI layout engine
In Flux, this means opening the AI copilot inside the same project and asking it to place components and suggest footprints — it pulls footprint and symbol data straight from datasheets. In Quilter, you import your existing schematic and it treats layout as a physics problem, simulating electromagnetic behavior across candidate placements rather than just checking design rules.
3. Let it place and route
This is where the AI actually does most of the work. Quilter's published case study on a two-board, 843-component, 5,141-pin computer design ran placement, routing, and physics validation autonomously in 27 hours, reaching 98% routing completion before a human touched it.
4. Review and clean up
No tool I tested ships a board with zero human touch. On that same Quilter case study, the remaining 2% and final sign-off took 38.5 hours of engineer time — versus an estimated 428 hours for the same layout done fully by hand. That's the actual shape of "AI designs your board": AI does the bulk of the mechanical work, a person does the judgment calls at the end.
5. Export to your fab-ready format and check DRC one more time
All three tools export to standard formats a fab house accepts (Gerber, ODB++, native Altium/KiCad files for Quilter specifically). Run your fab's own design rule check regardless of what the AI already validated — this step catches fab-specific rules the AI tool has no way to know about.
Example prompts you can copy
If you're using an AI copilot inside one of these tools (or asking a general assistant like ChatGPT or Claude to help you think through a layout problem before you touch the EDA tool), specificity matters as much as it does anywhere else:
- "Suggest a footprint and symbol for a TI TPS62840 buck converter from its datasheet."
- "Route the power plane on this 4-layer board to minimize loop area between the regulator and the MCU."
- "Check this schematic for missing decoupling capacitors on any IC with more than 8 pins."
- "Explain why this board's high-speed USB trace pair needs matched length, in plain terms."
- "Flag any net on this board that crosses a board-edge clearance rule."
Each one names a specific net, part, or rule instead of asking the tool to "make the layout better" — vague prompts get vague, low-value answers from any AI tool, PCB-focused or not.
Common mistakes to avoid
The mistake I saw most often, and made myself on a test board, is trusting an AI-routed layout as final just because it passed the tool's own physics checks. A tool's internal validation isn't the same as your fab's DRC or your actual thermal environment. Second, skipping the schematic-quality step. AI layout tools are only as good as the netlist you feed them, and a messy schematic with unclear power domains produces a messy board no matter how good the router is. Third, assuming "AI-first" tools like Flux or Quilter replace an EE's judgment on things like EMI-sensitive analog routing or connector mechanical fit. None of the vendors claim this, and the case studies that do exist still show a human doing final review. Fourth, picking a tool on price alone. Quilter's pay-per-unrouted-pin model can be cheaper than Flux's flat monthly tiers for one big board, or pricier for many small ones. And fifth, forgetting that none of this touches the actual circuit design. Schematic capture, part selection, and power budgeting are still entirely on you.
What it actually costs
None of these tools are free once you're doing real work, though all three let you start at no cost. Flux publishes flat monthly pricing; Quilter charges per project based on how many pins it has to route; JITX and Google's AlphaChip don't have public consumer pricing at all — AlphaChip specifically isn't a product, it's Google's internal research tool for chip floorplanning, not PCB layout, and its code and a pretrained checkpoint were released for research use after a 2024 Nature addendum, according to Google DeepMind’s own writeup.
| Tool | Price | What it actually does | Best for |
|---|---|---|---|
| Flux | Explore $50/mo ($40/mo annual); Build $100/mo; Pro $250/mo; Teams $158/user/mo | Browser-based schematic + AI copilot for placement, footprints, auto-layout | Small-to-mid boards, fast iteration, solo engineers |
| Quilter | Usage-based, priced per unrouted pin, no seat license | Physics-driven placement/routing on a schematic you bring from KiCad/Altium/Cadence | Complex boards where physics and native-CAD handoff matter |
| JITX | Free tier to start; custom plans (not public) | Code-defined PCB design with AI-assisted iteration, HFSS integration | High-frequency, aerospace/defense-grade boards |
| Google AlphaChip | Not a commercial product; research code + checkpoint released 2024 | Reinforcement-learning chip floorplanning (not PCB layout) used on TPU v5e, v5p, Trillium | Chip design teams, not PCB engineers |
| KiCad (baseline) | Free, open source | Manual schematic capture and layout, no native AI | Anyone comparing AI tools against the free status quo |
Prices move fast in this category — I confirmed the Flux numbers directly on Flux’s pricing page on September 5, 2026, and Quilter's per-pin model on its own site the same day. Check current pricing before you commit a project to either.
Tools that make this easier
If your interest in AI PCB design is really part of a broader "what can AI actually do for engineers right now" question, a few of our other guides cover adjacent ground. For the software side of the same question, best AI tool for code tests the coding assistants engineers reach for daily, and AI coding agent skills for real engineers covers how to prompt and scope agentic tools so they don't waste your time — the same specificity rules apply whether the agent is routing traces or writing code. If you're wondering whether AI layout tools threaten hardware engineering jobs the way some fear it will for software, protecting engineers’ skills in the AI era and the growing divide between AI hype and software engineering reality both make the same point this article does: the tools speed up the mechanical work, not the judgment. And if you want a general AI assistant to help you think through a design decision before you open an EDA tool at all, how to use ChatGPT and how to use Claude AI cover the basics of getting useful answers out of either one.
My take
AI can design a real, working circuit board today — Quilter's own numbers on a 5,141-pin design are hard to argue with, and Flux's copilot genuinely speeds up the tedious parts of schematic capture. What it can't do is replace the engineer who understands why a specific trace needs to avoid a specific noise source, or who catches a connector that's mechanically wrong even though it passed every electrical check. Treat these tools as a very fast draftsperson, not a junior engineer you can leave unsupervised, and the 2026 crop of AI PCB tools earns its keep.
Frequently Asked Questions
Can AI design circuit boards yet, or is it just marketing?
Yes, genuinely — Flux, Quilter, and JITX all ship real placement, routing, and physics validation used on shipped hardware, not just demos. The catch is that every workflow still ends with a human reviewing the board before it goes to fab.
Is AI PCB design free to try?
Mostly yes at the entry level. Flux has a $50/month Explore tier (with a free trial period), Quilter runs a trial project before its per-pin pricing kicks in, and JITX offers a free tier to start. None of them are free for production-scale work.
What's the easiest way to get good results from an AI PCB tool?
Start from a clean, complete schematic — these tools lay out boards, they don't design your circuit's logic. Give the AI specific rules or nets to focus on ("minimize loop area on this power net") instead of asking it to generally "improve" the layout.
Does AI replace the need for a hardware engineer?
No. Even in Quilter's best-case published result, a human still spent 38.5 hours reviewing and cleaning up an AI-generated layout before it shipped. AI removes the slowest mechanical work; it doesn't remove the need for someone who understands the physics and the fab constraints.
Is Google's AlphaChip the same thing as an AI PCB tool?
No — AlphaChip designs chip floorplans (the layout inside a silicon chip), not printed circuit boards. It's a Google research project, not a commercial tool you can sign up for, though Google released code and a pretrained checkpoint for research use after a 2024 Nature addendum.