A former SpaceX engineer taught an AI the laws of physics and pointed it at the most stubborn job in electronics: laying out a circuit board. It does the whole thing - placement, routing, validation - in hours.
PROJECT SPEEDRUN: a hand holds a small green computer board. No human routed a single trace on it. The AI did - and then checked its own physics homework.
In a Los Angeles office, an engineer typed a netlist, pressed go, and walked away. A few hours later a full circuit board came back - parts placed, traces routed, electromagnetics and heat already checked. In under a week, that board became a working computer. The engineer never moved a single trace by hand.
This is Quilter's normal Tuesday. The company makes the first physics-driven AI that lays out printed circuit boards on its own. Not an autorouter that nudges a few wires. The whole job - the part most hardware teams dread, the part that quietly eats weeks of every product cycle.
"Just as the compiler made programming faster and more accessible, Quilter aims to remove the last manual barrier in electronics design."
Every gadget you own starts as a board, and every board starts as a person hunched over CAD software, dragging copper lines around a rectangle for weeks. The chips get faster. The schematics get smarter. And then layout - the slow, manual, error-prone part - hits the brakes. It is the step that turns "we have an idea" into "see you next quarter."
Sergiy Nesterenko watched this up close. For five years at SpaceX, he built radiation-tolerant electronics for the Falcon 9 and Falcon Heavy second stage. He saw schedules bend around board spins. A trace too close to another, a thermal mistake, an electromagnetic surprise - any of it could mean another revision, another week, another launch slipping. The hardware was extraordinary. The way it got laid out was, frankly, medieval.
The usual fix is to throw more engineers at it, or buy an autorouter that does maybe 30% of the job before a human takes over in frustration. Quilter's bet was different: don't teach a machine to imitate tired humans. Teach it the physics, and let it find layouts a human would never think to try.
"Electrons obey physics, not opinions."
Most AI learns by copying examples. Quilter refused the obvious shortcut - and there is something pleasingly contrarian about a company whose entire edge is that it ignores the data everyone else fights over. It trains on no human design data at all. Instead, reinforcement-learning agents place components, route traces, then run physics simulations to grade their own work. Millions of iterations later, they have learned what actually produces a board that works.
Nesterenko - a UC Berkeley triple major in math, physics, and chemistry - calls the long-term goal "the compiler for hardware." Programmers stopped hand-assembling code decades ago. Hardware engineers never got that gift. Quilter wants to hand it to them.
The AI never trains on people's layouts. It learns routing from physics simulation, so it inherits none of our bad habits - and keeps customer designs private.
Electromagnetics, thermodynamics and manufacturing rules are checked while the board is being built, not flagged after.
Dozens of candidate layouts generate at once. You pick from real options instead of waiting on one.
Sergiy Nesterenko leaves SpaceX and founds Quilter to attack PCB layout at the root.
Benchmark leads the Series A, with partner Eric Vishria backing the physics-first thesis.
The autonomous layout platform opens to engineers in public beta.
Quilter launches a free tier - autonomous layout with unlimited iterations.
An engineer uses Quilter to autonomously design and verify a working computer in under a week.
Index Ventures leads; Nina Achadjian joins the board. Total raised reaches about $40M.
You hand Quilter a netlist and your constraints. It returns fab-ready boards - differential pairs matched, DDR timing honored, clearances respected, the whole layout placed and routed. The jobs that used to define a quarter now fit inside an afternoon.
Quilter charges by pin count - the complexity of the board - not by how many engineers log in. The AI does the routing, so you are paying for the work, not the chairs.
"Designs correct by construction - surfacing solutions a human would miss."
Speed is the whole pitch, so it had better be true. Here is what customers report on the kind of work Quilter targets - the same board class, by hand versus by AI.
Bars scaled to the manual baseline. Figures are customer-reported ranges, not lab guarantees - your board, your mileage.
The proof is not only on screen. Quilter says Fortune 500 aerospace, defense, automotive and consumer-electronics teams - roughly half a trillion dollars in market cap - are adopting the platform. Boards designed with it have been manufactured, tested, and put to work on Earth and in orbit.
"Quilter-designed boards have been tested, manufactured, and put into use - on Earth and in orbit."
The money agrees. A $10M Series A from Benchmark in 2023, then a $25M Series B led by Index Ventures in October 2025, bring the total to about $40M. The cap table reads like a hardware hall of fame: Benchmark, Index, Coatue, Root Ventures, and semiconductor veteran Lip-Bu Tan.
Quilter has a name for the world it wants: "Hardware-Rich Development." Software teams ship dozens of builds a day because compiling is free. Hardware teams ship a board a quarter because layout is expensive. Quilter wants to collapse that gap until a physical board is as casual to spin as a code commit.
It is a big claim, and the company is refreshingly blunt about why it might be right: electrons do not negotiate. If an AI genuinely understands the physics, it does not need permission, taste, or twenty years of intuition. It just needs to be correct - over and over, in parallel, overnight.
That reframes the engineer's job, too. Less time dragging copper, more time deciding what to build. Quilter is not pitching the end of hardware engineers; it is pitching the end of the part of their week they like least.
The competition - DeepPCB, Flux.ai, and incumbents like Cadence, Altium and Zuken - is circling the same prize. Quilter's wager is that physics-first beats pattern-matching when the stakes are a board that has to actually work.
Return to the office, the netlist, the engineer who walked away. A few hours later, a working computer - designed by an AI that never saw a human do it first. That image used to be a demo. Quilter is trying to make it a habit.
If they are right, the slowest step in building electronics quietly disappears, the way hand-assembling code did. Robots, satellites, medical devices, the next strange gadget - all of them spend less time waiting on a rectangle full of copper. The hand in the photo stops being a milestone and starts being a Tuesday.
The boards still have to work. They have to fly. So far, some already do.
Quilter taught a machine physics so engineers could stop routing traces and start designing the things those traces are for.