A Los Angeles-based startup has demonstrated what it calls a breakthrough in hardware development: a man-made intelligence system that designed a totally functional Linux computer in a single week — a process that will typically devour nearly three months of expert engineering labor.
Quilter, which has raised greater than $40 million from investors including Benchmark, Index Ventures, and Coatue, used its physics-driven AI to automate the design of a two-board computer system that booted successfully on its first attempt, requiring no costly revisions. The project, internally dubbed "Project Speedrun," required just 38.5 hours of human labor in comparison with the 428 hours that skilled PCB designers quoted for a similar task.
The announcement also marks the primary public disclosure that Tony Fadell, the engineer who led development of the iPod and iPhone at Apple and later founded Nest, has invested in the corporate and serves as an advisor.
"We didn't teach Quilter to attract; we taught it to think in physics," said Sergiy Nesterenko, Quilter's chief executive and a former SpaceX engineer, in an exclusive interview with VentureBeat. "The result wasn't a simulation — it was a working computer."
Circuit board design stays the forgotten bottleneck that delays nearly every hardware product
The announcement shines a light-weight on an unglamorous but critical chokepoint in technology development: printed circuit board layout. While semiconductors and software have received enormous attention and investment, the green fiberglass boards that connect chips, memory, and components in virtually every electronic device remain stubbornly manual to design.
"Besides auto-routers, the technology really hadn't modified for the reason that early '90s," Fadell told VentureBeat. "One of the best boards are still made by hand. You go to Apple, they've got the tools, and these guys are only pushing traces, checking every thing, doing flood fills—and also you're like, there's got to be a greater way."
The PCB design process typically unfolds in three stages. Engineers first create a schematic — a logical diagram showing how components connect. Then a specialist manually draws the physical layout in CAD software, placing components and routing hundreds of copper traces across multiple layers. Finally, the design goes to a manufacturer for fabrication.
That middle step — the layout — creates a persistent bottleneck. For a board of moderate complexity, the method typically consumes 4 to eight weeks. For classy systems like computers or automotive electronics, timelines stretch to a few months or longer.
"The timeline was at all times this elastic thing—they'd say, 'Yeah, that's two weeks minimum,'" Fadell recalled of his experience at Apple and Nest. "And we'd say, 'No, no. Work day and night. It's two weeks.' But it surely was at all times this fixed bottleneck."
The results ripple through hardware organizations. Firmware teams sit idle waiting for physical boards to check their code. Validation engineers cannot begin debugging. Product launches slip. In accordance with Quilter's research, only about 10 percent of first board revisions work appropriately, forcing expensive and time-consuming respins.
Project Speedrun put Quilter's AI to the test with an 843-component computer that booted on the primary try
Project Speedrun was designed to push the technology to its limits while producing an easily understood result: a working computer that would boot Linux, browse the web, and run applications.
The system consists of two boards based on NXP's i.MX 8M Mini reference platform, a processor architecture utilized in automotive infotainment, industrial automation, and machine vision applications.
The major system-on-module incorporates a quad-core ARM processor running at 1.8 gigahertz, 2 gigabytes of LPDDR4 memory, and 32 gigabytes of eMMC storage. A companion baseboard provides connectivity including Ethernet, USB, HDMI, and audio.
Together, the boards incorporate 843 components and 5,141 electrical connections, or "pins," routed across eight-layer circuit board stackups manufactured by Sierra Circuits in California. The minimum trace geometry reached 2 mils (two-thousandths of an inch) on the system-on-module — tremendous enough to require advanced high-density interconnect manufacturing techniques.
Quilter's AI accomplished the layout with roughly 98 percent routing coverage and 0 design rule violations. Each boards passed power-on testing and successfully booted Debian Linux on the primary attempt.
"We made a whole computer to reveal that this technology works," Nesterenko said. "We took something that's typically quoted at 400 to 450 hours, automated the overwhelming majority of it, and reduced it to about 30 to 40 hours of cleanup time."
The cleanup time is figure that human engineers still perform: reviewing the AI's output, fixing any issues, and preparing final fabrication files. But even with that overhead, the full elapsed time from schematic to fabricated boards collapsed from the standard 11 weeks to a single week.
Unlike ChatGPT, Quilter's AI learns by playing billions of games against the laws of physics
Quilter's technical approach differs fundamentally from the massive language models which have dominated recent AI headlines. Where systems like GPT-5 or Claude learn to predict text based on massive training datasets of human writing, Quilter's AI learns by playing what amounts to an elaborate game against the laws of physics.
"Language models don't apply to us because this just isn’t a language problem," Nesterenko explained. "In case you ask it to really create a blueprint, it has no training data for that. It has no context for that."
The corporate also rejected the seemingly obvious approach of coaching on examples of human-designed boards. Nesterenko cited three reasons: humans make frequent errors (explaining why most boards require revisions), the very best designs are locked inside large corporations unwilling to share proprietary data, and training on human examples would cap the AI's performance at human levels.
As a substitute, Quilter built what Nesterenko describes as a "game" where the AI agent makes sequential decisions — place this component here, route this trace there — and receives feedback based on whether the resulting design satisfies electromagnetic, thermal, and manufacturing constraints.
"What you're really changing just isn’t the probability of getting a really specific final result of the model, however the probability of selecting a certain motion based on that have," Nesterenko said.
The approach mirrors DeepMind's progression with its Go-playing systems. The unique AlphaGo learned from human games, but its successor AlphaZero learned purely through self-play and ultimately surpassed human capability. Quilter harbors similar ambitions.
"In the long run, to give you higher designs for circuit boards than humans have ever tried to do," Nesterenko said.
Fadell drew a parallel to an earlier technological transition: "I remember this with assembly. You had assembly and compilers, and engineers would say, 'I can't trust the compiler. I'm going to do the loop unrolling myself.' Now very, only a few people write any assembly."
He expects PCB design to follow an analogous arc: "I hope the identical thing happens with PCB design. Sure, a couple of people will hold out, but these tools are going to get so good that everybody else will move on."
Fadell and Nesterenko spent months solving a fragile problem: easy methods to automate design without stripping engineers of control
Automating a task that expert professionals have performed manually for many years raises an obvious query: how do engineers maintain control over designs that may ultimately ship in products where reliability matters?
Fadell said he spent significant time with Nesterenko working through this tension. The answer, he said, lies in allowing users to decide on their level of involvement at each stage of the method.
"In case you're a control freak, you may be a control freak. If you wish to say 'just do it for me,' you may do this too—and every thing in between," Fadell said. "You’ll be able to walk through each phase of the design and become involved wherever you wish, or let the AI handle it."
The workflow breaks into three phases: setup, where engineers define constraints and requirements; execution, where the AI generates candidate layouts; and cleanup, where humans review and refine the output. Engineers can intervene at any point, adjusting constraints and regenerating designs until they're satisfied.
"That is something Tony and I speak about rather a lot," Nesterenko said. "How can we give users control while still automating a lot of the work?"
Quilter's technology has clear boundaries: 10,000 pins and 10 gigahertz mark the present limits
The technology has clear limitations. Quilter currently handles boards with as much as roughly 10,000 pins — sufficient for a big selection of applications but well wanting essentially the most complex designs, which may exceed 100,000 connections.
Physics complexity also creates boundaries. The system handles high-speed communications as much as roughly 10 gigahertz, covering typical consumer electronics and lots of industrial applications. But advanced systems like sophisticated radar, which may operate at 100 gigahertz, exceed current capabilities.
"There are boards where Quilter won't make enough progress to make the cleanup time worthwhile," Nesterenko acknowledged. "We're just not that helpful yet with essentially the most advanced, sophisticated designs."
The corporate has focused initially on categories where speed matters greater than extreme complexity: test fixtures, evaluation boards, design validation boards, and environmental test hardware. These boards often sit in long queues behind higher-priority production designs, delaying engineering programs.
The corporate bets that engineers pays the identical price for a 10x speed improvement
Quilter prices its service by pin count, matching the billing conventions that exist already when corporations hire outside layout specialists. The pitch to customers is cost neutrality with a ten-fold improvement in speed.
"We're going to charge you roughly the identical that you simply would pay for the pins that you simply would with an individual," Nesterenko said. "But the rationale you select us is that we do that 10 times faster."
For an organization waiting three months for a board layout, receiving it in per week fundamentally changes what's possible. Engineering teams can run multiple design experiments in parallel. Firmware developers get hardware faster. Products reach the market sooner.
The corporate offers free access for hobbyists, students, and small businesses with lower than $50,000 in revenue — a technique to construct familiarity while targeting enterprise customers for business revenue.
The iPod creator waited years to connect his name to Quilter — until he could prove the technology actually works
Fadell said he selected this moment to publicly acknowledge his investment since the Project Speedrun demonstration provides concrete evidence that the technology works.
"It's not about being comfortable—I used to be at all times comfortable with the team," he said. "This was about waiting until we had something you could possibly hang your hat on. Now I can say, 'I've used the tool. I've seen it.'"
He contrasted his approach with typical investor announcements: "Every investor goes, I invested on this, it's gonna change the world. It's like, no, I do know higher. I've used the tool. I do know individuals who use it. I asked my startups to make use of the tool."
Fadell's involvement goes beyond capital. He described email exchanges running to "a dozen pages of details" covering product design, user experience, enterprise sales, and technical architecture.
"Of all of the investors I work with, Tony by far goes deepest with me on the product side," Nesterenko said.
If Quilter succeeds, it could unlock a brand new generation of hardware startups that were never economically viable before
The stakes extend far beyond one company's product roadmap. If Quilter's technology scales, it could fundamentally alter the economics of constructing physical products.
Fadell argued that hardware development has historically moved slowly because each step in the method — schematic design, PCB layout, manufacturing, assembly — created friction. Other innovations have already smoothed schematic tools and manufacturing. Layout remained the stubborn holdout.
"When you shrink that from weeks to hours, you may iterate a lot faster because all the opposite friction within the chain has been reduced," Fadell said.
He predicted the technology would eventually extend upstream into schematic design itself, with AI that understands each logical connections and physical constraints helping engineers avoid problems earlier in the method.
At MIT, where Fadell now spends time, he encounters would-be founders who’ve abandoned hardware ambitions because the method seemed insurmountable.
"I talk over with professors and startup founders, they usually say, 'I'm never doing hardware. It's too hard,'" he said. "I hope we are able to make it easier for more people to leap in and take a look at things."
Industry veterans remain skeptical. Auto-routing tools — previous attempts at automation — became notorious for producing unusable results, spawning T-shirts proclaiming engineers would "never trust the auto-router."
Nesterenko has seen the skepticism dissolve in real time. He described a recent meeting with executives from a serious customer who got here to debate Quilter's capabilities. Because the conversation unfolded, one executive picked up the Project Speedrun boards and started photographing them from every angle, turning them over in his hands.
"He was just fascinated by the indisputable fact that this is feasible now," Nesterenko said.
The query isn’t any longer whether AI can design circuit boards. A working Linux computer, assembled from 843 components and booted on the primary attempt, answers that definitively. The query now’s what engineers will construct when layout stops being the bottleneck — when hardware, as Fadell put it, finally "moves on the speed of thought."
On that time, Nesterenko offered a prediction. "In case you ask the common electrical engineer today whether automation or AI could in any respect help with the board of this complexity, they’d say no," he said. For many years, they’d have been right. As of last week, they're not.
