Designing a chip is hard. It takes years, hundreds of millions of dollars, and a tiny pool of experts. Most of them sit inside a few large companies.
Architect Labs wants to change that with AI. The Palo Alto startup left stealth on Thursday with a $24mn seed round. Its plan is to build an AI system that designs and verifies custom chips from end to end.
Kindred Ventures led the round. TQ Ventures, Race Capital and Together Fund joined in. So did a notable group of angels. Per Reuters, they include Google DeepMind chief scientist Jeff Dean, plus executives from OpenAI and Nvidia. Kindred founder Steve Jang takes a board seat.
A ‘designless’ chip industry #
The pitch leans on semiconductor history. Two decades ago, the fabless model split design from manufacturing. TSMC then made world-class fabrication available to anyone with a design.
Architect Labs wants to repeat that trick one layer up. This time, the target is design itself. It calls the result a “designless” semiconductor industry. In that world, a company no longer has to become a chip company. It does not have to bet a decade on one architecture. It simply brings a workload and gets the silicon to run it.
Taking aim at Broadcom and Marvell #
That goal collides with two of the industry’s richest franchises. Reuters reports that Architect Labs is chasing the custom-chip business of Broadcom and Marvell. Both firms design bespoke AI accelerators for cloud giants such as Amazon and Google. Together they earn tens of billions of dollars a year.
Demand for custom silicon keeps climbing. AI labs, hyperscalers and robotics makers all want chips tuned to their own workloads. Off-the-shelf hardware can no longer keep up. The squeeze is also fuelling a wider hunt for new chip architectures.
“AI models have advanced dramatically across nearly every field, yet chip development cycles remain equally slow and painful,” said co-founder Ebrahim Hussain. His fix is not to bolt AI agents onto a decades-old design “flow”. Instead, he wants to rebuild the process from scratch, with AI as a “first-class actor”.
Founders who dropped out to build it #
The founders are betting their CVs on the idea. Hussain skipped high school and started college at 15, the company says. He then worked on custom chips at Apple and Tesla. His co-founder, Aaditya Subedi, researched AI code verification at Harvard. The two met at Stanford, then dropped out to launch the company.
Their team runs to about 18 people, split between machine learning and hardware. Together, the company says, they have taped out more than 80 production chips. The group also includes alumni of Intel, Meta’s custom-silicon work, and machine-learning teams at Anthropic, DeepMind and xAI.
What comes next #
Architect Labs says it has already deployed with semiconductor partners. It expects AI-generated designs to tape out on leading-edge nodes later this year. No one has independently verified either claim yet.
The new money will scale its compute, deepen its research and fund co-design with early partners. The bigger vision is a tighter loop. In it, models, software and hardware improve together. If the bet pays off, hardware stops being the thing AI has to work around.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.