Opportunity assessment: high. AI models improve every few months, but the chips they run on still take years to build. Two researchers who already showed that AI can design production silicon are now building a company to close that gap.
According to TechCrunch AI, Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini will speak on the Disrupt Stage at TechCrunch Disrupt 2026, which runs October 13-15 at Moscone West in San Francisco. Their session is called “When AI Starts Designing Its Own Hardware.” The goal behind it is simple to state. Designing a chip takes two to three years today, and Ricursive wants to get that down to weeks.
🎯 Situation Report
Ricursive is building AI that designs chips, learns from each design and applies what it learned to the next one. The company wants to automate the whole pipeline, from component placement through design verification.
What stands out is the learning across designs. Most chip-design tools work on one project at a time. Ricursive’s system is meant to carry what it learned from one chip over to the next. That sets up the loop the founders are betting on:
- AI designs better hardware.
- Better hardware runs more powerful AI.
- More powerful AI designs the next generation of hardware.
That’s the “recursive” in the name. It’s also why the talk leans so hard on self-improving systems.
🧭 The Operators
They’re not newcomers pitching a theory. Before Ricursive, Goldie and Mirhoseini co-led AlphaChip at Google. That AI system generated chip layouts in hours, a job that takes human designers far longer. Their work helped design several generations of Google’s Tensor Processing Units, the custom chips behind much of Google’s AI work.
The rest of their record, per TechCrunch AI:
- They co-founded Google’s ML for Systems team.
- Both were early employees at Anthropic.
- Both were senior staff research scientists at Google DeepMind.
- Goldie (CEO) holds a Stanford PhD in computer science and made MIT Technology Review’s 35 Innovators Under 35 list.
- Mirhoseini (CTO) is an assistant professor of computer science at Stanford and founded its Scaling Intelligence Lab.
AlphaChip also had critics. After the 2021 Nature paper, some chip-design researchers publicly challenged its results, and Google defended the work. That history matters. Ricursive will have to show real-world results, not just promise them.
💰 The Money
Investors moved fast. Ricursive launched in late 2025. Within four months it had raised $335 million at a $4 billion valuation, including a $300 million Series A.
Nvidia is one of the investors. That’s the key signal here. The company that dominates AI hardware is backing a startup that could speed up how everyone designs chips, possibly including its competitors.
⚡ Why It Matters
- Hardware sets the pace. Models improve on a cycle of months. Chips improve on a cycle of years. If Ricursive shrinks that gap even partway, hardware stops being the slowest part of AI progress.
- Custom silicon keeps spreading. Hyperscalers and AI labs all want their own chips. Faster design means more teams can afford to try.
- New architectures become practical. When a design costs weeks instead of years, teams can take more risks on unusual chip designs. The founders say this could lead to more capable and more efficient AI.
- Incumbents get pressure. Chip-design software giants Synopsys and Cadence already sell AI-assisted tools. A well-funded startup aiming at full automation raises the bar for them.
🔭 What to Watch
- Proof on real silicon. Weeks-long design cycles are a bold claim. Watch for chips that actually tape out (get finalized for manufacturing) using Ricursive’s tools.
- Verification. Placement is where AlphaChip made its name. Verification is where bugs get caught, and where they get expensive. Automating it would be a much bigger result.
- Customer names. Nvidia’s investment doesn’t make it a customer. The first public design partners will show who trusts this approach enough to use it.
- Disrupt details. The October session is the founders’ chance to put numbers behind the loop they’re describing.
📌 Assessment
The AI industry has spent the past two years treating compute as its main constraint. Most of the response has been more money, more data centers and more power. Ricursive is going after a different lever: how fast the hardware itself gets made. If AI can speed up the design of its own chips, progress could compound faster than current roadmaps assume.
Goldie and Mirhoseini speak at Disrupt in mid-October. TechCrunch AI’s original piece has more on the session.