Ricursive Intelligence bets on AI that designs the chips behind AI

Ricursive Intelligence is building software that uses artificial intelligence to help design semiconductors, a move that could shorten chip development from years to weeks and reshape how future AI hardware is built. The startup’s co-founders, Anna Goldie and Azalia Mirhoseini, will explain the approach at TechCrunch Disrupt 2026 as investor interest in AI-driven chip design accelerates.

The company’s premise is simple but consequential: if AI can help create better chips, those chips can in turn support even more capable AI systems, creating a feedback loop that may speed up the entire industry.

Why AI chip design is becoming a bottleneck

AI models have been improving quickly, but the hardware needed to train and run them still follows a slow, highly specialized development cycle. Semiconductor design is expensive, technically demanding and time-consuming, often requiring multiple years from concept to production. As AI developers push for larger models and lower inference costs, chip architecture has become one of the most important constraints in the field.

Ricursive Intelligence is trying to attack that constraint directly. Rather than relying only on teams of human engineers to place components, optimize layouts and verify designs, the startup wants AI systems to take on more of that work and improve with every project they complete.

The company says the long-term goal is to compress the chip-design timeline dramatically, from the current two-to-three-year cadence to a matter of weeks. If it succeeds, the impact could extend beyond faster product cycles. It could also create new possibilities for custom chips, higher efficiency and more rapid experimentation with AI hardware.

What Ricursive Intelligence is building

Ricursive Intelligence is developing AI tools aimed at automating major parts of chip design, including component placement and design verification. The system is designed to learn from each chip it works on so that knowledge gained on one architecture can carry forward to the next. In effect, the software is meant to become better at chip design by repeatedly doing chip design.

That learning loop is the core of the startup’s strategy. Traditional design workflows can be highly repetitive and hard to scale. Ricursive argues that a machine learning system can absorb patterns across multiple generations of chips and apply them faster than a human-only process, potentially improving both speed and performance.

The company’s founders are not starting from scratch. Before launching Ricursive, Goldie and Mirhoseini helped lead AlphaChip at Google, the AI system that demonstrated layout generation in hours rather than the much longer timelines associated with conventional methods. Their work contributed to multiple generations of Google’s Tensor Processing Units, giving them a track record in applied AI-for-hardware research.

Goldie and Mirhoseini’s earlier work showed that AI could meaningfully help design real chips, and Ricursive is now trying to turn that research into a broader platform for hardware development.

How did the founders get here?

Goldie and Mirhoseini have spent much of their careers at the intersection of AI research and systems design. They co-led Google’s ML for Systems team and each held roles at major AI labs before founding the new company. Their combined background gives Ricursive unusual credibility in a field where technical proof matters as much as ambition.

Goldie is Ricursive’s founder and chief executive. She holds a Ph.D. in computer science from Stanford University and was previously recognized by MIT Technology Review as one of its Innovators Under 35. Mirhoseini, Ricursive’s founder and chief technology officer, is an assistant professor of computer science at Stanford and founder of the university’s Scaling Intelligence Lab.

Both founders also spent time as early employees at Anthropic and later as senior staff research scientists at Google DeepMind, experiences that place them squarely inside the modern AI elite. That background likely helped Ricursive move quickly from concept to funding.

Key milestones in Ricursive Intelligence’s rise

The company’s progress has been unusually fast, even by startup standards. Ricursive was launched in late 2025 and raised a large round within months of coming out of stealth. Investors have been willing to back the thesis that AI can help solve the bottleneck of chip design, especially as demand for advanced accelerators continues to surge.

The funding total stands at $335 million, including a $300 million Series A, at a reported $4 billion valuation. Nvidia is among the investors, a notable endorsement given the chipmaker’s central role in the AI boom and its direct interest in advances that could improve future hardware design.

Why investors are paying attention

Chip design has become strategically important because every major AI breakthrough now depends on faster, more efficient compute. Training frontier models requires enormous processing power, while deployment at scale demands chips that can deliver performance without driving energy and infrastructure costs too high.

That makes the ability to design better hardware faster more than a niche engineering challenge. It is increasingly a competitive advantage. Companies that can iterate on chip architecture quickly may be better positioned to build models more efficiently, cut operating expenses and customize hardware for specific workloads.

Ricursive’s thesis is that AI can compress the traditional R&D cycle enough to make chip innovation more dynamic. Instead of a drawn-out process where designs remain static for years, the company imagines a continuous loop in which software helps invent the next generation of the tools it runs on.

• AI helps automate chip layout and verification.
• Each design feeds new data back into the system.
• The system improves as it encounters more architectures.
• Better chips can then support larger, faster and cheaper AI models.

What happens when AI designs its own hardware?

That question sits at the center of Ricursive’s pitch. The company is not claiming that AI will replace every chip engineer. Instead, it is betting that machine learning can eliminate the slowest and most repetitive parts of the process while giving human designers better tools and shorter iteration cycles.

In the best case, the result would be a new development model for semiconductors. Chip teams could test more ideas, iterate more quickly and uncover architectures that might be difficult to find manually. That matters because small improvements in chip efficiency can have outsized effects across AI infrastructure, from cloud costs to data-center energy use.

There is also a strategic dimension. If AI systems themselves are helping define the next wave of hardware, the industry could become more tightly coupled. Progress in software and hardware would no longer move on separate clocks. Instead, they could reinforce one another in near real time.

How fast could the loop move?

Ricursive says the goal is to reduce chip development from years to weeks, but that remains an ambitious target. Semiconductor workflows involve multiple layers of validation, fabrication constraints, testing and reliability checks. Any AI system that participates in that process has to work within strict engineering requirements, not just generate promising ideas.

Still, even partial gains could be meaningful. If AI can cut weeks or months out of design stages, the cumulative effect across several generations of chips could be substantial. Faster turnarounds may also help startup customers and large labs respond more quickly to shifts in model size, performance targets and manufacturing constraints.

That is why Ricursive’s pitch resonates beyond chip enthusiasts. It speaks to a broader race in AI: the fight to remove bottlenecks that slow progress in training, deployment and infrastructure.

What Goldie and Mirhoseini will discuss at Disrupt 2026

At TechCrunch Disrupt 2026, the two founders are scheduled for a session titled “When AI Starts Designing Its Own Hardware.” Their appearance will give attendees a closer look at how Ricursive thinks about self-improving systems, hardware acceleration and the future of AI infrastructure.

The session is expected to focus on the practical side of the story as much as the vision. That includes the challenge of closing the loop between AI and chip development, the engineering limits of automation and the reasons investors believe this area could be one of the next major frontiers in AI.

Disrupt itself has become a key venue for startup founders, engineers and investors looking to spot the next breakout category early. This year’s event, set for October 13 to 15 at Moscone West in San Francisco, will feature more than 200 sessions across six industry stages, along with roundtables and breakouts. Organizers say they expect more than 10,000 attendees, 250-plus speakers and over 300 exhibiting startups.

For Ricursive, the appearance is both a technical forum and a signal of momentum. The company is emerging at a moment when AI leaders are increasingly focused not only on model quality, but on the hardware pipeline that determines how far those models can scale.

Why this startup matters for the next phase of AI

Ricursive Intelligence is important because it targets one of the least glamorous but most consequential parts of the AI stack. The field has spent years obsessing over model parameters, training data and product interfaces. But the next wave of progress may depend just as much on better silicon and faster design loops.

If the company’s approach proves effective, it could alter how semiconductor teams work and how AI infrastructure evolves. It might also create a template for using AI to optimize other complex engineering processes where iteration speed is a competitive edge.

For now, Ricursive sits at an intersection of research, hardware and startup ambition. Its founders are known for helping show that AI can contribute to chip design; now they are trying to prove that the same idea can become a scalable business.

That is what makes the company worth watching. In a sector where every advance eventually runs into hardware limits, a startup promising to make the hardware faster to design is aiming at one of the industry’s deepest chokepoints.

And if the thesis holds, the next leap in AI may begin not with a new model, but with the chip that helps build it.