AI Chip Design Enters a Recursive Phase

On September 25, 2026, TechCrunch Disrupt 2026 confirmed that Recursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini will present “When AI Starts Designing Its Own Hardware” at the event taking place October 13-15 in San Francisco. Attendees can save up to $200 by purchasing tickets before midnight PT on September 26, and a second pass is available at 50% off for select ticket types; online live streaming has not been announced.

• Session topic: “When AI Starts Designing Its Own Hardware”
• Date & venue: October 13-15, 2026, Moscone West, San Francisco
• Ticket deal: $200 off for single pass until Sept 26; 50% off for second pass (select types)

From AlphaChip to a $4 Billion Startup

Goldie and Mirhoseini previously co-led AlphaChip at Google—an AI system capable of generating chip layouts in hours rather than years—which has already supported multiple generations of Google’s Tensor Processing Units. Their work reduced traditional 2–3 year chip design cycles to hours, demonstrating AI’s potential to transform hardware development.

Founded in late 2025, Recursive Intelligence aims to extend this concept: building AI that learns across chip designs, allowing experience from one project to directly improve subsequent iterations. Goldie and Mirhoseini seek to compress the full development cycle to weeks.

A key irony: while AI models evolve monthly, hardware development persists on yearly cadence. Recursive’s mission is to close this gap—when AI doesn’t just rely on human-designed chips, but helps design them, the entire innovation rhythm shifts.

Founding Team Credentials

• Anna Goldie (CEO): PhD in CS, Stanford; MIT Tech Review “35 Innovators Under 35”; ex-early employee at Anthropic and senior staff research scientist at Google DeepMind
• Azalia Mirhoseini (CTO): Assistant Professor of CS, Stanford; founder of Scaling Intelligence Lab

The pair co-founded Google’s ML for Systems team. Recursive raised $335 million in four months, hitting a $4 billion valuation, with Nvidia among investors.

Automating the Full Chip Design Pipeline

Recursive’s approach spans the entire pipeline: placement, optimization, timing closure, and verification. Its distinguishing feature is cross-chip learning—AI acumulates patterns from one design to guide future ones, creating compounding efficiency.

Modern chip design involves arranging billions of transistors, where minor errors cause costly failures. Silicon tape-out can cost tens of millions of dollars, making human experience crucial but slow. Recursive’s AI aims to internalize yield and performance patterns, reducing costly re-spins.

At Disrupt, the founders will detail how this loop unlocks novel architectures. When hardware design is no longer bounded by human cognitive cycles, approaches like Chiplet, 3D stacking, and compute-in-memory may mature faster.

Practical Entry Points for Stakeholders

• Immediate interest: AI-chip startups, hardware-acceleration vendors, and cloud-provider technology leaders tracing compute supply chains;
• Wait-and-see: Companies relying on off-the-shelf AI models, unless custom silicon is essential to your use case; Recursive’s tools remain in development.

A Final Note

Recursive Intelligence’s milestone isn’t about replacing engineers—it’s about amplifying human capability at scale. When hardware development becomes computational, learnable, and reusable, the bottleneck shifts from execution to vision: the true essence of recursion in the AI era.