Ricursive Intelligence founders will discuss AI-designed chips at Disrupt 2026

Why it matters

Ricursive is applying proven AI research in chip layout to a much broader ambition: software that improves across chip-design cycles. Whether it can make that leap will matter to companies seeking faster, specialized silicon for AI.

Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini are scheduled to discuss AI-assisted chip design at TechCrunch Disrupt 2026, taking place October 13th-15th in San Francisco. Their session, announced by TechCrunch on September 25th, centers on a bet the pair have pursued from research into company-building: AI can help design the chips that, in turn, power more capable AI.

For Goldie, the bet grows out of work that already crossed the line from research paper to industrial use. She co-led AlphaChip, a reinforcement-learning system for chip floorplanning, after earning a computer science PhD at Stanford and degrees in computer science, linguistics and electrical engineering and computer science at MIT. She later co-founded Google Brain's ML for Systems team, worked at Anthropic, and became a senior staff research scientist at Google DeepMind, according to her biography.

Mirhoseini, Ricursive's co-founder and CTO, is an assistant professor of computer science at Stanford, where she directs the Scaling Intelligence Lab. She and Goldie developed AlphaChip together at Google Brain. Their partnership gives Ricursive an unusual starting point: its founders have experience both publishing the research and working inside organizations building frontier AI models and custom chips.

From chip placement to a broader design platform

AlphaChip addressed a specific, costly part of chip development: deciding where components should sit on a chip. The research paper in Nature described a machine-learning approach to chip floorplanning. Ricursive's founders have said AlphaChip helped design multiple generations of Google's Tensor Processing Units; Goldie's site says the work was used across four TPU generations and other chip designs.

Ricursive wants to extend that approach across more of the design process. The startup describes its goal as a system that designs chips, verifies them and improves from one design cycle to the next. The proposed loop is straightforward to explain and difficult to deliver: an AI system helps create better hardware, that hardware supports more capable models, and those models help develop subsequent chips.

In its September 25th announcement, TechCrunch said designing a chip can take two to three years and described reducing that cycle to weeks as Ricursive's goal. That is an ambition, not a reported production result. AlphaChip's experience in component placement provides a meaningful foundation; extending an AI system across a chip-design flow is a larger engineering and validation challenge. Chip designs still have to satisfy power, performance and manufacturing constraints, and customers have to trust the result before committing expensive fabrication runs.

That gap between demonstrated research and the company's broader product is the consequential subject behind the conference session. The founders will be discussing what it takes to turn a successful design method into a system that learns across different chips and can handle more of the work. The conference announcement names the theme, but does not describe a new product release or technical milestone.

Investors are funding the team and the cycle time

The capital behind the thesis arrived well before this speaker announcement. Ricursive launched in late 2025 and raised a $35 million seed round led by Sequoia Capital. In January 2026, it announced a $300 million Series A led by Lightspeed Venture Partners at a reported $4 billion post-money valuation. TechCrunch reported in February that the two rounds brought the startup's total funding to $335 million. Nvidia's venture arm, NVentures, was among the Series A participants.

That backing reflects investor confidence in Goldie and Mirhoseini's research record as well as in the market for custom silicon. Ricursive sells design software, not finished chips, so its prospective customers include the chipmakers and technology companies that need specialized processors. The strategy depends on convincing those buyers that AI can shorten design work without compromising the reliability that chip production demands.

Goldie has described chip performance as a way to advance AI; Mirhoseini has framed the long design cycle as a brake on how quickly models and their hardware can improve together. The premise gives Ricursive a clear reason to exist beyond another AI-tooling pitch. Its founders are trying to make the infrastructure underlying AI development improve at a pace closer to the models themselves.

Their prior work gives that argument substance, but the next stage asks for a different kind of proof. AlphaChip demonstrated AI's role in a defined chip-layout task. Ricursive's larger proposition is that learning can compound across designs and reach further into the engineering process. The company has raised enough to pursue that expansion; the technical challenge is showing that each cycle produces improvements chip customers can use and trust.

Goldie and Mirhoseini are due on the Disrupt Stage during a conference running October 13th-15th at Moscone West in San Francisco. Their conversation is a chance to explain how far the AlphaChip approach can stretch, and where the boundary sits between AI-assisted chip design and a system that can meaningfully design the hardware used to build its successors.