Ricursive Intelligence raises $335M to put AI to work designing chips

Ricursive Intelligence is an AI chip-design company founded by Anna Goldie and Azalia Mirhoseini, who previously helped create Google’s AlphaChip. On January 26, 2026, Ricursive announced a $300 million Series A at a $4 billion valuation; TechCrunch later reported that the company had raised $335 million in total, including its seed round. Its ambition is to build AI into the semiconductor design process so that each design can inform the next.

Key points

• Lightspeed led the $300 million Series A; NVIDIA’s venture arm NVentures and Sequoia were among the investors.
• The $335 million total combines a $35 million seed round and the $300 million Series A. It was not the size of the Series A alone.
• Goldie told TechCrunch that chip design can take six months to two years and said Ricursive aims to do a design in about a week. That is a founder’s claim, not a published independent benchmark.

The funding story

The two headline figures describe different things. TechCrunch’s February 2026 profile traced a $35 million seed round led by Sequoia and a $300 million Series A led by Lightspeed, announced in January. Together they make $335 million; the company was valued at $4 billion. Ricursive’s January announcement said the financing would support growth in its research and engineering team and computing infrastructure. The funding establishes that the company has substantial early backing. It does not establish that its platform is already deployed at scale.

The founders’ record is central to the investment case. Goldie and Mirhoseini worked together on AlphaChip at Google. Lightspeed’s investment thesis says the system was used across multiple generations of Google’s TPU and adopted by MediaTek. TechCrunch’s founder profile also describes AlphaChip as their earlier machine-learning effort for chip layout. That history is relevant evidence of experience, but it is separate from proof that Ricursive’s current platform can deliver equivalent results. The distinction matters: AlphaChip is a prior system; Ricursive is developing a broader platform. Our coverage of OpenAI and Synopsys’s chip-design partnership covers another route into AI-assisted chip design, centered on a model and EDA-tool collaboration.

How the system works

Ricursive’s stated goal is a feedback loop between AI models and the chips that run them. Its January announcement frames the work as AI systems helping design better hardware, with that hardware in turn supporting more capable computing. TechCrunch’s February report says the platform combines reinforcement learning and large language models, spanning component placement through design verification. The founders say it should learn across chip designs, so work on one project can inform another.

A later TechCrunch interview added detail on how the founders see the design process changing. Goldie described conventional workflows as splitting a chip into many smaller blocks, while Ricursive wants its AI to optimize the whole chip and explore less conventional layouts. She cited an earlier AlphaChip result that arranged components in a donut-like pattern as an example of an unusual layout that could meet design goals. That example belongs to AlphaChip, not a disclosed test of Ricursive’s new platform. The available reports do not publish a new-platform benchmark, named customer list, or chip that has entered production, so the technique cannot yet be compared on measured performance.

The speed claim

Shorter design cycles are a major part of the pitch. In October, Goldie told TechCrunch that conventional chip design can take six months to two years, while Ricursive can design a chip in about a week. Lightspeed’s investment article uses a different frame: leading chips can take two to three years and hundreds of millions of dollars to design, and the firm’s stated goal is to compress that work from years to weeks. Those estimates are not a controlled comparison of the same chip, team, and workflow. Goldie’s one-week figure is a company claim, while Lightspeed’s account describes the broader challenge and target.

If AI can automate repeated placement and verification work, chip teams might be able to test more workload-specific designs. That could make custom silicon more accessible to companies without a large in-house design group, which is part of Lightspeed’s thesis. But the articles provide no project-level cost savings, reproducible results, or public product interface to test. The one-week figure should be treated as an ambition or reported capability, not a delivery guarantee for any arbitrary chip.

A supply-side bet

Goldie told TechCrunch at Disrupt that she expects compute demand to grow by at least 10,000 times over the next five years, adding that the estimate might be conservative. This is her forecast, not a consensus industry projection. Her argument is that the answer cannot be only to build more data centers; the industry also needs to use bandwidth, wafers, and fab capacity more efficiently. AI-assisted chip design is one proposed way to improve that supply side.

That places Ricursive in a different part of the compute conversation from companies that train large models or sell accelerators. The product it aims to sell is design software for semiconductor and electronics makers. NVIDIA’s NVentures participated in the financing, but that does not establish NVIDIA as a customer. The distinction is useful when reading the investment story: strategic participation signals interest, not a commercial deployment. We have also covered a leadership change in OpenAI’s compute organization, a separate development on the infrastructure side of the same supply constraint.

What builders should watch

For people working in chip design, EDA, or AI infrastructure, Ricursive is worth tracking because it targets the tooling bottleneck instead of launching another accelerator. Its founders have relevant AlphaChip experience, and the financing brings backing from Lightspeed, Sequoia, and NVentures. Those factors may help the company attract talent and industry partners; they do not substitute for evidence from the new system.

There is no public download, API, or developer trial in the material reviewed, so there is no one-minute setup to recommend. The practical first step is to read the company’s financing announcement and the founder reporting to understand the product scope and the team’s track record. For a technical evaluation, look for customer collaborations, reproducible design outcomes, which parts of the design flow the tool covers, and the class of chips behind the one-week claim.

The honest caveat is that layout speed alone does not tell a buyer whether the finished chip is faster, more power-efficient, cheaper to manufacture, or easier to validate. Manufacturing process constraints, verification requirements, and the target workload still matter. Until Ricursive publishes results on actual designs, the strongest conclusion is that experienced AlphaChip researchers have raised significant capital to pursue a credible but unproven approach to the compute crunch.