Partnering with Ricursive Intelligence: A Premier Frontier Lab Pioneering AI for Chip Design
Sequoia by Stephanie Zhan
Sequoia is backing Ricursive Intelligence, a new lab focused on AI that designs chips. The bet is that faster chip design could unlock a lot more compute, and a lot less waste.
Based on reporting by Sequoia, by Stephanie Zhan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sequoia says it is leading Ricursive Intelligence’s first round, backing a company built around one very specific bet: AI should help design AI chips. The founders, Anna Goldie and Azalia Mirhoseini, are the people behind AlphaChip, the effort that showed floorplanning could move from months to hours. Now they want to push that idea much further.
The pitch starts with a bottleneck everyone in the industry already feels. Chip design is slow, taking 12 to 24 months at mature nodes and 18 to 36 months at the leading edge for 5nm or 3nm. It is also expensive, with average design costs of $200 million to $250 million for 7nm, $450 million to $500 million for 5nm, and $600 million to $650 million for 3nm. Sequoia points out that 50% to 70% of that cost is human labor, while another 5% to 15% goes to electronic design automation tools.
That matters because the companies selling the tools are already huge. The article names Cadence and Synopsys as the long-dominant players in EDA, each with $5 billion to $6 billion in annual revenue and roughly $90 billion to $100 billion in market value. But the more interesting part is not the size of the old market. It is the suggestion that the whole flow — architecture, RTL, verification, and physical design — could eventually be automated.
Sequoia also frames the problem in plain hardware terms: compute is the scarce resource, and chip design is the thing slowing it down. The article cites reports from August 2024 that a multi-month Blackwell delay could mean more than $10 billion in lost 2025 revenue. Faster design cycles, in that view, are not just an efficiency win. They can change what gets built, how quickly it ships, and how many teams can afford to try.
Ricursive’s bigger ambition is to move the industry from “fabless” to “designless.” Fabless companies already outsource manufacturing; Ricursive wants to take on the design work too, turning an idea into a manufacturable chip. Sequoia says the company has already assembled a highly dense team in its first weeks, and it is betting that this category gets defined now, not later.
My take — AI-written commentary, not fact-checked reporting
This is the kind of bet the AI industry should be making: not another chatbot wrapper, but the machinery underneath the whole stack. Chip design is where the money and the bottleneck actually live, which is exactly why it’s a better target than the usual demo-friendly nonsense. The open question is whether the field wants real automation, or just the comforting illusion of it wrapped in a pitch deck.
Read more about this at: Sequoia