led by Anna Goldie
AI lab from AlphaChip's creators, maker of Ripple open models
Worth following if you track open models: Ricursive's coding model Ripple runs locally through Ollama and is free to use commercially. Behind it is a chip-design lab with a pedigree from Google and Anthropic that has not named its early customers, so treat its benchmark claims as self-reported.
| Model | Lab | Context | In / out per 1M | Reads and writes |
|---|---|---|---|---|
| Ricursive Intelligence | 1M | $0.10 / $0.30 | text → text, code |
| Product | What it is |
|---|---|
| Ripple-65B Preview | Open weight language model for agentic coding |
| AI chip design platform | Proprietary B2B chip design platform |
Every model Hokai holds for this company that carries a release date, newest first.
Its latest round is a Series A of $300 million, announced on 26 January 2026 and led by Lightspeed Venture Partners, which valued the company at $4 billion. Before it came a $35 million seed round in December 2025 with Sequoia Capital as lead, at a $750 million valuation. TechCrunch cites $335 million in total; some outlets later reported a $500 million Series A, which the January coverage does not support.
Ricursive Intelligence, Inc. is private, with no stock ticker. Anna Goldie is the CEO and Azalia Mirhoseini the CTO. Crunchbase News names Sequoia, Lightspeed, DST Global, NVentures, Felicis, 49 Palms and Radical Ventures among its investors.
Ripple-65B Preview is an open-weight language model for agentic coding, trained from scratch on 12 trillion tokens and licensed under Apache 2.0. It is a general-purpose version of the proprietary model behind Ricursive's chip platform. Ricursive reports 80.5 on Terminal-Bench 2.1; it runs locally with the command ollama run ripple-preview.
It uses AI to design chips, handling component placement through design verification, and is aimed at companies that need custom or standard chips rather than at competing with them. The founders told TechCrunch it should learn across chips, so each design improves the next. TechCrunch notes that the early customers are not named.
Anna Goldie and Azalia Mirhoseini founded it. They met at Stanford, worked together at Google Brain, were early Anthropic employees, and created AlphaChip, the AI layout tool used on Google's TPU chips. Both were named to the TIME100 AI 2026 list, according to the company's site.
Ricursive Intelligence, founded in 2025 and based in Palo Alto, is an AI lab led by Anna Goldie and Azalia Mirhoseini, the creators of AlphaChip. It builds AI tools that design chips and released Ripple-65B Preview, a 65-billion-parameter open-weight model made for agentic coding under an Apache 2.0 licence.
Build self-improving AI systems, starting with chip design, to close the loop between AI and the hardware that fuels it.
Ricursive Intelligence is a Palo Alto AI lab that describes itself as a frontier lab building self-improving systems, starting with chip design. It was founded in 2025 by Anna Goldie (chief executive) and Azalia Mirhoseini (chief technology officer), who created AlphaChip at Google, the AI tool that generates chip layouts, and were early employees at Anthropic. TechCrunch reports that AlphaChip helped design three generations of Google's TPUs, and the company's own site lists AlphaChip (Nature, 2021) among the team's papers.
Its main product is a platform that uses AI to design chips, from component placement through design verification, and it is not building chips itself. TechCrunch's February 2026 report says the platform targets chip makers, that its early customers are unnamed, and that NVIDIA is both an investor and a potential customer. The platform is proprietary, with an internal language model behind it.
On the model side, Ricursive released Ripple-65B Preview, whose Hugging Face repository was created on 4 November 2026. In its launch post the company describes it as a 65-billion-parameter dense model pretrained from scratch on 12 trillion tokens, with up to 1 million tokens of context, released under Apache 2.0. Ricursive reports a score of 80.5 on Terminal-Bench 2.1, a benchmark of terminal-based coding agents, and puts its full research and development cost at $32 million, both its own figures. The weights are available on Hugging Face, Together AI, OpenRouter, Fireworks and GMI Cloud, and in the Ollama library.
Funding: Ricursive raised a $35 million seed round led by Sequoia Capital in early December 2025, then announced a $300 million Series A on 26 January 2026, led by Lightspeed Venture Partners at a $4 billion valuation. Crunchbase News lists DST Global, NVentures (NVIDIA's venture arm), Felicis, 49 Palms and Radical Ventures among the Series A participants, and TechCrunch cites a $335 million total.
The team is small and hiring: its careers page lists on-site roles in Palo Alto, and the company was named to the TIME100 AI 2026 list for its two founders. Its post thanks open-source work from the Allen Institute for AI (Ai2), whose Olmo model ladder it used as a starting point, and from Prime Intellect, RadixArk and NVIDIA.
Where it is weaker: it is a company about a year old whose chip-design customers are unnamed, and Ripple is a Preview. The 'U.S. open-weight state of the art' headline rests on one coding benchmark, measured by Ricursive and compared with larger models from Reflection AI and Thinking Machines Lab. A team that wants an established, supported model API from a larger provider has other places to look. Hokai's guide to open-weight models and its guide to running AI locally put Ripple in context, and the startup directory lists comparable labs. Developers can try Ripple through Ollama, OpenRouter or hosted inference such as Together AI.
Everything on this page (funding, leadership, compliance posture, the product list) is read from Ricursive Intelligence's own pages and filings, or from the Hokai records for the products it ships. A field the company does not publish is left off the page rather than guessed at.