AI chip startup Ricursive Intelligence hits $4B valuation two months after launch

AI chip startup Ricursive hits $4B valuation two months after launch

Ricursive Intelligence, a startup building an AI system to design and automatically improve AI chips, has raised $300 million at a $4 billion valuation. The company said Monday the round was led by Lightspeed. Ricursive says the system will be able to create its own silicon substrate layer and speed up AI chip improvements. Rinse and repeat to get to AGI, the founders say. The Series A comes just two months since the company formally launched with a seed investment led by Sequoia. It has raised $335 million total, reports the New York Times. Ricursive was founded by former Google researchers CEO Anna Goldie and CTO Azalia Mirhoseini. Their work on a novel reinforcement learning method for designing chip layouts, called AlphaChip, has been used in four generations of Google's TPU chip, the startup says. DST Global, NVIDIA's venture capital arm NVentures, Felicis Ventures, 49 Palms Ventures and Radical AI are also investors. Ricursive is not to be confused with the similarly named startup Recursive, reportedly founded by well-known natural language processing neural networks researcher Richard Socher. That Recursive is also in talks to raise a giant round at a $4 billion valuation, Bloomberg reported last week. And it is also working on AI systems that improve themselves. And these two are not the only new startups working on the concept. As TechCrunch previously reported, Naveen Rao's new AI hardware startup, named Unconventional AI, is also working on an intelligent substrate. In December it raised a $475 million seed round at a $4.5 billion valuation led by Andreessen Horowitz and Lightspeed Ventures, with participation from Lux Capital and DCVC.

Ricursive Intelligence nabs $300M to speed up chip design with AI

Ricursive Intelligence nabs $300M to speed up chip design with AI Ricursive Intelligence Inc., a startup that plans to use artificial intelligence to speed up chip development, has raised $300 million in funding at a $4 billion valuation. Lightspeed led the Series A investment. Recursive disclosed in its funding announcement today that the venture capital firm was joined by Nvidia Corp.'s NVentures, DST Global, Felicis Ventures and several others. The raise comes less than two months after Ricursive was launched by former Google LLC researchers Anna Goldie and Azalia Mirhoseini. In 2020, the duo co-created AlphaChip, an AI system that the search giant uses to speed up internal chip projects. The software helped Google engineers accelerate the development of several TPU accelerators. Ricursive plans to train AI models with a similar purpose as AlphaChip: speeding up the development of AI accelerators. Currently, designing a cutting-edge data center processor can take several years. AlphaChip can design some semiconductor components in under six hours. The most advanced processors on the market include well over 100 billion transistors. Engineers must determine where each transistor should be placed on the host chip, how it should be connected to the other transistors and how much power it should receive. There are trillions of potential combinations, which makes it difficult to find the best design. AI can speed up development by quickly evaluating a large number of potential chip layouts. Doing so manually is time-consuming partly because chip performance is not the only factor that engineers must consider. Processors must stay below certain heat, power consumption and surface area thresholds. Chip projects usually also have a long list of more granular requirements. For example, engineers may wish to limit a processor's wirelength, the aggregate length of the tiny networking wires that link together its circuits. Reducing a chip's wirelength lowers power consumption and helps decrease the occurrence of manufacturing errors. Ricursive will use the proceeds from its Series A round to hire more engineers and researchers. Additionally, it will upgrade the infrastructure that it uses to train AI models. The company faces competition from established chip design software providers such as Synopsys Inc. and Cadence Design Systems Inc. Both companies provide AI features that automate manual aspects of the chip development process. Those features can also speed up subsequent engineering tasks such as testing a newly created design for manufacturability issues. The AI-powered chip design software market could become even more crowded over time. Last year, OpenAI Group PBC revealed that it's using its large language models to design a custom AI chip. Anthropic PBC's Cloud Claude is also capable of automating certain electrical engineering tasks.