Cognichip raises $60M to let AI design the chips for AI

The most advanced silicon chips have significantly accelerated the development of artificial intelligence. Now Cognichip is attempting to return the favor by building a deep learning model that works alongside engineers in designing new computer chips. The problem they are addressing is one the industry has faced for decades: chip design is enormously complex, prohibitively expensive, and slow. Advanced chips can take three to five years to go from conception to mass production, with the design phase alone taking up to two years before physical layout begins. For instance, the latest line of Nvidia GPUs, Blackwell, contains 104 billion transistors — a lot to align. According to Cognichip's CEO and founder Faraj Aalaei, by the time a new chip is created, the market can shift, rendering all that investment futile.

Aalaei's goal is to bring the type of AI tools that software engineers have used to expedite their work into the semiconductor design realm. “These systems have now become intelligent enough that by just guiding them and telling them what result you want, it can actually produce beautiful code,” Aalaei told TechCrunch. He claims that the firm’s technology can reduce chip development costs by over 75% and cut the timeline by more than half.

The company emerged from stealth mode last year and announced on Wednesday that it had raised $60 million in new funding led by Seligman Ventures, with notable participation from Intel CEO Lip-Bu Tan, who will join Cognichip’s board. Umesh Padval, a managing partner at Seligman, will also join the board. To date, Cognichip has raised a total of $93 million since its founding in 2024. However, the company cannot yet point to a new chip designed using its system and did not disclose any of the customers it claims to have been collaborating with since September.

Cognichip asserts that its advantage lies in using its own model trained on chip design data, rather than starting with a general-purpose LLM. This required gaining access to domain-specific training data, which is no small feat. Unlike software developers, who openly share vast amounts of code, chip designers closely guard their IP, making the kind of open-source trove that typically trains AI coding assistants largely unavailable. Cognichip had to develop its own datasets, including synthetic data, and license data from partners.

The firm has also developed procedures allowing chipmakers to securely train Cognichip’s models on their proprietary data without exposing it. Where proprietary data isn’t available, Cognichip has leaned on open-source alternatives. In one demonstration last year, Cognichip invited electrical engineering students at San Jose State University to try the model in a hackathon. The teams were able to use the model to design CPUs based on the RISC-V open-source chip architecture — a freely available design that anyone can build on.

Cognichip is competing against established players like Synopsys and Cadence Design Systems, as well as well-funded startups like ChipAgents, which closed a $74 million extended Series A in February, and Ricursive, which raised a $300 million Series A round in January. Padval noted that the current influx of capital into AI infrastructure is the largest he has seen in 40 years of investing. “If it’s a super cycle for semiconductors and hardware, it’s a super cycle for companies like Cognichip,” he said.