Chip design is dirty work.
Not literally dirty — it’s the kind of dirty that’s time-consuming, expensive, and mentally exhausting. It takes three to five years to go from concept to mass production for an advanced chip, with the design phase alone taking two years — and that’s before any actual wiring starts. Just think about it: Nvidia’s Blackwell GPU packs 104 billion transistors. Simply arranging and combining all those components is enough to wear engineers out.
But there’s a startup called Cognichip that thinks this hard life shouldn’t continue.
Solving the “Chicken-and-Egg” Problem of AI Helping AI
The logic is simple: Advanced chips run AI, so why can’t AI be used to design better chips?
Cognichip’s technology is essentially a deep learning model that assists engineers. CEO Faraj Aalaei puts it plainly: today’s AI is smart enough that “you just tell it what result you want, and it can write beautiful code for you.” He wants to bring the AI assistants that software engineers rely on directly into the semiconductor design field.
The goal is clear: cut development costs by more than 75% and reduce the development cycle by more than half.
Honestly, if they can really pull this off, the entire industry will be shaken.
$60 Million, with Intel’s CEO Personally Investing
Cognichip announced a new $60 million funding round on Wednesday, led by Seligman Ventures. Intel CEO Lip-Bu Tan not only participated in the investment but is also joining the board. From founding to now, Cognichip has raised a total of $93 million.
However, there’s one thing they currently can’t account for: Where are the chips designed using their system?
Cognichip can’t show specific finished products, nor have they disclosed any named customers. They only say that “after emerging from stealth mode last year, they’ve been working with customers since last September.”
To be fair, it’s normal not to see results at this stage, but when investors are spending money, this is still a question mark that makes people nervous.
The Hard Part Is Data, Not Technology
Many people think AI writing chips is as simple as AI writing code. It’s actually completely different.
Why can software engineers use Copilot to write code so fast? Because there’s a massive amount of open-source code worldwide to train on. Chip designers? Every company’s IP is locked up tight. The training data you need simply doesn’t exist in the open-source community.
Cognichip’s solution is building their own dataset — combining synthetic data with patented data licensed from partners. They’ve also developed a mechanism that allows chip manufacturers to safely train models on their own servers, without leaking core IP.
No data? Use the open-source RISC-V architecture as an alternative. Last year they held a hackathon where electrical engineering students from San Jose State University used their model to design CPUs. The results were pretty good.
None of the Competitors Are Weak
Cognichip isn’t the only company that wants to do this.
The established EDA giants Synopsys and Cadence have long held firm positions in this market. Startups aren’t sitting idle either — ChipAgents just completed a $74 million extended Series A in February, and Ricursive raised a $300 million Series A in January.
Seligman partner Umesh Padval said something quite interesting. He said that in 40 years of investing, he’s never seen the scale of capital flowing into AI infrastructure like it is now. “If semiconductors and hardware are a supercycle, then companies like Cognichip are the supercycle within the supercycle.”
My Take
From my perspective — I work in cell therapy process development — this story of using AI to accelerate high-complexity design processes follows the same logic as using AI to optimize bioprocessing in our field. Data is the barrier, speed is the value, but the real bottleneck is always in real-world validation.
For Cognichip to prove itself, it doesn’t rely on PPTs or valuations. It relies on whether it can deliver a chip that truly goes from concept to silicon.
At that point, the rules of the chip design game may be completely rewritten.
