How long does it take to build an advanced chip? The answer is typically three to five years. The design phase alone consumes two years. And those two years are not idle—they involve hundreds or thousands of engineers working intensively, where a single verification error can reset the entire pipeline.
This is the problem Cognichip aims to solve. The startup, founded in 2024, has just closed a new $60 million funding round, bringing its pre-money valuation to $93 million. The round was led by Seligman Ventures, and Intel's current CEO, Lip-Bu Tan, has joined its board.
Not using GPT to write RTL code
Cognichip's approach differs from most efforts that plug large language models into IDEs. Instead of using a general-purpose LLM to handle hardware description languages, the company trained its own deep learning model from scratch, specifically for chip design data.
This distinction is critical. General-purpose models have very limited ability to understand chip design specifications—timing constraints, power budgets, process library rules—things that cannot be learned by scraping internet text. Cognichip signed agreements with multiple chip companies to access proprietary design data, introduced synthetic data pipelines, and built an isolated training environment that allows chipmakers to fine-tune models with their private data without fear of leaks.
The company also used the RISC-V open-source architecture for demonstration and validation—showing that an AI can run through a complete design flow, even for electrical engineering students.
CEO Faraj Aalaei explained their goal: "These systems are now smart enough that you can tell them what result you want, and they can actually produce good code."
The numbers: 75% and 50%
The company claims it can reduce chip development costs by more than 75% and cut development time by more than half.
These two figures are staggering in the semiconductor industry. Current mainstream EDA (electronic design automation) tools are mature, but the workflow largely follows paradigms established decades ago: manually writing RTL, synthesis, placement and routing, timing analysis, tape-out verification—each step requires dedicated engineers. The potential of AI intervention is not to replace a single step but to compress the entire iteration loop—shifting engineers from writing code to reviewing results.
However, these numbers are the company's own claims and have not been publicly verified in a large-scale production environment. Seligman managing partner Umesh Padval said this is the largest capital influx he has seen in 40 years of investing in AI infrastructure—but that also underscores how hot the sector is, and funding amounts should not be mistaken for technical validation.
Several heavyweight players are already in this race
Cognichip is not alone. In the first few months of 2026, competitor ChipAgents completed a $74 million Series A, and Ricursive raised a $300 million Series A. Together with Cognichip, more than $400 million has flowed into the AI chip design track in just the first four months of 2026.
Established players Synopsys and Cadence Design Systems, the two EDA giants with a combined market cap exceeding $100 billion, are also heavily investing in AI-assisted design features. The pressure on startups is considerable.
Lip-Bu Tan joining Cognichip's board is worth noting. Intel itself is one of the world's largest chip companies, and having its CEO personally endorse an AI chip design startup sends a signal that goes beyond money. He brings deep understanding of chip design workflows and potential willingness for Intel to collaborate in this direction.
AI helping to build its own brain
The most fascinating aspect of this track is its recursiveness: AI models themselves need chips for training and inference, and now people are using AI to design those chips. If design cycles can truly be compressed significantly, the iteration speed of next-generation AI accelerators will no longer be limited solely by the number of engineers, but also by the capability of AI design tools themselves.
This is a positive flywheel and the fundamental logic behind the massive investments.
The real test for Cognichip going forward is whether it can produce a case where a major company uses its tools in a real tape-out project. With $93 million in the bank, it has enough ammunition to burn for a while.
Sources: CocoLoop, Cognichip wants AI to design the chips that power AI, and just raised $60M to try (TechCrunch)