Skip to main content

AI and the Future of Chip Design: A Q&A With ChipAgents Founder William Wang

ChipAgents avatar
ChipAgentsNovember 13, 2026

ChipAgents founder and CEO William Wang joined an interview with theCUBE at the New York Stock Exchange to discuss how AI agents are changing semiconductor engineering, from digital verification and analog design to custom chips for robotics.

This Q&A has been lightly edited for clarity, removing filler words and repetition while preserving the substance and sequence of the conversation.

Tell us about your background and what led you to start ChipAgents.

Essentially, we’re seeing the complexity of chips growing exponentially over the years, whereas the fundamental technologies and software supporting the solutions are lagging behind. Another big issue is that we don’t have enough engineers to build the chips. So we see a huge opportunity.

My background is AI research. I started building some of the earliest neural network language models back in 2011. In 2016, I finished my PhD at Carnegie Mellon University in AI and machine learning, and then I became a professor at UC Santa Barbara.

In the last 10 years, we built some of the earliest reinforcement learning models for reasoning. That started in 2017. In 2024, I realized there was a huge opportunity to really reimagine the whole stack and think about AI-native research for semiconductor and chip design. That’s when we started the company.

Two years later, we’re serving more than 200 global semiconductor companies. We’re backed by Micron, MediaTek, Samsung, Ericsson, and many strategic investors. It’s an incredible time to think about how we can use AI-native methodologies to enable more people to build chips.

There are many different chips for different needs. What kinds of companies come to you, and why?

When ChipAgents came to market two years ago, we started in an area called digital verification. This is verification specifically for digital chips. It used to be the case that verification would take about five or six months of the entire 18- to 24-month tape-out cycle. That was one of the biggest pain points and how we entered the market.

Right now, we’re seeing much broader use cases for our solutions. We went from verification to design architecture, firmware, embedded systems, and now digital implementation. This is about improving and optimizing your chip for power, timing, and performance, and minimizing the area.

ChipAgents has become an end-to-end solution to optimize these digital chips. But in recent quarters, we’re also moving into the analog space. There’s a growing number of analog engineers using ChipAgents to do schematic design and layout, which used to be a very painful process, with everything done by hand.

With our customers, we’re seeing tape-out cycles shrink from 18 months to 12 months, and even further.

The beauty of our solution is that it doesn’t really depend on a particular type of chip. We work with hyperscalers, GPU companies, CPU companies, IP companies, memory companies, and FPGA companies. We’re very broad and very general. That’s how we can scale our solutions to a large number of customers.

What does it mean to condense a verification cycle, and what are you testing for?

There are two things. One is digital implementation. You can shrink the cycle, but even if you finish the project in two months, if it doesn’t meet the power requirement or the timing requirement, it still doesn’t work. We have to take all of the constraints into consideration while shrinking the cycle.

Going back to verification, we’ve dramatically accelerated the tape-out cycle for formal verification, functional verification, and many of the solutions that used to be very demanding.

You have to hire formal verification engineers who understand software, know your hardware, and have to be mathematicians at the same time. These people don’t exist on the market.

With our technology, we’re enabling a lot more engineers to run the full verification flow. Chip designers are now able to offload some of the tedious implementation work to ChipAgents, where they can focus on higher-level verification planning and architectural planning.

I think this is truly a revolutionary technology that is changing how people work and changing job titles. The legacy tools were buried in different silos, but we’re breaking down the barriers and doing things end to end.

What does collaboration with a customer look like?

We first understand exactly what the customer needs. We have to understand what type of chips they’re building. Are they building analog chips? Are they building digital chips? If it’s mixed-signal, what part is analog and what part is digital?

We have to understand their product and their pain points. For a large number of our customers, their pain point is verification, because verifying the digital part was the most time-consuming part of their process.

Once we start engaging, we set up a plan for evaluation with the customer. Our teams execute together and deploy our solutions in the customer’s environment. We measure the performance of our solution compared with their in-house solutions, or sometimes other third-party solutions.

At the end of the day, customers choose ChipAgents because they believe in “buy and build.” They license our technology and build with us. They also put their roadmap on our roadmap, and we accelerate together to build the next generation of their chips.

How do you handle customer IP and the knowledge gained from working across different chip designs?

Data security and IP security are among the industry’s top priorities. We would never train on customer data. We would never use one customer’s data in another customer’s environment.

We build an ontology layer that is deployed in the customer’s environment. We don’t have access to the network. This ontology layer does test-time exploration. It tries to understand your prior design, your new design, and what your prior test cases look like. It can improve as you interact with the solution.

That’s extremely critical because it explores the testing environment and gets better over time as users start using it.

Customers can draw on the enterprise know-how they’ve had for the last 20 years, bring everybody together under the same project, and bring everybody’s AI know-how up to the same level as their superusers. Then they’re seeing this acceleration in their tape-out cycles.

Everyone seems to want to build chips. Do you think that makes sense for more companies?

We have to separate design and manufacturing. Manufacturing is really tough, and it’s now very concentrated in terms of the fabs.

But here we’re talking about design, and design is really interesting. Right now, it’s data centers. But think about five to 10 years from now, especially physical AI and robots.

One of the biggest challenges with physical AI and robotics is that a lot of these companies started in the last few years. So far, they don’t have the capability to build their own chips.

That’s a big issue. When you wait for a robot to open your fridge and it takes two minutes, that’s too late. You can do it in 10 seconds yourself.

That agility and those latency requirements really require robotics companies to be able to build application-specific chips. I think this is where the industry is going.

If you look back 10 or 20 years, there were only semiconductor companies building chips. Right now, there are system companies, hyperscaler companies, and automobile companies. We’re predicting that in the next three to five years, there will be many more robotics companies building their own chips.

But who are they going to hire? Are there enough engineers they can hire to build the chips? The answer is no. There aren’t enough chip design engineers in the world.

You have to use agentic AI and ChipAgents technologies to accelerate development in physical AI beyond the current data center wave.

You’ve raised significant funding. What’s ahead for the company?

We raised $136 million up to our Series A2 round. I think it’s an incredibly exciting time to build a startup in this area.

People used to stay away from this area. Now hardware has become really hot, and we’re at the intersection of software supporting hardware.

We’re taking a fundamentally new approach. People see a huge opportunity for AI to come in and not just disrupt the industry, but fundamentally change how people think about the process and how they build chips. That’s why we’ve been getting a lot of interest from investors, including strategic semiconductor investors.

But I think this is just the beginning. Two weeks before this interview, we did a benchmark with NVIDIA. It showed that frontier models are still lagging behind on many physical design problems, including timing, design rules, and other aspects.

We think there’s a huge opportunity in the next few years to break barriers in front-end logical design, back-end design, manufacturing, and testing.

Having enough capital is very important, especially as we proceed along our roadmap into 3D IC and advanced packaging. It’s not easy. We’re only at the beginning of our journey.

Watch the full interview with William Wang on YouTube.