Summary of Key Points
In July 2026, the Kimi K3 large model released by Moon's Dark Side utilized an AI agent to complete the entire process of chip design from architecture to simulation verification in 48 consecutive hours. This was achieved using open-source EDA tools and a 45nm manufacturing process, resulting in a 4mm² chip containing 1.46 million standard cells. This development has caused a stir within the semiconductor industry: stock prices of EDA leaders Synopsys and Cadence have declined, leading to questions about whether AI will redefine the value of these companies. However, it's important to maintain a calm perspective. Kimi K3 is not intended to replace chip design teams; rather, it demonstrates the potential for AI-automated design. Its real impact lies in lowering the barriers to chip design and changing the production methods, enabling smaller teams to participate in complex chip development. The key to future competition will be the data and knowledge ecosystems within the chip industry.
Detailed Analysis
1. No need to panic! Kimi K3 didn't create a “high-end flagship chip”
Many people mistakenly believe that this means EDA companies are doomed or that anyone can design 3nm GPUs. There are two significant limitations with Kimi K3's chip:
- Limited process and scale: The chip was manufactured using a 45nm process (while current smartphone chips use 3nm), and its area is only 4mm² (about one-tenth the size of a fingernail). The number of transistors is far fewer compared to Apple's A-series or NVIDIA GPUs, which have billions of transistors.
- Commercialization challenges unresolved: Real-world chips must balance factors such as power consumption, performance, area, yield, and manufacturability (known as PPA optimization). For example, smartphone chips need to have long battery life, low heat generation, and be producible at a profit. Kimi K3 only demonstrated that the design process can be automated; it has not produced a usable product.
Therefore, the significance of Kimi K3 is in proving that AI can automate the entire design process, not replacing engineers.
2. AI is changing the way chip design is done
Traditionally, chip design involved humans operating software: engineers had to learn various EDA tools (such as Design Compiler and Innovus), manually write code, adjust parameters, and run simulations, with dozens of people working together for months. With Kimi K3, the process becomes more efficient: users specify their requirements, and AI handles the rest—generating the architecture, writing code, optimizing with tools, and running simulations. This represents a shift in the production paradigm.
EDA giants have long been integrating AI into their tools (e.g., Synopsys has incorporated AI into its design software). However, the capital market is concerned that if AI becomes the primary point of entry for chip design (users first use AI and then it uses EDA tools), the value of EDA companies could shift from selling tools to providing services for AI. This is likely the underlying reason for the decline in their stock prices.
3. Why does this impact EDA companies?
It's not that software will be phased out, but rather that the barriers to entry have been lowered. In the past, mastering EDA tools required years of training; now, AI can handle the tedious tasks, allowing engineers to focus on understanding the design requirements. This is similar to how Photoshop has made graphic design more accessible to non-professionals and Claude Code has accelerated coding. AI in chip design won't eliminate EDA software, but it reduces the barriers, potentially enabling more people to enter the industry and weakening the monopolistic position of existing companies.
4. The greatest opportunity: Smaller teams can now engage in chip design
Previously, starting a chip business was extremely challenging, requiring large teams, millions of dollars in EDA licenses, and years of experience. Now, AI can handle 80% of the basic tasks (such as code writing and parameter tuning), leaving only 20% for human decision-making (e.g., chip architecture and market demand). For example, a task that previously required 100 people could now be completed by 10 individuals.
This is good news for China's chip industry, which lacks high-end talent. AI can help fill the gap in basic skills, giving smaller companies and startups more opportunities to participate in chip design without relying on large corporations.
5. The key to AI-driven chip design: data, not model size
AI in chip design is different from ChatGPT writing articles. While ChatGPT can learn from internet text, chip design requires specialized data, such as real chip code libraries, process rules, and feedback from manufacturing processes (issues that arise after chip production). This data is primarily held by giants like NVIDIA, TSMC, and Synopsys.
Therefore, the winner in the future will not be the company with the largest model parameters but the one with the most extensive knowledge and data on chip design. For instance, if TSMC provides its process data to AI, its AI-driven designs will likely outperform those of companies without such data. This is why EDA giants and chip manufacturers are racing to accumulate this valuable information.
Conclusion
Kimi K3 marks a significant milestone, but it's not the beginning of an era where AI completely replaces human engineers in chip design. Instead, it opens the door to more efficient and accessible design processes, allowing smaller teams to participate. EDA companies will still be necessary, but they must adapt to a new role where AI serves as a tool for design. The future competition will focus on data and knowledge ecosystems. For the general public, this could lead to more innovation in the chip industry, possibly resulting in chips designed by smaller teams that become more common.