- Shanghai startup is developing a dedicated chip for AI agents and using AI to automate the chip-design process itself
- Backers bet faster design cycles will become a competitive edge as AI models evolve rapidly
Shanghai-based AI chip startup Trace Intelligence (淬思科技) has completed a seed funding round co-led by Monolith Capital and Qiying Tongchuang Fund, as it develops a specialized inference chip for AI agents and seeks to use AI to shorten the notoriously lengthy process of chip design.
Financial terms of the transaction, announced on June 21, were not disclosed.
The proceeds will be used to develop the company’s first agent-focused inference chip, fund tape-out efforts and expand its engineering team.
Struggling to keep pace
Founded in May 2026 by Pan Hongyang, a PhD graduate from Fudan University’s State Key Laboratory of ASIC & System, Trace is targeting a problem that has become increasingly acute in the AI era: custom chips often struggle to keep pace with rapidly evolving model architectures and fragmented deployment scenarios.
As AI models change, chip specifications become harder to lock in, extending development timelines and forcing companies to choose between cutting-edge performance and rapid tape-out.
“The real bottleneck in this generation of AI chips isn’t compute power, but design speed,” Pan said.
AI-powered workflows
Trace’s approach is to embed AI throughout the chip-design workflow. Tasks traditionally reliant on engineering experience — including iterative testing, design decisions and optimization — are delegated to AI agents that can plan workflows, run tools, evaluate results and refine designs through continuous feedback.
The company said its proprietary Agentic EDA platform covers the entire process from design specifications to final Graphic Data System layout files.
Rather than manufacturing chips itself, the platform automates repetitive engineering work, aiming to reduce development costs and accelerate custom chip creation.
The startup argues that the key metric is not how much computing power a chip delivers, but how quickly a viable design can be brought from concept to tape-out.
Commercialization underway
According to the company, the platform has already generated commercial revenue and its design methodology has been validated through successful tape-outs of real chips.
“The competition in AI inference is shifting from raw computing scale to delivery capability,” said Cao Xi, founding partner of Monolith Capital. “The Trace team has managed to bridge both the tooling and chip-design sides, giving inference chips a better chance of keeping up with changing demand and model architectures.”
Trace is currently developing its first inference chip designed specifically for AI agents. The company said the entire design process is being handled by AI and that it plans to complete tape-out before the end of the year.
