A veteran chip engineer from Huawei has launched a startup focused on solving one of edge AI's biggest challenges: running massive AI models on a single device without relying on cloud servers. Jiuwanli Future Technology, founded by Peng Chen, a former chief of HiSilicon's Kunpeng processor division, has raised more than $100 million across seed and angel funding rounds to develop high-performance inference chips for on-device and edge AI.

Why Does Running Large AI Models on Devices Matter?

Today's most powerful AI models live in data centers. When you ask your phone or smartwatch a question, it sends your request to the cloud, waits for a response, and sends the answer back. This creates three problems: network latency slows things down, privacy concerns emerge because your data leaves your device, and costs add up for companies running massive servers. Edge AI chips have historically been limited to smaller models, making it difficult to bring advanced AI agents and reasoning capabilities to local devices.

Jiuwanli's mission is to change that equation. The startup, founded in May 2026, aims to run ultra-large-parameter models efficiently on a single chip by combining high computing power, energy efficiency, and low latency to support long-horizon tasks and next-generation agentic AI. This approach could unlock AI capabilities on phones, wearables, and industrial devices without the delays and privacy risks of cloud processing.

Who Is Behind This Startup and Why Does It Matter?

Chen brings nearly two decades of chip development experience. During his time at HiSilicon, he oversaw the development of Kunpeng and Ascend processors and led teams of more than 1,000 engineers. His track record includes work on what the company describes as the industry's first TSMC CoWoS-packaged chip, the first ARM64 server chip, and an in-house ARM CPU core competitive with Arm's own designs. He also led the development and mass production of a series of AI training and inference chips within two years.

More recently, at Horizon Robotics, Chen built the company's Journey 6 series product line, which spans computing power from 10 to 560 TOPS (tera operations per second), covering a broad range of computing requirements. His reputation has attracted heavyweight investors including HongShan (formerly Sequoia China), Lenovo Capital and Incubator Group, and Walden Hi-Tech, a local fund affiliate of Walden International.

How to Understand the Technical Challenge Jiuwanli Is Solving

• Model Size vs. Device Power: Large language models with billions or trillions of parameters require enormous computing resources, but phones and edge devices have limited power budgets and processing capacity, forcing a trade-off between model capability and device efficiency.
• Latency and Privacy Trade-offs: Cloud-based AI introduces network delays and requires sending sensitive user data off-device, whereas on-device inference keeps data local and eliminates round-trip communication delays.
• Energy Constraints: Running complex AI models on battery-powered devices drains power quickly, so specialized chips must balance computational performance with minimal energy consumption to remain practical for everyday use.

The startup's focus on edge and on-device inference represents a relatively unexplored area for Chinese AI chipmakers, according to reporting on the funding. While companies like Huawei, MediaTek, and others have released chips with AI capabilities, few have specifically targeted the challenge of running ultra-large models locally. Jiuwanli's $100 million funding round signals investor confidence that this gap represents a significant market opportunity as AI agents become more sophisticated and demand more local processing power.

The timing aligns with broader industry trends. As AI models grow more capable, companies are increasingly interested in keeping inference workloads on devices rather than sending them to cloud servers. This shift is driven by concerns about latency, privacy, cost, and the need for AI systems that can operate reliably without constant internet connectivity. Jiuwanli's approach of combining high compute, energy efficiency, and low latency on a single chip directly addresses these demands.

The startup's success could reshape how AI reaches consumers and industrial applications. If Jiuwanli can deliver chips that run large models efficiently on edge devices, it could enable new categories of AI-powered products, from smarter phones and wearables to autonomous robots and industrial equipment that operate independently without relying on cloud infrastructure. For now, the company remains in development, but its funding and leadership suggest the edge AI chip market is entering a new phase of competition and innovation.