Companies are shipping AI hardware to China

Hello, I am Evelyn, writing to you from Beijing. Welcome to the latest edition of The China Connection — A brief snapshot of what I’m seeing and hearing from local businesses.

In Hangzhou, startups are working on both software and hardware while building devices to run AI. How does this change the AI ​​technology race?

As Chinese cloud companies There was a rush to promote OpenClaw in early March, with a startup in Hangzhou already making the device.

iinclaw It shipped its first 100 units on Friday of the $43 clip-on mic, which lets users send voice commands to an OpenGL AI agent. Co-founder Arvin Chen told me when I visited a WeWork-style office in Hangzhou last week that just two people had developed and assembled the device from parts sourced from across China.

OpenClaw functions are also coming to robots. In nearby Suzhou, startup Joyin claims its Zeroth M1 humanoid is the first to do so. using the Tencent The start-up said that through the cloud tool, people can send commands to the robot and control it remotely. Pre-orders are expected to begin by July.

Together, these point to a shift across industries – from Internet-only AI to hardware.

“Cloud-native is a little outdated. The technology is useful, but the business model is a little outdated,” Ray Vaughn, founder, CEO and chairman of Tencent-backed OpenPie, told me last week. “Data sovereignty is a concern right now.”

That’s a big deal, coming from a startup founded on building cloud data systems.

from china millions of factories Describe the limitations of cloud-only AI. While they’re interested in unlocking AI efficiencies, Vaughn said manufacturers are concerned about sending proprietary information to the cloud. So, he said, OpenPie is building tools that enable AI tools to run locally using low-cost Chinese chips.

The goal is to ship 10,000 of these boxes at a price of 100,000 yuan ($14,627) by the end of this year, Vaughn said, before it is increased.

Expanding into the physical world is also changing software-first companies like Style3D, which was launched in 2015 to help textile companies speed up the design-to-production process by using AI.

So many companies asked Style3D for its data on physical materials and textures that the company decided to enter the business itself — launching the robotics platform SynReal last year, CEO Eric Liu told me on the sidelines of the Hangzhou Venture Capital Association conference on Thursday. He says that to work well in the real world, humanoid robots will need a range of specialized information on texture to understand objects ranging from oranges to silk scarves – data his company can provide.

Startups aren’t the only ones chasing the hardware trend. Electric car companies, including German automaker Volkswagen, announced last week that they are bringing on-vehicle AI tools to respond to driver voice commands.

alibabawhich has focused primarily on in-app AI tools, also revealed this month that its maps unit, Amap, is developing a four-legged robot.

Then, the company aims to use 20 years of specialized data from digital mapmaking to gain an edge in robotics. Its initial goal is to help blind people, given the shortage of guide dogs in China.

Map data can assist robot sensors with navigation, while AI tools enable the robot to find nearby convenience stores, for example, based on signals like “I’m thirsty,” said Mu Xu, head of AI algorithms embedded in Amap.

But he warned that, especially for robotics, the ability to process powerful AI on devices becomes critical – and the biggest challenge.

Once that hurdle is solved, the question will not be how capable AI models are theoretically, but rather what the technology can do once inside every device.

Foreign car companies bet on technology to capture China’s once-lucrative auto market
US, Korean and German automakers rushed to announce The new lineup of models for China — using locally developed AI from ByteDance and other Chinese companies — debuted around the Beijing auto show that started on Friday.