Newshot

After 20+ Years in Consumer Hardware, Why Is Aoni Electronics Building a Token Factory?

AI summary

"What is this for? How do we use it?" After entering the AI inference compute market, a company with over 20 years of manufacturing experience found itself stumped by this question from customers. IDC and CAICT's 2026 joint survey shows that in manufacturing, government, and other sectors, fewer than 35% of enterprises have intelligent compute resource utilization rates of 70% or above.

Translation(model translation, not the source's own text)

Article title: After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?

Body:

After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?
After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?

"What is this for? How do we actually use it?"

A company with more than 20 years of manufacturing experience entered the AI inference computing power market, and was stumped by this question from customers.

Over the past two years, the AI computing power market has been buzzing. Some build computing clusters, some sell servers, and some provide computing power services. But when it comes to actual use, some enterprises find that they have bought the equipment, yet haven't found the applications, and can't connect it to their business.

A 2026 joint survey by IDC and the China Academy of Information and Communications Technology shows that in industries such as manufacturing and government affairs, less than 35% of enterprises have achieved an intelligent computing resource utilization rate of 70% or above.

The supply side is busy selling computing power, while the demand side is still figuring out how to use it.

For Aoni Electronics, which entered the AI inference computing power market not long ago, more than 20 years of manufacturing and supply chain accumulation provide the foundation for hardware production, customization, and delivery in its exploration of the Token factory.

And the "how to use it" raised by customers is precisely the question that past manufacturing experience struggles to answer.

Over 20 years in hardware

Breaking into AI, only to be stumped by software

In June of this year, Aoni Electronics began offering free hardware adaptation testing resources to Agent and AI application development companies.

Why would a company with more than 20 years in consumer electronics manufacturing enter the AI inference computing power market? And why open hardware adaptation to software companies?

This starts with Aoni Electronics' transformation a few years ago. This manufacturing company, which started with products such as smart cameras and automotive cameras, went public in 2021. Affected by intensifying industry competition, it began looking for new business directions.

In 2023, the company entered the computing device field through notebook OEM; in 2025, it further supplied high-end servers to enterprises in cloud gaming, rendering, and other fields, and only clarified its transformation toward the high-performance inference computing market by the end of that year.

From cameras to servers, the products changed, but the mindset for doing business still carried inertia.

Ruan Jian, general manager of Aoni Electronics' AI computing power business unit, told Leiphone.com that the company initially still followed the traditional manufacturing mindset: design the product well, make the appearance good, and ensure solid manufacturing quality, believing this would win over customers.

Customer feedback was blunt: the appearance and product design of the equipment were not their main concern. Customers wanted to know what AI applications these devices could run, and how to connect them to their own business.

To this end, Aoni Electronics tried developing applications itself. But customer demands kept growing—today they want AI-assisted office work, tomorrow they want to connect finance, legal affairs... Needs vary greatly across industries and enterprises, and it is hard to satisfy them with one set of equipment or a few pre-installed applications. "We were actually quite troubled by this too," Ruan Jian said.

A "veteran" of hardware manufacturing ran into new troubles in software development.

However, in exchanges with software companies, Ruan Jian found that they had their own difficulties. "We found that software development companies face even greater competitive pressure than we hardware companies do. But if they want to invest in hardware R&D, they don't have the manufacturing foundation we have."

Hardware companies are familiar with production and manufacturing, but may not understand specific application scenarios; software companies have development capabilities, but find it hard to build hardware R&D, supply chain, and delivery systems in a short time.

Both sides see opportunities in the other's field, and both hit the limits of their own capabilities. (If you know about or have experienced the stories behind enterprises crossing over into AI, feel free to add the author's WeChat treelog10 to share)

Aoni Electronics lacks applications, software companies lack hardware—and this has instead created an opportunity for cooperation.

"We take responsibility for making the hardware well, and open adaptation testing resources for free," Ruan Jian said. The other side lacks a hardware foundation, while Aoni Electronics has manufacturing and customization capabilities, so both sides can play to their strengths.

This is also why Aoni Electronics tried opening up hardware resources. As cooperation progressed, Aoni Electronics also gained support from some model and chip manufacturers.

"We don't call this an 'ecosystem barrier,' but 'ecosystem cooperation,'" Ruan Jian said, noting that Aoni Electronics' capabilities are mainly concentrated in hardware architecture, while the application layer is not where its strengths lie.

At present, Aoni Electronics has developed more than 60 ecosystem partners. Hardware manufacturing and adaptation testing resources are provided by Aoni, while AI applications for specific scenarios are completed by software development companies.

"Rather than developing all kinds of applications on our own, it's better to find professional partners. Our capabilities have boundaries, but every field has its experts."

From a one-off deal to three to five years of service: Token Infra changes the business logic

Token Infra changes the business logic

Finding software partners filled in Aoni Electronics' capabilities, but this still did not truly answer the question of "how to use it."

Ruan Jian put it more bluntly: some enterprises invest funds to purchase equipment but fail to find suitable business applications. "If we only wanted to sell equipment to customers, it would be hard to sell, and we wouldn't be able to sell it."

Aoni Electronics began trying a different way of doing business with customers: instead of rushing to recommend equipment, first understand what the other party actually wants to do with AI, whether it is necessary to purchase equipment, and how much they plan to invest.

Ruan Jian feels this model is somewhat similar to the working style of an FDE (Forward Deployed Engineer): first understand the business, then discuss technology and deployment plans.

Selling computing power equipment is only the beginning; follow-up services may last three to five years.

To achieve long-term management, the first step is to manage computing power equipment well. As Aoni Electronics gradually rolled out desktop workstations, enterprise micro-nodes, and cloud clusters, customers also hoped to connect hardware of other brands purchased earlier into the same system. After all, if every additional piece of equipment requires deploying another management system, it is not only complicated to use but may also lead to wasted resources.

Ruan Jian gave an example: computing power equipment purchased at the beginning of the year may no longer be the latest model by the end of the year, but it still has computing power and value. Only if old equipment can be managed and scheduled together with new equipment can earlier investments avoid sitting idle.

Aoni Electronics' TokenPond platform was gradually built under such demand. The platform initially managed only edge data center equipment, later connected desktop workstations, then linked cloud clusters, attempting to be compatible with hardware from different vendors and centralize computing power scheduling, metering and billing, and operations and maintenance management into one system.

In addition to managing computing power resources, Aoni Electronics has also begun providing inference computing power services through the Token factory. Ruan Jian introduced that Aoni Electronics currently has established computing clusters in three cities, among which the Qingyang project serves autonomous driving business, while the Yuxi project mainly targets rendering scenarios.

Shenzhen is slightly different: Aoni Electronics has built its own Token factory model there, deploying models through GPU clusters and providing inference services, attempting to apply AI to its own manufacturing business.

"Building an enterprise's private Token Infra was initially because we ourselves encountered data security troubles," Ruan Jian said. As a listed manufacturing company, Aoni Electronics' core operations and customer data cannot leave its domain and cannot be directly handed over to external AI platforms for processing.

This is a real problem manufacturing enterprises face when introducing AI today. Involving trade secrets, production data, and customer information, enterprises must consider data security and compliance requirements, and also need to solve issues of model deployment, data management, and the connection between AI applications and existing business systems.

According to Aoni Electronics' vision, enterprises can choose to deploy computing power in desktop workstations, internal enterprise data centers, or dedicated cloud environments based on business scale and data security requirements, and then manage these resources uniformly through TokenPond. In this way, enterprises can manage computing power, deploy models, and connect the needed AI applications within a private Token Infra.

Aoni Electronics has also used itself as a "guinea pig." In addition to building and using its own private Token factory, the company has also set up an internal enterprise Agent R&D team, and related applications will first be trialed in its own business; when customer enterprises raise new application needs, it will also consider verifying them in its own business first.

"Whatever customers need, we can try it ourselves first," Ruan Jian said with a smile.

Moving toward an intelligent computing service provider, while keeping up with technological change

In addition to application development and long-term service, Aoni Electronics also faces another challenge: AI technology is updating too fast.

Ruan Jian introduced that the company currently works with international chip ecosystems such as 英伟达 and AMD, as well as domestic manufacturers such as Huawei, Hygon, and MetaX. Different customers have different choices of chips and computing power solutions, and how to keep up with technological change and provide corresponding products and services has also become something that must be continuously learned during the transformation.

When talking about the accumulation of technical capabilities, Ruan Jian compared himself and Aoni Electronics to students: "We have found good teachers, and we can also count as a hardworking student."

Ruan Jian revealed that in the coming year, the company will continue to increase investment in teams, technology, and products, and gradually build a more complete AI inference computing power service capability.

"This whole journey is full of challenges," Ruan Jian admitted. Aoni Electronics' service system is still under construction, and when selecting projects, it must consider upstream chip supply and its own technical support capabilities.

Regarding this transformation from hardware manufacturing to intelligent computing services, he also admitted: "Actually, I feel quite uneasy inside."

Uneasy as he is, Ruan Jian still has a clear judgment about Aoni Electronics' future: only by truly moving from a manufacturing enterprise to an intelligent computing service provider will this transformation be considered successful. "I hope that in the future, when people mention Aoni Electronics, they think of a 'truly deliverable and trustworthy AI inference computing power service provider.'"

For traditional manufacturing enterprises entering the AI market, starting from the areas they are good at is a good choice, but finding new business on this new track goes beyond manufacturing: it requires understanding customer needs and finding ways to continuously provide services and create value.

Aoni Electronics' exploration continues, and this practical path also provides a new observational sample for traditional manufacturing enterprises expanding into AI business.

(Regarding the transformation of traditional manufacturing toward the AI industry, feel free to add WeChat treelog10 to share stories and views.) Leiphone.com Leiphone.com (official account: Leiphone.com)

After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?
After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?

20 rounds of financing in 10 months: why is third-party AI Infra more sought after the more controversial it becomes?

After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?
After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?

Where electricity is cheap, computing power runs there? It's not that simple

After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?
After more than 20 years in consumer hardware, why is Aoni Electronics building a Token factory?

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by 雷锋网
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗