news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Source: Times Finance Author: Xie Silin

Editor's Note: "China's Science and Technology Innovation Story" is an interview column that tells the science and technology innovation stories of Chinese listed companies. We focus on the ecosystem of science and technology innovation companies and convey the important role of finance in serving the real economy.Through vivid cases, we witness the rise of China's scientific and technological innovation. This is the second interview, looking at how Intellifusion, the "first AI stock in Shenzhen" and a company listed on the Science and Technology Innovation Board, breaks through the AI ​​market.

As the "first AI stock in Shenzhen", Yuntian Lifei (688343.SH) has great ambitions. After successfully listing on the Science and Technology Innovation Board in April last year, Yuntian Lifei has made multiple moves to rapidly expand its business territory.

On the one hand, Yuntian Lifei is trying to establish in-depth connections with more upstream and downstream partners in the artificial intelligence industry. It has not only invested in leading companies such as Smart Interconnect, Zhenshi Technology, and Shenzhou Yunhai in the fields of dynamic parking, static parking, holographic intersections, smart construction sites, and cleaning robots, but also participated in Shenzhen Guochuang Embodied Intelligent Robot Co., Ltd. as one of the founding shareholders, providing technical support in the fields of AI chips and large models.

In March this year, it will acquire Shenzhen Xicheng Technology Co., Ltd. (hereinafter referred to as "Xicheng Technology"), a smart wearable solution design company, for no more than RMB 180 million to complete the C-end entry map of AI large models. This move has attracted widespread attention and discussion in the market.

While deploying multiple tracks for terminal products, Yuntian Lifei has also received a large order for intelligent computing services. On July 2, Yuntian Lifei announced that it had signed a "Service Contract for AI Computing Power Operation Project" with Beijing Deyuan Fanghui Technology Development Co., Ltd. (hereinafter referred to as "Deyuan Fanghui") to provide it with a three-year AI training and reasoning computing power service, which will bring the company a total revenue of more than 1.6 billion yuan.

To ensure delivery, Yuntianlifei expects to use no more than 1.546 billion yuan of its own and self-raised funds to first pay for equipment purchases, and pointed out that if the user terminates the contract early and does not renew the lease, it may affect the company's liquidity.

As an AI chip company that started with the unique advantage of "algorithm chipization", Yuntian Lifei has two core technology platforms, algorithm and chip. It not only released the domestically produced 14nm large model inference chip DeepEdge10, which is the first in China, but also recently launched the IPU-X6000 accelerator card for large model cloud inference, which has a built-in efficient large model inference DeepEdge200 chip with 256T computing power; in addition, it has also developed a cloud Tiantianshu large model with hundreds of billions of parameters. The previous business model was also more focused on providing hardware and software integrated AI solutions for G-end and B-end customers.

From this perspective, Yuntian Lifei seems to be more suitable to be an AI "water seller" than a gold digger who personally goes out to dig. What prompted this company to invest in the computing power service field and hardware consumer products? How will Yuntian Lifei break through in the fiercely competitive market?

To find the answers to these questions, Times Finance interviewed Deng Haoran, senior vice president, CFO and secretary of the board of directors of Yuntian Lifei. Deng Haoran wore simple frameless glasses and a business casual outfit, giving people an easy-going impression. Before the interview, Deng Haoran, who had just rushed from the company's product seminar, excitedly introduced to Times Finance an AI application that Yuntian Lifei was exploring.

This is a prototype of a mobile APP. Users can easily identify objects in front of them and interact in multiple languages ​​by simply clicking and taking photos. In Deng Haoran's vision, this APP is just a pilot product for trial promotion. The algorithm behind it can be combined with the company's edge AI products in the future to form a new form of AI learning product. It is revealed that the product is expected to be launched on the market within the year.

On this special occasion of the fifth anniversary of the launch of the Science and Technology Innovation Board, Times Finance also reviewed with him the insights and experiences of Yuntian Lifei since its listing on the Science and Technology Innovation Board in April 2023, and discussed topics such as how the capital market and corporate development can promote each other, and promote the high-quality development of listed companies while developing new quality productivity.

AI must be productized to open up the market

Times Finance: The market still thinks of Yuntianlifei as a technology company that focuses on algorithms and chips. What is the business logic behind Yuntianlifei's plan to develop the above-mentioned AI application products in the future?

Deng Haoran:Algorithms and chips have always been the cornerstone of Yuntian Lifei's development, and this will not change. However, ordinary consumers usually do not directly purchase algorithm products. Within the scope of the overall purchasing power of society, including various industries, enterprises and government agencies, the purchase volume of algorithms is generally small. Generally speaking, only the information technology department within an organization will have the demand for purchasing algorithms.

Intellifusion must transform its core technology into a competitive commodity. This is particularly critical for a listed company, because only in this way can the company realize its earnings promise to shareholders. For example, the AI ​​application that the company is exploring is intended to transform the company's strong technical capabilities into a commodity that has broad appeal in the market.

Times Finance: Looking further, what kind of business model does Yuntianlifei want to create in the future?

Deng Haoran:Selling to government and enterprise customers, the integrated hardware and software AI industry solutions will account for up to one-third of the company's revenue in the future. Our requirement for this business segment is to increase profit margins, and in the future we will only accept orders with higher gross margins and good payment returns. Only in this way can this business achieve sustainable development together with customers.

Secondly, it is the computing power operation business. The company has been deeply deploying the intelligent computing industry since 2023. Before disclosing the service contract with Deyuan Fanghui, the company already had sufficient technical reserves, had gradually developed relevant business customer resources, formed some projects that have been implemented and obtained relevant income or orders in hand, and obtained good economic benefits from them.

We expect that the computing power operation business will contribute more revenue and a considerable part of the company's profits in the next three years. The development of large models is inseparable from the support of computing power operations. From a longer-term perspective, we are full of confidence in the computing power operation business.

The third part is standardized AI products. We hope to provide Copilot assistants to industries, enterprises and even individual consumers by creating more concrete AI products.

It may be presented in the form of software, such as the government-related large-scale model software independently developed by Yuntian Lifei, which has a higher accuracy in writing official documents than ordinary general-purpose large-scale models. In addition, it may also be presented in the form of smart hardware products. In the future, we will focus on smart wearable devices such as glasses, watches, and headphones. Currently, one or two demo versions of the products are being tested and polished.

In fact, the company has already occupied a certain market share through the acquisition of Xincheng Technology. Last year, the shipment volume reached 14 million sets, and it will increase further this year. At the same time, Xincheng Technology's own product design capabilities and channel recognition in the ToC market will greatly help Yuntian Lifei expand and develop in the field of smart hardware products.

Times Finance: In terms of new business, Yuntian Lifei's self-developed Yuntian Tianshu big model and the acquired Chengcheng technology can bring certain competitive advantages to the company. So in terms of computing power operation, how does Yuntian Lifei cope with the fierce competition in the market?

Deng Haoran:In essence, Yuntian Lifei's core advantage lies in its understanding of "algorithm chipization". We are very clear about what kind of infrastructure algorithm companies need and can provide them with sufficient relevant resources. This is why Yuntian Lifei was able to successfully win this 1.6 billion yuan order.

Intellifusion already has a strong model training foundation and relevant experience, and is able to understand what kind of products customers need and pre-classify them effectively. For example, Intellifusion's system and software can help customers process raw data, significantly reducing their manpower investment and time in data annotation, and achieving cost reduction and efficiency improvement.

The most important work in the cooperation with Deyuan Fanghui is now focused on delivery, and I am optimistic about this. The project team is actively promoting subsequent delivery, acceptance and other work according to the plan, and will disclose further progress in a timely manner if it meets the information disclosure standards.

There is no ceiling for the edge AI inference market

Times Finance: More and more industry giants are beginning to embrace edge AI. Google, Microsoft, and Meta have all launched lightweight large-scale models to support edge AI applications, and giants such as Apple and Nvidia have also launched edge AI chips. Why has this change occurred? In your opinion, what are the advantages of edge AI?

Deng Haoran:There are at least three advantages to performing AI reasoning at the edge. The first is data security. By introducing AI to the device side, data processing can be performed locally, avoiding the transmission of sensitive data to the cloud, which enhances data privacy and security.

Second, AI reasoning at the edge does not generate latency. For example, smart speakers can respond quickly when they are first awakened, but when we want to perform more complex tasks, the speaker needs to connect to the network to calculate and then transmit back to complete the task, resulting in response delays. The use scenarios of smart speakers are limited and far from the level of Copilot.

But if we can deploy AI directly on the edge, we can use the model without a connection, avoiding the network instability and transmission delay problems that may arise when traditional cloud models need to transmit data to the data center over the network, thereby improving the reliability and timeliness of AI application use.

The third point is lightweight and miniaturization. Traditional cloud model parameters are over 100 billion, and the computing cost of a single calculation call is huge, making it difficult to promote. However, by directly deploying AI on the edge, a lower and smaller-scale model can be used to meet the needs of segmented scenarios.

Times Finance: At this year's WAIC, Intellifusion announced a new edge AI strategic layout. Why did the company do this? What advantages does Intellifusion have in the development of edge AI?

Deng Haoran:The existing computing system cannot meet the market demand for high computing power, light weight, and low latency. From this perspective, edge AI chips will definitely replace existing terminal chips in large quantities in the future, and a new round of reshuffle is inevitable.

At the same time, as the training of large models gradually improves, the potential for combining artificial intelligence with various devices continues to expand. This is not only reflected in traditional devices such as smartphones and personal computers, but also in wearable devices such as smart watches, glasses, and headphones, as well as emerging terminals such as robots, drones, and self-driving cars. The market size of these terminals will undoubtedly exceed the market size of existing consumer electronics products.

From this perspective, as the number of smart terminal devices purchased by consumers increases, the demand for artificial intelligence applications also grows, and there is no ceiling for the market for edge AI inference.

Such a huge market cannot be monopolized by a single company, which provides opportunities for Intellifusion to develop. Whether it is the release of the Deep Edge 10 edge inference chip or the continued exploration of standardized intelligent hardware, it is based on this logic.

Moreover, Yuntian Lifei also has strong technical advantages and accumulated experience in this regard. Although we still have a certain gap with the world's advanced level in training chips and cloud inference chips, Yuntian Lifei's inference computing power at the edge has reached 64Tops, which can meet the market needs of low energy consumption, fast response and miniaturization at the same time. Its technical capabilities are definitely among the top three in the country.

Financing capabilities significantly improved after listing

Times Finance: This year marks the 5th anniversary of the launch of the Science and Technology Innovation Board. Counting from April last year, Yuntian Lifei has been listed for more than a year. How do you feel standing at this point in time?

Deng Haoran:Going public has brought a significant improvement to the company's business development, which is first reflected in its financing capabilities. The financing capabilities of listed companies are very different from those of unlisted companies. Only with sufficient financial strength can an enterprise achieve rapid development, which is crucial.

At the same time, going public also has a great positive impact on brand awareness. The influence of a brand is not only reflected in customer awareness, but also in government awareness, employer brand attractiveness and other aspects, which can provide support for the development of the company.

Times Finance: As a company listed on the Science and Technology Innovation Board, how is Yuntianlifei prepared to respond to the country's call for the development of new quality productivity? What successful cases or milestone events does the company have in technological innovation?

Deng Haoran:We believe that the computing power operation business can be regarded as a typical representative of new quality productivity.

First, building a computing power operation center not only promotes the development of the chip industry, but also drives the development of various computing component industry chains such as CPU, GPU, electronic components, storage devices, as well as the transmission industry, communication industry, and power industry. In other words, this 1.6 billion yuan investment may bring economic benefits of 30 to 40 billion yuan or even larger.

Secondly, computing power operation business will also become the cornerstone of the development of advanced artificial intelligence technology, and will eventually be fed back to society in the form of software or cloud services, greatly improving work efficiency at the social level.

Finally, computing power operations can also drive investment at the entire social level. Investing in high-performance computing centers not only brings considerable returns, but also has a relatively predictable payback cycle, providing a new outlet for market funds.