A：The total length of the roadshow and investor exchange for Paradigm Intelligence is 35 to 40 minutes. I am Zhang Mingxiao, an investor in Fan Shi Intelligence, and I am in charge of today's roadshow exchange.

A：Recently, there have been a lot of roadshow exchanges, and the stock price has performed poorly, and it has received greater attention and pressure. I feel a lot of pressure, but I will also be nervous, but we have made careful preparations and hope to share the company's latest business, technology and business model progress.

A：Last year, the company's total revenue reached 7.1 billion, the first full-year profit, in hand orders 9 billion. At the same time, it has maintained the first market share in the ML platform for eight consecutive years. At present, there are three main parts classified by charging mode: AI platform (revenue 6.5 billion, up 32% year-on-year), pay-as-you-go business (revenue 0.5 billion) and API business (revenue 80 million, up 120% year-on-year).

A：Total revenue rose 35 per cent in the first quarter from a year earlier, with gross margins of 35.1 per cent, up from 34.8 per cent last year. In addition, the computing resources have been improved, the self-sustaining computing power has been increased, and a major breakthrough has been made in the business of the relief factory. A framework agreement of 0.2 billion US dollars for three years has been signed with Huanxi Media.

A：The AI platform business maintained rapid growth and led to overall growth, with gross profit margins rebounding; pay-as-you-go orders grew significantly; API business grew the strongest, up 120 percent year-on-year, and is expected to grow tenfold this year.

A：Paradigm Intelligence is solving the pain points and difficulties of enterprise deployment AI through a standardized platform, especially to improve computing power utilization and resource scheduling efficiency, and is committed to building a heterogeneous computing power intelligent butler system. The company's long-term research and development of technology as a heterogeneous computing power intelligent steward, through the cloud-native computing virtualization middle layer to improve GPU utilization, has become the CNCF official sandbox project, has become one of the global organization's standard-setters.

A：The two major challenges facing China's deployment of computing power are insufficient utilization of computing power and fragmentation of computing power. The problem of low GPU utilization is common, and the heterogeneous computing environment makes this challenge even more difficult.

A：GPU virtualization technology (VGPU) is mainly used to finely divide GPU computing power with an accuracy of 1%, for example, the computing power of a card can be cut into 100 parts. Through this technology, the original situation that multiple tasks occupy multiple cards is optimized, and only fewer card resources are needed to complete the tasks, thereby improving efficiency, reducing costs, saving half of the cost or achieving cost reduction.

A：This technology not only improves the efficiency of domestic AI chips, but also significantly improves resource utilization, up to 5 to 10 times, and greatly reduces hardware costs. At the same time, due to its cloud-native characteristics, it can reduce operation and maintenance costs and realize cloud manual operation and maintenance.

A：The open source version focuses on the promotion and development of technologies in the community, while the commercial version includes exclusive technologies such as vRAM management and power isolation. Among them, vRAM technology can enhance the virtualization of display memory and enhance computing power, especially for domestic chips to fill the performance gap when running efficient models, reduce the demand for multiple cards and reduce costs. The open source community has gathered many companies and developers to form a strong ecology, making Hami VGPU one of the most influential companies in the world in the field of hardware management. Attracting more participants through open source, after the technology takes advantage, we will be the only winner in the industry.

A：100 or 200 large domestic enterprises have adopted our commercial products, including well-known enterprises such as Baidu and Huawei. They may need to use or adjust the commercial version technology provided by us due to their own technical requirements or compliance requirements.

A：After Huawei adopted our technology, it successfully realized the unified management of domestic GPU, reduced the platform cost, improved the utilization rate, and also met the requirements of domestic Xinchuang compliance supply chain security. Huawei cannot solve the problem of unified management of equipment from multiple vendors, and we achieve this goal through our unique ecological location and technological advantages.

A：This technology was originally developed to improve the utilization rate of the enterprise's existing GPU, and gradually developed into an industry pain point solution. Through continuous technology research and development and open source, it finally reached its current market position and capability.

A：Nvidia's model is already adapted at the time of coding, but domestic chips are not directly compatible. Domestic chip manufacturers are trying to do some adaptation on their own or through the community, but the number of adaptation models available is relatively small, compared to the hundreds of models of Nvidia, the number of adaptation models of domestic chips is still very limited, we do this work to adapt to more models. At present, the second quarter is close to the 100000's adaptation volume, reaching this year's target ahead of schedule, and is confident to reach the same target next year.

A：The goal of our open source community is to reach 100000 adaptation models this year and 1 million next year to match the ecology of heavy face and make domestic chips better adapted.

A：Hami VGPU technology can unify the management and cutting of computing power, just like a house that can only live for one person can now be allocated to more users or tasks, thus improving efficiency.

A：Xinchuang running-in mainly solves two problems: one is to solve the problem of adaptation between domestic chips and various models, that is, whether they can be used; the other is to solve the problem of how to make better use of chips to make them run faster. Xinchuang Magic Box helps customers adapt external open source models, build private models, deploy computing power and improve the utilization rate of computing power resources, and build customers' private clouds. At the same time, public cloud provides services to small and medium-sized customers and companies in need, supporting them to call or deploy intelligent bodies to meet the AI capability needs in different scenarios.

A：The Prophet model is a link or module in the entire set of AI operating systems provided by the company to enterprise customers, and its role is to help enterprises develop and train their own private models, rather than directly providing vertical models.

A：Enterprises own private models and deploy them locally to meet the needs of data security and compliance risk control, and build exclusive models for their own business flows and application scenarios, and run their own data optimization and iterative models to assist in decision-making.

A：By using our fancy cloud products, Litong Electronics has successfully transformed from an electronic component manufacturer to IDC business, and has solved the problem that the bare metal cards it bought can only be piled up and cannot be effectively used.

A：By providing fancy cloud capability, we provide customers with unified management and scheduling of GPU, and optimize the model to improve the running speed. In this way, customers can use our technical capabilities to transform it into a UI cloud role, effectively solving the problem of final accounting power acquisition and optimization.

A：This is because with the rise of the lobster intelligent ecological model, the demand for tokens has increased exponentially, which has completely opened the ceiling of the market. After several months of evaluation, the management believes that the company has the ability to seize this opportunity, through self-built computing power and the use of its own infrastructure capabilities, to achieve end-to-end access from the source to the final generation of token.

A：Transparent factory business is to transform the computing equipment of the physical world into virtual form, and through additional models and other technologies into model inference call services in the virtual world. At present, there are mainly three business forms: first, large model companies output token through leasing or self-built computing power; The second is that traditional cloud vendors cooperate with model vendors to convert token after grafting or accessing the model. The third is the open API transfer station, which does not directly own the computing power and model, but distributes it through the aggregation API mode and collects from it.

A：The company has strong supply chain capabilities and maintains good cooperative relations with Huawei, domestic chip manufacturers and other startups. For example, Dr. Dai, chairman of the board of directors, has worked with Huawei and has a deep understanding of Huawei Shengteng series. At the same time, the company has reduced costs, improved efficiency and adapted more models through Hami VGPU and shape running-in technology, thus serving both ends of public cloud and private cloud, helping enterprises and developers to improve efficiency and integrate workflows.

A：The projection factory business is similar to that of companies such as Nebula, which superimpose various value-added services on the basis of their own computing power to solve the production and operation problems of hundreds of thousands of industries. Unlike AWS, which only provides pure computing power, we not only provide computing power, but also enrich our capabilities through self-research and acquisition, directly providing managed services to customers to meet different types of customer needs, including long-term contracts, instant pricing, and various charging models such as to B and to C.

A：our cooperation with joyful media adopts the real-time pricing mode. by using our fancy movie products to create movies and TV dramas, customers pay in advance and deduct fees according to the traffic token. In this mode, we do not lock in a specific number of servers. As the technology iterates, even if only 800 servers are needed in the future, the remaining 200 servers can be provided to other customers, thus helping to improve the gross profit margin in the future.

A：The sharp decline in the number of employees is mainly due to the shift of the company's product system to a standardized platform model, the reduction of outsourcing personnel such as front-end engineers in the factory, and the use of digital employees to improve research and development efficiency, resulting in a more than tenfold increase in code output efficiency. Research and development personnel are now more inclined to adjust the structure, by leading the project, master the algorithm modeling and other aspects of the engineer to lead the team work, reduce the proportion of basic code writers.

A：In recent years, the company's revenue has maintained a relatively high growth, turning losses into profits for the first time, and achieving profitability at the operating level. The increase in cash and cash equivalents ensured that debt funds and liquidity were used for expansion calculations. At the same time, hand orders have increased compared to last year's revenue, ensuring the visibility of revenue growth this year.

A：The founder of the company, Dr. Dai, served as the chairman and secretary-general of the Artificial Intelligence Committee of the All-China Federation of Industry and Commerce, and won the national-level important award for outstanding builders of socialism with Chinese characteristics for non-public economic people. This award is selected every five years by the United Front Work Department of the People's Republic of China and five ministries and commissions. It is a high recognition of the contribution of private entrepreneurs in socialist construction.

A：The customer categories of token factory business mainly include three categories: the first category is the company's former large and medium-sized customers; The second category is sold on third-party aggregation platforms such as open node, mainly small developer groups; The third category is a new mode in which the company directly signs contracts with enterprises and outputs token, such as Happy Media.

A：In tests, the company found that users can help small studios, self-media or artist companies create economic benefits more efficiently when using products such as lobster farms. For example, through automatic collection of popular material, writing and publishing novels, interaction, and even the application of video generation models, these can bring real benefits to users. In addition, the company also combines models with other links to provide one-stop service for large-scale film production, which reflects the value of token in social and economic activities.