Shang Tang’s headquarters building in caohejing, Shanghai (Source: edited by Titanium Media App)

Shang Tang is accelerating its transformation.

On the afternoon of March 26th, Shang Tang Group (0020.HK), the world’s leading artificial intelligence (AI) software company, released its financial report for fiscal year 2023 as of December 31st.

The financial report shows that in 2023, Shang Tang Group’s revenue reached 3.4 billion yuan, down 11% from 3.8 billion yuan in 2022; Gross profit reached 1.5 billion yuan, down 41% year-on-year; The Group’s EBITDA (loss) was RMB 5.4 billion, a year-on-year decrease of 2%; The main reason is that Shang Tang increased the transformation and capital investment of large-scale models and large-scale devices last year.

It is worth noting that the revenue business of Shang Tang Group has been redefined as three new business sectors-generative AI, traditional AI and smart car, from four sectors: smart city, smart business, smart car and smart life. Shang Tang said that the company takes generative AI business as its core development direction, while maintaining the advantages of traditional AI business and smart cars.

According to the financial report, in 2023, the revenue of Shang Tang Group’s generative AI business reached 1.2 billion yuan, soaring by 200% year-on-year, and the revenue proportion increased from 10% in 2022 to 35%, which is the fastest new business with a revenue volume of over 1 billion since the establishment of Shang Tang in the past ten years.

Dr. Li Xu, Chairman and CEO of Shang Tang Group, said, "Generative AI is not only a revolutionary innovation in the technical field for Shang Tang, but also a core business of the company. The growth of Shang Tang’s generative AI business benefits from the extensive demand for large-scale model training and reasoning from all walks of life, which indicates that a new cycle of hard technology investment in China has officially started. Shang Tang is winning new customers and promoting the overall improvement of efficiency and productivity by deeply integrating generative AI capabilities at all business levels. "

In the new era of AI 2.0, Shang Tang is accelerating its transformation and making a brand-new "Shang Tang 2.0" with the goal of General Artificial Intelligence (AGI).

The transformation and upgrading of Shang Tang AI 2.0 originated from a year ago.

In March 2023, Shang Tang announced that the company’s vision and strategic goal changed from "connecting the physical world and the digital world with artificial intelligence, promoting the sustainable development of social productive forces, and bringing people a better life experience of combining reality with reality" to "taking AGI as the core strategic goal, with a view to achieving a major breakthrough in AGI technology in the next few years".

Based on this change, Shangtang Technology established and implemented the development goal of AI for All, and focused on SenseCore Al as the core platform for large-scale model production, and Shang Tang AIDC artificial intelligence computing center platform as the AI infrastructure capacity, thus creating a multi-modal large-scale model with general capabilities and a professional large-scale model with more expertise in vertical industries, fundamentally reducing the downstream application cost and threshold of large-scale models, and ensuring that the development of AI technology benefits all mankind.

On April 10th last year, Shang Tang officially announced for the first time the "SenseNova" large-scale model system and the self-developed Chinese language large-scale model application platform "Consultation", with hundreds of billions of parameters, which can realize text generation, image generation, multimodal content generation and other capabilities and scene applications.

In July and August, 2023, the large model was upgraded to version 2.0 and 3.0, and the model version pages with different parameters such as "Consultation" version 2.0 and "SenseChat S" version of small model were released at the same time, which can adapt to the application requirements of different terminals and scenarios such as mobile terminals, and the total performance exceeds the basic model of GPT-3.5-Turbo.

Nissin V4.0 released in January this year, by improving the quality of training data and significantly improving the basic language ability, achieves the matching ability with GPT-4 in many scenarios such as code writing, data analysis, medical question and answer, and at the same time, opens up two parameter base models corresponding to 7B and 20B.

Li Xu said that under the guidance of Scaling Law, the big model is in the golden age of technological revolution and performance improvement. Since it was released in 2023, the ability of Shang Tang’s "Day by Day" big model has been significantly improved every three months, and Wanka Wanshen’s big model training ability has been realized, reaching the domestic leading level in basic model, multimodal, programming and tool calling, one million words lossless context and terminal small model.

At the product level, Shang Tang has launched the AI programming assistant "Code Little Raccoon" and the AI data analysis tool "Office Little Raccoon", which greatly improves the programming efficiency of programmers and the efficiency of office workers. At the same time, Shang Tang also invested thousands of calories to train the best model of Wensheng Diagram in China, Seconds Drawing 4.0. For example, the built-in "AI Chat Partner" in mobile QQ is the technical service provided by Shang Tang.

Specific to the performance level,According to the financial report, in 2023, based on Shang Tang’s core service leading ability of "big model+big device", the Group’s revenue of generative AI reached RMB 1.2 billion, achieving a rapid growth of 200%. At the same time, SenseCore, a large-scale infrastructure facility in Shang Tang, has finally achieved a breakthrough in scale growth to 12000 petaFLOPS, and the number of operating GPUs has reached 45,000.

Based on Shang Tang’s new "Model as a Service" (MaaS) business model, customers can easily fine-tune and call all kinds of generative AI capabilities in large devices without building and managing the underlying infrastructure themselves.

Shang Tang’s financial report shows that,More than 70% of the customers in the generative AI business are new customers in Shang Tang in the past 12 months, and the unit price of the remaining 30% of the existing customers has also recorded a growth rate of about 50%. Up to now, the number of customers with orders exceeding 10 million RMB has reached dozens, and the call volume of the C-terminal empowered by the large model has increased by nearly 120 times. Typical customers include the three major operators of China Mobile Telecom Unicom, China Merchants Bank, Haitong Securities, JD.COM, Xiaomi, Yuewen, Tsinghua University, Shanghai Jiaotong University and other leading enterprises and institutions in all walks of life.

Not only large models, but also large computing devices in Shang Tang are developing rapidly.In addition to Shanghai Lingang AIDC, which opened in 2022, the computing power managed by Shang Tang has realized the unified scheduling of national networking and supported the production of large-scale models with trillions of parameters, and has expanded new computing nodes in Shanghai, Shenzhen, Guangzhou, Fuzhou, Jinan and Chongqing.

Shang Tang revealed that it is expected that by 2024, the Group will expand the computing nodes in the whole country in a larger area, and export rich "new quality productivity".

"For large-scale model training services, the scale of GPU interconnection, acceleration efficiency and stability of infrastructure are the three most concerned capability indicators in the industry. We have achieved the largest cluster interconnection of Wanka, maintaining 95% acceleration efficiency, which is very leading in the industry. From the aspect of training stability, we have realized the ability of continuous training for more than 30 days, and the time of diagnosis and recovery in the event of training interruption has been optimized to half an hour. In the large-scale model reasoning service, we have also achieved the ultimate optimization of leading industry reasoning, and the cost-effectiveness of reasoning services has increased by three times in one year, providing customers with the most cost-effective large-scale elastic reasoning service in the industry. " Li Xu said that Shang Tang is increasing the adaptation of mainstream domestic chips such as Huawei Ascension and CAMBRIAN, so as to support the training, fine-tuning and reasoning of large models.

The management of Shang Tang Group pointed out, "From the perspective of business model, Shang Tang’s generative AI is currently promoted in a way of’ big device+big model’ deep collaboration, so there are generally three business models on the revenue side: first, public cloud standards, calling API；; The second is private cloud, which provides exclusive model and model authorization services for customers with security needs; The third is model customization service. "

In addition to the generative AI business, Shang Tang’s traditional AI and smart car businesses are also accelerating changes.

The financial report shows that in 2023, the overall revenue performance of traditional AI business was stable, with revenue of 3.4 billion yuan, down 11% year-on-year. The main reason was that the Group actively contracted the smart city business, and the proportion of this business revenue was reduced from 30% to less than 10%, significantly reducing the Group’s dependence on smart city business; The gross profit is 1.5 billion yuan, with a gross profit margin of 44%. In terms of smart terminal services, the number of mobile phones enabled by AI has exceeded 2 billion.

In terms of smart cars, Shang Tang’s "Jueying" smart car business earned 400 million yuan last year, a year-on-year increase of 31%. Among them, in the two core areas of intelligent driving and intelligent cockpit, sustained growth has been achieved, and the gross profit of mass-produced bicycles has doubled. In terms of mass production and delivery, the smart car sector added 1.29 million vehicles throughout the year, a substantial increase of 163% year-on-year. So far, the company has delivered 1.95 million smart cars, covering more than 90 models. In addition, the company added more than 16 million fixed-point orders and expanded 41 new models.

According to the management of Shang Tang Group, in 2023, Shang Tang "Jueying" will fully absorb the basic capabilities of large devices and large models, and provide various services at the infrastructure layer, model layer and application layer for automobile enterprises. The commercialization of Jueying is still based on "mass production" and realizes the common evolution of AGI capabilities with car companies. In addition, with the continuous advancement of testing, the Group expects that the capabilities of the large model will be applied to more production vehicles faster than expected, providing users with an unprecedented intelligent cockpit experience.

In addition, the management of Shang Tang said that Shang Tang will hold the "Shangtang Technology Exchange Day" in April this year, and launch the "Day by Day" big model 5.0 and a variety of generative AI products.

On the whole, Shang Tang’s business income in 2023 is still "stable as an old dog". From AI 1.0 to AI 2.0 era, based on the new engine of "big model+big device" with a growth rate of 200%, Shang Tang has realized transformation and upgrading, and continuously reduced costs and increased efficiency.

According to the financial report, in 2023, the total amount of R&D, management and sales in Shang Tang decreased by 11% year-on-year to 5.8 billion yuan. At the same time, the capital expenditure (excluding the capital investment related to disposable office buildings) also decreased by 20%, and the Group’s EBITDA (loss) decreased by 2% year-on-year.

Looking into the future, the management of Shang Tang Group said that the long-term growth rate of generative AI business in Shang Tang is very fast, and it is predicted that by 2024, the revenue of generative AI will be doubled (200%), accounting for about half (50%) of the company’s total revenue in fiscal year 2024-less than 35% last year.

"Starting from this year, the company (Shang Tang) has truly become a company with generative AI technology as its core. So in the long run, it is really difficult for us to predict very accurately. However, I think that if there is 50% contribution income this year, by 2027 and 2028, the proportion of overall income will increase to, for example, close to three quarters (75%). "

Shang Tang Group stated in the company’s strategic outlook that in 2024, Shang Tang will fully promote the profitability of its core business, so as to ensure the company’s dominant position in the competition in the AI 2.0 era and realize long-term leap-forward development.

Stay ahead of technology. Give further play to the deep synergy advantage of "big device+big model", continuously expand the scale of computing power, improve the comprehensive service capacity of infrastructure and big model, and continue to invest in the iteration of "day by day" model system, so as to keep the model capacity ahead of the industry. At the same time, the company will strive to realize the "Moore’s Law" to improve the efficiency of large-scale model reasoning, so that the cost of model reasoning can be rapidly reduced, thus opening up a broader application prospect;
Promote business growth. The company will accelerate the commercialization process and market penetration of generative artificial intelligence, and provide users with generative AI solutions with the best cost performance. At the same time, the company will also promote the steady growth of traditional AI business and improve the quality of cash flow;
Promote the profitability of core business. The company will continue to optimize the operational efficiency of the Group and concentrate resources on the generative AI business to improve cash flow and reduce losses.

For the multi-modal model level,Li Xu mentioned that Shang Tang has deep technical accumulation in the multi-modal direction, and has some important technical advantages in the data, model architecture and computing power to be solved in the process of AI image generation and video generation, especially the synthetic data will become an important advantage for the industry in the future.

Today, Shang Tang’s "growing day by day" model has established outstanding application value and leading position in key fields such as finance, medical care, Copilot assistant, anthropomorphic interaction and intelligent terminal.

Li Xu emphasized that the Shang Tang team also hopes to continuously promote the changes of the whole AI industry, especially the transformation in the AI 2.0 era.

As of press time, Shang Tang’s share price has dropped to HK$ 0.78/share, and its market value is about 20% at the time of listing, which is currently smaller than the company’s net assets. From the perspective of assets, Shang Tang’s share price has been seriously underestimated.

Whether it is the technical strength of Shang Tang’s big models and big devices, or Shang Tang’s 200% revenue growth in the generative AI business, it has become one of the "leaders" in the whole AI 2.0 era.

"Regarding the stock price and market value, the current stock price is about 20% of the listing period. There are many reasons behind it. On the one hand, the external environment of the whole market has changed, and the most important thing is to reduce the holdings of multiple old shareholders. There are some systematic reasons. At present, from the perspective of management, it is actually seriously underestimated. At present, a large number of assets are actually computing power and infrastructure and AIDC corresponding to computing power. Although they are reflected in assets in the form of depreciation, their value is actually higher. We hope that the adjustment last year and this year will make our growth business have a very good development state to meet the future challenges, and also extend our commercialization ability to 2.0, and let the healthy and stable cash flow of the business meet the new AI 2.0 era. " Li Xu said.

"From the management point of view, we also hope to make use of our many efforts, including our management locking in our own stocks, and we will also promote the rise of stock prices to repay our investors, our shareholders and our employees." Li Xu said.

(This article is the first titanium media App, author | Lin Zhijia)

Reporting/feedback