#精品长文创作季#
Author: Yuntan, Editor: Xiaoshi Mei
"Generative AI business revenue reached 1.2 billion, a year-on-year increase of 200%, accounting for 35% of the group's total revenue. "After the glory of the AI1.0 era, SenseTime has grown in the AI2.0 era.
Due to industry difficulties such as heavy R&D and difficult commercialization, former AI head companies have encountered dual pressure on stock prices and performance. Now, in the new round of prosperity cycle opened by the large model, SenseTime took the lead in breaking through and commercialization has been fruitful.
Realize that the past is not admonished, and those who know the future can be traced. SenseTime set sail again.
【Soaring Generative AI Business】
On March 26, SenseTime announced its 2023 annual results report.
Among them, the most notable is the generative AI business, which achieved a rapid growth of 200% year-on-year, with annual revenue reaching 1.2 billion yuan, accounting for 35% of the group's total revenue.
For SenseTime, this is a qualitative change.
This means that generative AI is not only a transformative technological innovation, but also a pillar of the company's business. At the same time, it also represents that SenseTime has achieved a breakthrough in the AI 2.0 era through the accumulation of the AI 1.0 era.
Generative AI is also a new business that SenseTime has achieved from scratch at the fastest speed since its establishment ten years ago, with a revenue volume of more than 1 billion yuan. It shows the huge momentum of the new AI track.
In the 2023 financial report, SenseTime took the initiative to adjust its business segments, clarifying the three major sectors of generative AI, traditional AI and smart cars, with generative AI as the core.
The above three segments correspond to generative AI model training, inference and other services, traditional smart cities, smart business and smart life, non-generative AI business, and the original smart car business.
At the same time, relying on the infrastructure and R&D capabilities of large models, the Jieying intelligent car business has gained a lot of money. In the past year, the smart car business achieved revenue of 384 million yuan, a year-on-year increase of 31.1%, and the proportion increased from 7.7% to 11.3%.
SenseTime added 1.29 million vehicles to mass production and delivery, a year-on-year increase of 163%, and so far, the company has delivered a total of 1.95 million smart cars. Owners of brands such as BYD, NIO, Aion, Zeekr, and Audi will enjoy the convenience brought by SenseTime's AI technology.
Moreover, SenseTime has also added more than 16 million fixed-point orders, which will be converted into solid performance in the future.
The substantial growth of the two businesses has greatly restructured SenseTime's business structure in 2023, especially the dependence on "smart city" has been significantly reduced, which has been reduced to less than 10%, and the effect of business adjustment is significant.
In addition, the company will pay more attention to the improvement of financial efficiency in 2023, and its operating expenses will decrease, with a collection of 3.9 billion yuan, a year-on-year increase of 49%; EBITDA (before interest, taxes, depreciation and amortization) loss narrowed to $5.4 billion, down 2% year-on-year.
As Xu Li, chairman of the company, said: "The growth of the generative AI business is due to the extensive demand for large model training and inference in all walks of life, which indicates the official start of a new cycle of hard technology investment in China." ”
In this new boom cycle, SenseTime's generative AI continues to explode. At the results conference, SenseTime's management said that generative AI is expected to account for about 50% in 2024, becoming an absolute core growth driver.
This also means that SenseTime is about to usher in an inflection point in value revaluation.
[New Business Model of Generative AI]
SenseTime, which has re-embarked on the road, has significantly accelerated its commercialization.
In April 2023, SenseTime officially launched the "RiRixin SenseNova" large-scale model system, and in less than a year, SenseTime has received many olive branches from the B-end and C-end.
In the past year, 70% of the customers were new customers, and 30% of the existing customers also achieved a 50% increase in the unit price, which means that the volume and price have risen together, and the commercial flywheel is running rapidly.
At present, SenseTime has dozens of customers with orders worth more than 10 million yuan, including the three major operators in the field of telecommunications, financial institutions such as China Merchants Bank and Haitong Securities, as well as leading Internet companies such as JD.com, Xiaomi and China Literature, as well as Tsinghua University, Shanghai Jiaotong University and other institutions of higher learning.
In the past six months, the number of C-end calls of large models has increased by nearly 120 times, and the number of mobile phones empowered has exceeded 2 billion.
These achievements are driven by the company's model-as-a-service business model, which enables customers to easily fine-tune and invoke generative AI capabilities at large scales without having to build and manage the underlying infrastructure themselves. Drive the surging growth of generative AI business.
For example, the "daily new" model empowers the national office software Kingsoft WPS's Copilot, which makes it easy to generate PPT, form processing, and PDF documents with one click, so that WPS AI can benchmark Microsoft 365 Copilot. The enterprise Copilot assistant has also been adopted by large enterprises such as Haitong Securities and China Merchants Group.
The "100 model war" has entered a white heat, but based on the deep cultivation of the AI industry, SenseTime knows what cards to play in this track. With insight into the downstream application market, SenseTime took the initiative to bid farewell to the limited thinking of "bigger is better" and built a number of advanced small models.
According to the implementation of the "industry small model" in different scenarios, SenseTime makes the large model smaller, so that the large model can become a tool that can be used by small and medium-sized enterprises. For example, thanks to the GPU computing power of SenseTime's large device, users can use SenseTime's self-developed painting models to efficiently generate content, or import local models and third-party open-source models with one click to generate more diverse content.
In this way, the capabilities of large models can be adapted to vertical fields, lower the threshold for users to use, truly empower society with technology, and turn AI into a useful productivity tool.
SenseTime has carried out a forward-looking layout driven by the synergy of "large device + large model", which has become the driving force for the company's growth.
SenseCore is the infrastructure of large models, which can be understood as the basic platform for AI development, like an AI "power station", providing the "power" required for various AI applications.
Based on the AI infrastructure of large devices, SenseTime has been trained and optimized, and SenseTime's "daily new" large model system has been iterated at a high speed.
By the end of 2023, the total computing power of SenseTime's large devices will reach 12,000 petaFLOPS, and the number of operating GPUs will reach 45,000 cards, which will be the first echelon in China.
▲Source: the official website of the State-owned Assets Supervision and Administration Commission, the official website of the International Science and Technology Innovation Center, and the company's announcement
More importantly, the cost performance of its inference service has increased by 3 times within one year, and it is regarded as the most cost-effective large-scale inference service in the industry.
Thanks to the powerful computing power reserve of large devices, Ririxin version 4.0 will be launched in February 2024, which is comparable to GPT-4 in multiple scenarios such as code writing, data analysis, and medical Q&A.
It can be seen that through the technical advantages and hard technology thresholds, the strategic deployment of "large device + large model" has opened up a closed business loop for SenseTime. In the era of AI 2.0, it has embarked on a unique development path and has shown vigorous vitality.
[Focus on creating new quality productivity]
After ChatGPT set off a huge wave of AI transformation, China cannot afford to lag behind in the AI track.
Whether it is the top-level design or the level of enterprise business competition, Chinese enterprises have been catching up.
At the 2024 Global Developer Pioneer Conference (GDC), Xu Li pointed out: "In the era of AI 2.0, generative AI is regarded as an important technology to promote the progress of productivity. ”
In this change, SenseTime has been at the forefront of the industry, which stems from the following three dimensions:
First, the advantages accumulated by AI 1.0 are a huge gold mine, laying the foundation for the accumulation and development of the 2.0 era.
The era of AI 1.0 roughly began in 2015, when the wave of AI entrepreneurship was dominated by CV computer vision technology.
AI1.0 is mainly perceptual intelligence and decision-making intelligence, and the downstream mainly focuses on the To G or To B market, and the commercial application space is relatively limited. However, the accumulation of technology and applications in the 1.0 era is crucial, otherwise such a rapid leap cannot be achieved.
Since the beginning of the AI 1.0 era, SenseTime has accumulated a large number of high-quality customers. For example, in 2015, Xiaomi launched the baby album in the operating system MIUI7, which is a new selling point jointly created by SenseTime. In the same year, SenseTime won the bid for China Mobile's real-name authentication project.
Coupled with the accumulation of smart city and smart business fields, SenseTime's unique insights and insights are now accelerating the transformation of these customers into generative AI customers.
Second, rapid iteration and continuous improvement of performance.
With the support of super computing power, in February this year, "Ririxin" was iterated to version 4.0, and it is expected to be released in April, with performance fully benchmarked against GPT4-Turbo, supporting millions of words of lossless context, and multimodal model capabilities will also be fully benchmarked against GPT4V.
In the code evaluation held by SuperClue, the "Daily New" large model topped the list, with a first-time pass rate of 76% in the HumanEval test set, which even surpassed GPT-4.
According to the "2023 China AI Development Platform Market Report" released by Frost & Sullivan, SenseTime ranks first in China in the comprehensive score of the two evaluation dimensions of growth index and innovation index, second only to Amazon.
SenseTime has also open-sourced InternLM2-7B and 20B base models, which surpass Meta's Llama2 and Google's Gemma in terms of performance.
Third, have a deeper understanding of the industry and market, and have stronger product strength.
Facing multiple application scenarios, SenseTime has been deeply engaged for many years, and can quickly follow up with market demand and provide targeted generative AI solutions.
Little Raccoon Intelligent Assistant is a popular product in the current SenseTime product system, including Office Little Raccoon and Code Little Raccoon. Code Raccoon can greatly improve the programming efficiency of programmers, such as the "Little Raccoon" code subsidy tool, which can save 30% of the workload.
For example, when a company needs to make a budget for the promotion of a product, it only needs to input all the documents such as financial statements, account information, and product introductions, and it can give a reasonable and scientific budget plan according to the existing data and needs, showing strong reasoning ability.
Fourth, the long-term accumulation of end-to-end AI allows SenseTime to penetrate into the AI business market.
As early as 2015, SenseTime has cooperated with leading mobile phone manufacturers and mainstream car companies to implant AI models for perception and decision-making into more than 2 billion mobile phones and millions of cars. In the era of generative AI, SenseTime has taken the lead and launched a leading device-side small model, opening up new application scenarios.
If 2023 is the first year of AIGC, 2024 is the year of the explosion of device-side large model applications. In particular, Apple, which missed out on AI, turned to AI research and development after giving up car manufacturing, and is said to be cooperating with domestic companies to launch AI mobile phones, which are used in the iPhone 16, Mac system and iOS 18 released this year.
Think about it, if the world's smartphones were integrated with AI technology, what a huge market it would be?
In 2023, both Qualcomm and MTK demonstrated the application effect of SenseTime's generative AI device-side model on their flagship chips, and the SenseTime 7B small model reached an inference speed of 16 tokens/sec on Qualcomm chips.
These terminal models are a key step in the implementation of AI mobile phones, which can finally invoke various applications to complete complex tasks after listening to user instructions. At present, SenseTime has cooperated with Xiaomi, Honor and other mobile phone companies in research and development.
In the face of such a vast market, its imagination space is predictable.
In the past, SenseTime's impression in the capital market was that R&D and commercialization were still in the empowerment stage, so the stock price was suppressed.
Based on the closing price on April 2, SenseTime PB is only 0.96, the market value falls below the net assets, and the PS used by growth stocks is only 7 times. It is unthinkable in any mainstream capital market to give such a low valuation to a growth-stage technology stock. Moreover, it is still in the hottest AI track at the moment.
SenseTime has strategically anchored generative AI, showing explosive growth, and its overall operating efficiency has improved significantly, gradually entering a positive cycle of commercialization.
It's time to revalue SenseTime.
disclaimer
The content of this article related to listed companies is the author's personal analysis and judgment based on the information publicly disclosed by listed companies in accordance with their legal obligations (including but not limited to temporary announcements, periodic reports and official interactive platforms, etc.), and the information or opinions in this article do not constitute any investment or other business advice, and Market Value Watch does not assume any responsibility for any actions arising from the adoption of this article.
——END——