On August 27th, leading AI company SenseTime (00020) released its interim performance report for the first half of 2024 fiscal year ending June 30th. The data shows that SenseTime Group achieved operating revenue of 1.74 billion yuan in the first half of the year, a year-on-year increase of 21%, demonstrating strong growth momentum.

Of particular note is the significant increase in revenue from the generative AI business, which has grown by 256% to nearly 1.1 billion yuan, accounting for a high proportion of 60% of the total revenue, and has become the core engine driving the company's growth. This performance not only demonstrates SenseTime's deep accumulation in the AI field, but also signifies the significant effectiveness of its strategic transformation.

SenseTime Group Chairman and CEO Xu Li said, "The interim performance reflects that SenseTime's strategic goal of fully transitioning and upgrading to generative AI has been partially achieved."

In Xu Li's view, the core advantage of SenseTime's growth in generative AI business lies in the deep synergy of "large-scale equipment + large-scale models", which has the ability to build first-class large models, and has unique advantages in native multimodal, video stream natural interaction, and low-cost model inference architecture.

In terms of profitability, SenseTime achieved a gross profit of nearly 0.8 billion yuan in the first half of the year, a year-on-year increase of 18%, and maintained a steady gross margin of 44%. The net loss narrowed by 21% year-on-year, and the EBITDA loss narrowed by 26.5% year-on-year, sending a strong positive signal to the market - SenseTime is gradually emerging from the shadow of loss and steadily moving towards profitability.

Leading the future with technology: the core reason for the rapid surge in the generative AI business

The growth of SenseTime's performance in the first half of the year is mainly due to the explosive growth of its Generative AI business.

The soaring of SenseTime's Generative AI business is not accidental, but the inevitable result of its long-term technological accumulation and forward-looking layout. Looking back at SenseTime's development process, it is found that its persistence and investment in the 'large equipment + large model' strategy are the key factors that enable it to maintain a leading position in the field of AI.

Through deep collaboration between its large-scale computing infrastructure (large equipment) and advanced large model product (SenseNova), SenseTime has achieved a leap from laboratory technology to large-scale commercial application.

Specifically, SenseTime's large equipment SenseCore provides powerful computing support, enabling the company to support training of multimodal models with trillion-scale parameters and large-scale AI inference. In addition, SenseTime's large model system 'SenseNova' iteratively upgrades, significantly enhancing its comprehensive capabilities.

During the reporting period, SenseTime's large equipment intelligent computing services achieved significant scaling, with the total operating computing power exceeding 20,000 PetaFLOPS, an increase of approximately 70% compared to March 2024; the number of GPU deployed reached 0.054 million, ranking stably in the industry. The computing power managed by SenseTime's large equipment has achieved unified scheduling across the country, and new computing nodes have been expanded in Shanghai, Shenzhen, Guangzhou, Fuzhou, Jinan, Chongqing, and other places. Among them, Shanghai Lingang AIDC has reached 14,000 petaFLOPS, far exceeding the projected total computing power of 3,740 petaFLOPS. While supporting the development of SenseTime's own large models, it also supports the training and deployment of large models for external customers. The powerful computing power can support the simultaneous training of more than 20 trillion super-large models and the full lifecycle generation of trillion-parameter large models.

The company has the design, construction, and operation capabilities from kilo-card cluster to million-card cluster, which can efficiently support the training and inference of MoE multimodal models with up to 2 trillion parameters. In addition, through technological innovation, SenseTime has achieved elastic on-demand scaling of inference services, significantly reducing the overall cost of large-scale AI inference and improving the overall operational ROI of its Intelligent Computing Center.

Empowered by large equipment, SenseTime's generative AI technology is developing rapidly. The SenseNova large model system, released in April 2023, iterated to version 5.0 within a year, and became the first model to surpass GPT-4 Turbo in the SuperCLUE evaluation. It has demonstrated outstanding performance in traditional scenarios such as code writing, data analysis, and medical Q&A, and has gradually expanded its application boundaries to emerging fields such as electric vehicles, finance, and robotics. These technological advantages and leading edge in implementation have not only won SenseTime market favor, but also laid a solid foundation for its future development.

Differentiation advantage: Leading technology, driving easy implementation of applications for customers.

In today's fierce competition of large-scale models, differentiation becomes the key for AI enterprises to stand out. SenseTime is well aware of this, so it continues to enhance its inferential and multimodal capabilities based on large-scale models and devices.

In the inference scenario, SenseTime's innovative technological architecture enables the same computing power and electrical cost to achieve a 4-fold increase in inference concurrency (QPS) and elastic on-demand scaling of inference services, significantly optimizing the overall cost of large-scale AI inference. In addition, by continuously improving GPU resource utilization rate to more than 80% in the Wanka cluster through original technology innovations such as train-inference integration and algorithm-electricity coordination, SenseTime enhances the overall operational return on investment of its intelligence computing center.

In terms of multimodal capabilities, SenseTime has achieved seamless integration and interaction of various data types such as text, images, and audio. After the upgrade to version 5.5 in July 2024, SenseTime's comprehensive capabilities have increased by approximately 30% compared to version 5.0, and its multimodal capabilities have been significantly strengthened, achieving real-time interactive experience comparable to GPT-40 and becoming China's first multimodal real-time interactive large-scale model.

Multimodal capabilities not only greatly broaden the application scenarios of large-scale models, but also enable AI systems to comprehensively and accurately understand human needs and intentions. For example, in the medical field, SenseTime's large-scale models can provide more accurate diagnostic suggestions for doctors based on comprehensive analysis of patient records and examination reports; in the financial field, they can provide more scientific investment advice for investors through in-depth mining and analysis of market data.

In the field of smart automobiles, SenseTime's smart automobile business achieved a revenue of 0.17 billion yuan, a year-on-year growth of 100%, with the addition of 15 designated vehicle models and approximately 6 million designated vehicles. With the continuous expansion of the smart automobile market, SenseTime's smart automobile business is expected to continue to grow rapidly.

In the field of autonomous driving, Jueying is the earliest Chinese company to take the pure vision large-scale model route. In April of this year, at the Beijing Auto Show, it completed full autonomous driving without relying on high-definition maps and relying solely on visual perception. In the field of smart cabins, relying on the Wanka 5.5 large-scale model and multimodal capabilities, collaborations with well-known enterprise brands have been established, such as in smart voice interaction, instant image generation, AI diagnosis, etc., creating a more comfortable and safe driving experience.

So far, more than 3,000 leading companies in various industries, including internet, smart hardware, new energy vehicles, robotics, medical, and finance, have used SenseTime's large-scale models and intelligent computing services. SenseTime has also established deep collaborations with well-known domestic and foreign automotive companies such as Honda, BYD, and Great Wall, promoting the widespread application of AI large-scale models in smart cabins and autonomous driving.

The overlooked value: High potential behind undervalued stocks.

Compared to foreign AI innovation companies, SenseTime has demonstrated considerable competitiveness and growth potential in multiple fields.

In terms of product technology, SenseTime's humanoid interaction business, compared to Character.ai, enterprise-level services compared to Cohere, instant drawing compared to Midjourney, multimodal large model compared to ChatGPT, all demonstrate differentiated value and advantages. In terms of market share and customer resources, SenseTime has not only achieved breakthroughs, but also has strong growth potential.

However, despite SenseTime's significant achievements in technology and the market, its market cap is undervalued. This may be due to factors such as market sentiment and industry perception bias. But with the continuous innovation of the company's technology and the continued expansion of the market, SenseTime's market cap is expected to be reassessed.

The market has given positive feedback on SenseTime's performance. Many brokerage institutions, including CICC, Galaxy, CITIC, and Haitong, have given positive evaluations of SenseTime's future development prospects and have provided target prices. For example, CICC has set a target price of HK$2, corresponding to forecasted P/S multiples of 14 times and 10 times for the current and next year.

In the future, with technological breakthroughs, mature product expansion, and important customer collaborations, investor attention and recognition of SenseTime are expected to increase. After all, as an excellent investment target, SenseTime still has great potential for market cap growth.

Conclusion

According to an IDC report, China's intelligent computing service market will continue to grow at a high rate in the next five years, with an expected average annual compound growth rate of 57.3%. By 2028, the overall market size is expected to reach nearly 200 billion RMB. As a leading company in the industry with a forward-looking layout, SenseTime will benefit from the rapid market growth.

With the continuous maturity of generative AI technology and the continuous growth of market demand, SenseTime is expected to accelerate its expansion into profitability in the next few years. In order to achieve this goal, SenseTime will continue to increase its investment in technology research and development and market expansion, consolidating and expanding its leading position in the field of large models and large devices.

At the same time, SenseTime will actively explore new application scenarios and business models to promote the popularization and application of AI technology, which will further expand SenseTime's business and industry boundaries.