Break and Rebuild

Editor: Chu Yi

Feng Pin: Dapeng

Source: Lao Cai - Lao Cai Research Institute

The warmth of the 10th anniversary celebration still lingers, but the company seems to have felt the chill of layoffs both inside and outside.

On October 22, 2024, SenseTime celebrated its 10th anniversary. Chairman Xu Li mentioned the newly established trinity strategy of 'Large Devices - Large Models - Applications' in an internal letter to all employees for the first time and announced the acceleration of organizational and management streamlining. He added, 'We will build a more centralized and efficient organizational structure to promote the centralized and intensive investment of resources.'

This adjustment has also become SenseTime's unified propaganda line. According to Jiemian News, a recently unnamed employee revealed that although the overall scale of layoffs is unknown, independent research institutes, autonomous driving, security, and other business units are all affected; some departments have even seen 'the entire third-tier department eliminated,' with affected employees receiving 'N+1' months' salary compensation. Among them, the RAG (Retrieval Augmented Generation) technology-related personnel plan to reduce staff by 50%.

It is said that every ten years is a cycle. Standing at a new development node, it is understandable to have a new strategic image. However, why such a sharp turn inward? What signal does it convey? And how much of the accumulated problems can it resolve, truly emerging from the quagmire of loss?

01

Why Layoffs Again?

It's not too much to ask. According to media reports citing former SenseTime employees, besides the four major business units of Smart Living and Smart Auto, SenseTime initially also had independent 'Large Devices' and 'Research Institute' departments and recently added a dedicated 'Large Models' department.

However, this round of layoffs does not involve the Large Models and Large Devices departments. The Research Institute, Intelligent Driving, and other business scenario-oriented units have been hit hardest.

In response, SenseTime told CNR that it was optimizing and adjusting its organization and talent structure to better meet business development needs. At the same time, SenseTime's social and campus recruitments are proceeding normally, and the overall business is developing steadily.

Although that's the case, some public opinions are still unsatisfied. After all, this is not the first time SenseTime has been embroiled in layoffs. There were related media reports in June 2022, April 2023, and August 2023. There is also a decrease in staff size in the financial reports.

At the end of 2021, SenseTime had 6,113 employees, reaching a peak in public data. By the end of 2022 and 2023, the numbers were 5,098 and 4,531, respectively. A total of 1,582 employees were laid off in two years. The latest financial report shows that there were 4,672 employees in the first half of 2024, an increase from the end of 2023 but still a 6.9% decrease compared to the same period in 2023. Meanwhile, sales and administrative expenses decreased by RMB 418 million and RMB 735 million, respectively, mainly due to 'reduced employee benefit expenses.'

What signal does it convey? In fact, looking back at the 10th-anniversary internal letter, 'SenseTime's 10th Anniversary: Focus, Integration of Knowledge and Action,' it already shows a strong intention of organizational adjustment and strategic shift.

Xu Li mentioned in the letter that SenseTime looks forward to the era of general artificial intelligence and has gone through the two stages of traditional AI 1.0 and generative large model AI 2.0, namely specialized intelligence and general intelligence.

SenseTime believes that a key difference between the two is the change in AI cost structure. The former's main cost lies in developer investment, while the latter invests in computing resources. As scaling laws are validated in large language models, multimodal models, video generation models, and slow-thinking reasoning processes, the cost of producing and using large models can be directly equated to the consumption of computing resources.

It can be seen that SenseTime is making 'Large Devices - Large Models - Applications' the core of its strategic transformation, leading to the pain of layoffs. That said, change is inevitable. Well-known economist Pan Helin believes that the current AI industry is indeed facing a profit dilemma. With the Hundred Models War intensifying competition among AI companies, many AI companies have yet to find a breakthrough in profitability and are forced to make long-term plans through staff adjustments.

02

Behind the Unappealing Stock Price

Losses Exceeding RMB 49 Billion in Five and a Half Years

Indeed, regardless of the ultimate success or failure of this transformation, SenseTime, which has long been burdened by losses, urgently needs to find a breakthrough.

Public information shows that SenseTime was founded in 2014. Its founding team originated from the Multimedia Laboratory of The Chinese University of Hong Kong, established in Hong Kong in 2001. Core members include Professor Tang Xiao'ou and other key members of the laboratory. Thanks to its strong research and development background, the company has been favored by capital since its inception, completing a total of 12 rounds of funding before going public, with a cumulative funding amount exceeding USD 5.2 billion. Investors include Alibaba, SoftBank, Suning.com, Wanda Group, and others.

On December 30, 2021, SenseTime officially went public with an issue price of HKD 3.85. As the first AI stock on the Hong Kong stock market, its market value once reached a high of HKD 320 billion. However, it then began to fluctuate and decline, reaching HKD 1.56 per share as of October 31, 2024, with a market value of HKD 54.82 billion, a decrease of over HKD 250 billion from its peak.

As the stock price cooled, some early investors who were optimistic about SenseTime began to withdraw, with Alibaba, SoftBank, Global Capital, and others reducing their holdings. During the previous large-scale share unlocking period, public opinion analysis pointed out that these institutional investors participated in investments at a lower valuation during the early financing stage and gained significant profit space after the IPO, so they had a strong willingness to cash out after the unlocking period. Considering the fundamentals of SenseTime and the general decline in the valuation of global AI companies, it is not surprising that investors have left one after another.

Adjusting strategies based on expectations is normal for capital. According to Frost & Sullivan's report, the market size of China's computer vision software increased from RMB 7.9 billion in 2018 to RMB 16.7 billion in 2020 and is expected to reach RMB 101.7 billion by 2025, with a compound annual growth rate of 43.5%.

Focusing on SenseTime, its revenue in 2020 already ranked first in the Asian industry. From 2019 to 2021, the company's revenue was RMB 3.027 billion, RMB 3.446 billion, and RMB 4.7 billion, with corresponding growth rates of 63.3%, 13.86%, and 36.39%. It has strong growth potential in scale. As a leading enterprise in a popular sector, it is naturally sought after by capital.

However, after going public, SenseTime's performance has been less impressive. From 2022 to 2023, revenue decreased by 18.97% and 10.57% to RMB 3.809 billion and RMB 3.406 billion, respectively. This significant change in growth potential has somewhat disappointed expectations, especially considering the basic investment logic of 'strategic losses,' where large-scale commercialization spreads costs and profitability relies on scale effects. The aforementioned decline is not good news.

Fortunately, in the first half of 2024, revenue increased significantly by 21.39% to RMB 1.74 billion, and net profit increased by 21.33%. However, it has still failed to emerge from the quagmire of losses. From 2019 to the first half of 2024, net profit attributable to shareholders was RMB -4.963 billion, RMB -12.16 billion, RMB -17.14 billion, RMB -6.045 billion, RMB -6.44 billion, and RMB -2.457 billion, accumulating losses of RMB 49.205 billion.

Operating cash flow was also negative, with net outflows of RMB -1.229 billion, RMB -2.485 billion, RMB -3.084 billion, RMB -3.234 billion, and RMB -3.524 billion from 2020 to the first half of 2024, respectively, continuously negative and increasing.

Lacking self-financing ability, no matter how much funding is received, it will eventually run out. From a results-oriented perspective, performance is an important factor in the cold reception of the stock price. Against this background, it is not difficult to understand the practice of layoffs to save costs and business replacements to create room for profit improvement.

03

The Pros and Cons of Betting on Generative AI

However, cost reduction is only a temporary measure, and the fundamental solution lies in boosting the main business.

In 2023, with the successful launch of the 'Rixin' large model, SenseTime restructured its business segments and established three new businesses: Generative AI, Traditional AI, and Smart Auto. Xu Li pointed out that the company's annual revenue decline was mainly due to industrial transformation, including the active reduction of the 'Smart City' business within the traditional AI business.

Before the business restructuring, SenseTime's main businesses included Smart Commerce, Smart City, Smart Living, and Smart Auto, with Smart Commerce and Smart City being the primary revenue sources, accounting for 38.4% and 28.8% of revenue in 2022, respectively.

From an operational perspective, the four businesses focus on four different application scenarios. The large number of scenarios can easily lead to resource dispersion, high sales expenses, and costs, which is one of the cruxes of profitability. For example, from 2021 to 2022, sales expenses increased from RMB 680 million to RMB 900 million, an increase of 32.2%, and research and development expenses increased from RMB 3.614 billion to RMB 4.014 billion.

At the same time, these emerging businesses are in their infancy and rapidly evolving, susceptible to industry laws, environmental changes, and technological scenario iterations. As mentioned in the 2022 annual report, due to the impact of the pandemic, all traditional AI businesses declined.

Currently, SenseTime's main business structure revolves around Generative AI, with Smart Auto and Traditional AI businesses as supplements. When discussing traditional AI businesses, the company has clearly mentioned, 'Actively promoting our generative AI capabilities to traditional AI business customers.'

As we all know, since 2024, ChatGPT has ignited the generative AI sector. Many companies are shifting from traditional computing to intelligent computing worth trillions, with large models and intelligent computing becoming new productive forces. Behind the change in business structure lies SenseTime's pursuit of hot concepts. Taking the first half of 2024 as an example, the company's research and development expenses were RMB 1.892 billion, a year-on-year increase of 6.1%, mainly due to increased investment in base model training and fine-tuning, the development of generative AI applications, as well as server operation and cloud service fees.

In the first half of 2024, total revenue increased by 21.4% year-on-year; net profit loss was RMB 2.457 billion, a year-on-year increase of 21.33%. Among them, Generative AI business revenue reached RMB 1.05 billion, a year-on-year increase of 255.7%, accounting for 60.4% of revenue, surpassing the traditional AI business for the first time in history to become the dominant business. Smart Auto business revenue was RMB 168 million, a year-on-year increase of 100.4%; traditional AI business revenue was only RMB 520 million, a significant decrease of 50.6% year-on-year, accounting for 29.9%.

The company stated that the growth of the Generative AI business is mainly due to the explosive growth in market demand for generative AI model training, fine-tuning, and reasoning and plans to continue commercializing generative AI products for application in the internet, smart hardware, robotics, healthcare, and finance sectors.

With revenue growth and net profit reduction, the effect of business restructuring is worth acknowledging. Presumably, this is the confidence behind the company's intensified transformation and initiation of a new strategy. However, this does not mean it is flawless.

For example, although Generative AI business revenue reached RMB 1.051 billion, a year-on-year increase of 255.7%, it can be inferred that the business revenue in the first half of 2023 was only about RMB 300 million. The high growth is predicated on a relatively low comparable scale and base proportion. Moreover, popular sectors often face fiercer competition and faster iterations. Whether such growth can be maintained in the future, how strategic resolve will hold up, and whether there is potential to compete for new trends remain to be seen.

Moreover, traditional AI business has declined by 50.6%. Smart Auto revenue is less than RMB 170 million, and the scale is still relatively small. With one AI business increasing and the other decreasing, there is a possibility of 'shifting funds from one pocket to another.' The company needs to provide more evidence in the future.

Looking deeper, while the change in business structure has led to significant growth in some businesses, the crux of poor customer bargaining power and collection ability still exists and may even be worsening. In the first half of 2024, SenseTime's trade receivables were RMB 8.325 billion, higher than the RMB 7.909 billion in the same period last year, with impairment provisions reaching RMB 4.791 billion, also higher than the RMB 4.191 billion in the same period last year. From the first half of 2021 to 2024, the number of days for accounts receivable turnover was 326.59, 493.02, 474.75, and 623.70 days, showing an overall increase. As of the first half of 2024, more than 60% of the accounts receivable had an aging of over two years.

Coupled with the continuous negative increase in net operating cash flow and persistent net losses, SenseTime's business structure transformation is also a heavy burden. It is still in a critical stage of struggling uphill, with mixed fortunes. Perhaps this is why Xu Li had to implement a dual-track approach of 'transformation + layoffs.'

04

Small Test vs. Short-lived

Calling for a New Chapter

Of course, change is the only constant. Especially in the strategically emerging AI sector, everything is in its infancy. Adjustment and transformation are also manifestations of a company's development vitality, flexibility, and keeping pace with the times.

Looking back at 2023, SenseTime's investment and deep cultivation in Generative AI are evident. At the end of August that year, the Rixin large model became one of the first approved large model products, opening up more business opportunities for SenseTime. 'Our generative AI business is the fastest-growing new business with revenue exceeding RMB 1 billion since our inception ten years ago.' Generative AI has been given a special status in the 2023 financial report.

Since 2024, it has accelerated further. Rixin large model press conferences have been frequent, growing and iterating at a visible speed, with only three months passing from version 5.0 to 5.5. Focusing on version 5.5, the overall capability has improved by about 30%, making it the first multimodal real-time interactive large model in China. Based on version 5.5, models of different scales have been extracted, allowing SenseTime to gradually show its prowess in five areas: finance, healthcare, Copilot assistants, anthropomorphic interaction, and smart terminals.

A workman must sharpen his tools if he is to do his work well. The improved performance in the first half of 2024 demonstrates the monetization potential and imagination space of Generative AI. As of the end of June, over 3,000 leading enterprises in various industries have used SenseTime's large models and intelligent computing services, covering industries such as smart hardware, robotics, and healthcare. The call volume of the Rixin large model increased by 400% in the first half of the year.

As of the end of 2023, the company operated 45,000 GPUs. As of August 2024, the total operating computing power exceeded 20,000 PetaFLOPS, with the number of GPUs increasing to 54,000, a roughly 70% increase from March this year.",

Not too demanding. After all, generative AI is also a large-scale business. If it cannot effectively cover costs, it may still fall into a vicious cycle. Looking at the entire AI field, it can be said to be full of ups and downs. After experiencing the bursting of the bubble, we are now witnessing another wave triggered by the emergence of ChatGPT. However, once this wave subsides, can practitioners still meet market expectations and avoid being left stranded? This is also a question that SenseTime, Xu Li, and the company need to consider as they decisively shift focus and go all in.

Managing a business requires both courage and determination. It is said that one cannot establish something new without breaking the old. It is also said that managing a business is like cooking a small fish, requiring great caution. Will the improved performance in the first half of 2024 be a mere trial or a fleeting moment? Standing at the new starting point of its 10th anniversary, where will SenseTime go in the next decade and can it open a new chapter?

This article is originally created by Rhodium Wealth

Please leave a message if reprinting is needed