Each cycle in the AI industry is marked by the interplay of technological innovation and commercial implementation.

Not long ago, SenseTime released its full-year 2025 performance announcement.

The most iconic signals in this financial report were not the record-high revenue of RMB 5.015 billion, nor the narrowed net loss year-on-year, but rather two instances of "turning positive for the first time": In the second half of 2025, the company's earnings before interest, taxes, depreciation, and amortization (EBITDA) turned positive for the first time, reaching RMB 380 million, while operating cash flow also achieved its first positive net inflow since the company's listing.

For a company established 12 years ago that has fully experienced two cycles of rise and fall in China's AI industry, these two positive milestones resemble a watershed.

It signifies that this company, rooted in academic DNA, has finally moved from relying on financing and R&D investment-driven technological exploration to reaching the core threshold of commercial sustainability.

Behind this watershed lies two difficult transitions SenseTime has undergone amidst dramatic technological paradigm shifts and industry cycles—from the golden age of computer vision (CV) to the paradigm shift of generative AI. During this process, the company has enjoyed industry dividends while also experiencing the pain of bursting valuation bubbles.

The release of the financial report coincides with China's AI industry completing a full minor cycle.

ChatGPT triggered a large model frenzy at the end of 2022, and by 2024, the industry collectively returned to commercial rationality. With the full implementation of the "AI Plus" initiative in 2025, the AI industry has shifted from the first half of competing on parameters and computing power to the second half of competing on implementation and profitability.

The former "Four Little Dragons of AI" have collectively faced transformation challenges, with CloudWalk, Yitu, and other peers still struggling in the quagmire of losses. Meanwhile, SenseTime has completely restructured its business architecture within two years, increasing the revenue share of its generative AI business from 34.8% in 2023 to over 70% in 2025, transforming from a CV leader to one of China's top large model companies.

However, the phased results of the transformation do not mean SenseTime has successfully navigated the cycle. Behind the impressive financial data lies intensifying competition in the industry.

At the same time, the valuation logic for AI companies in the capital market has fundamentally changed, shifting from prioritizing "technological narratives" to prioritizing "profit certainty." SenseTime's market value has significantly shrunk from its peak post-listing, and even after the financial report release, its stock price failed to sustain a continuous rise.

The wheel of the AI industry's cycle continues to roll forward, and SenseTime's second half has just begun.

01

Two Cycles: From Lab to Industrial Deep Waters

SenseTime's starting point was the first enlightenment era of China's AI industry.

In 2014, when Tang Xiao'ou, a professor at the Chinese University of Hong Kong, founded SenseTime with his students Xu Li and Wang Xiaogang, deep learning had just achieved breakthrough applications in computer vision, and the entire industry was still in the early stages of "using AI technology to solve single-point problems."

At that time, most domestic AI startups followed the path of "securing projects, implementing solutions, and rapid monetization," as the demand for computer vision in security, finance, and other fields was already clear—stable orders could be secured as long as qualified algorithms were delivered. However, from its inception, SenseTime chose a less-traveled path: heavy investment in foundational technological infrastructure and building a complete AI technology stack.

While most peers directly used open-source frameworks like Google's TensorFlow and Meta's PyTorch, SenseTime invested core resources in developing its own deep learning framework, SenseParrots. When the industry was still competing on single-algorithm accuracy, SenseTime proposed the concept of an "AI infrastructure," preemptively deploying supercomputing centers and stockpiling computing power resources.

These investments, which seemed "reinventing the wheel" and "not cost-effective" at the time, essentially reflected the Tang Xiao'ou team's judgment on the future of the AI industry: AI's future must be generalized, and single-point algorithmic advantages cannot form long-term barriers. Only foundational capabilities in computing power, frameworks, and models can support companies through technological cycles.

This long-term technological investment allowed SenseTime to quickly widen the gap with peers during the golden age of computer vision. By 2019, SenseTime had become the absolute leader in China's computer vision sector, with business covering smart cities, smart commerce, intelligent vehicles, and other fields. By the time it listed on the Hong Kong Stock Exchange at the end of 2021, its market value once approached HKD 370 billion, marking its heyday.

However, the industry cycle turned faster than anyone expected.

Commercial implementation in the CV era heavily relied on project-based delivery, with SenseTime's smart city business once accounting for over 40% of revenue. While such business brought stable revenue, it also suffered from long payment cycles, high client concentration, and weak replicability, leaving the company unable to escape losses despite sustained revenue growth—its cumulative losses exceeded RMB 24 billion by the time of listing.

More critically, the technological paradigm of the AI industry was undergoing disruptive change. The emergence of ChatGPT at the end of 2022 propelled the global AI industry into the large model era, with artificial general intelligence becoming the new industry consensus. Traditional AI companies centered on single-point visual technology collectively faced the risk of being left behind.

To make matters worse, the departure of the founder in early 2024 not only shattered the company's moral support but also left it at a strategic crossroads: Should it stick to the familiar computer vision sector or fully commit to large models, undergoing a painful transformation?

Xu Li and the management team, thrust into crisis, answered with a complete organizational and strategic restructuring. In December 2024, SenseTime officially launched its "1+X" strategy, comprehensively adjusting its business architecture and organizational model.

The "1" refers to the group-level focus on dual core businesses of generative AI and visual AI, concentrating resources on R&D in computing power infrastructure and core models to safeguard technological foundations. The "X" involves spinning off vertical businesses such as intelligent vehicles, smart healthcare, and home robots into independent ecological enterprises with separate CEO mechanisms and fully market-driven operations.

Accompanying this was a company-wide cost-cutting and efficiency-enhancement drive. Over the past two years, SenseTime's total workforce has sharply declined, with sales and administrative expenses Year on year decline . It also decisively divested loss-making non-core businesses—by the second half of 2025, innovative businesses like intelligent vehicles were deconsolidated, generating RMB 1.313 billion in gains from selling subsidiaries alone.

This transformation essentially represents SenseTime's proactive adaptation to cyclical changes in the AI industry.

Competition in the large model era has shifted from single-point technological competition to a full-stack battle of "computing power-models-applications."

Internet giants have entered the fray with traffic, scenarios, and hundreds of billions in capital, leaving independent AI companies unable to compete across all sectors. They must focus limited resources on their most advantageous core areas while capturing vertical opportunities through ecological approaches.

Judging by the 2025 financial report, this transformation has achieved phased results.

Beyond narrowed losses and positive cash flow on the financial end, SenseTime's business structure has fundamentally changed: Generative AI revenue grew significantly year-on-year, accounting for 70% of total revenue and becoming the absolute growth engine. Meanwhile, traditional visual AI business also completed its transformation and optimization, achieving profitability for the first time in 2025 and positive cash flow for two consecutive years, becoming a stable cash cow.

02

Dividends and Hidden Reefs in the Cycle

SenseTime's ability to transform amid an industry downturn essentially stems from capturing dividends from two cycles of the AI industry: Its technological strengths accumulated during the CV era perfectly aligned with the multimodal trends of the large model era, while its preemptively built computing power infrastructure provided a solid foundation for transformation.

In the large model era, multimodal fusion has become a widely recognized technological direction, and SenseTime's nearly decade-long accumulation in computer vision has become its core differentiator from other large model vendors.

In 2025, SenseTime's "SenseNova" multimodal large model ranked first in multiple domestic authoritative benchmarks, while its spatial intelligence model topped global rankings in several international authoritative metrics. The NEO native multimodal model architecture launched at the end of the year achieved top-tier performance with just one-tenth the training data and computing power required by comparable industry models.

This technological edge is translating into tangible commercial implementation capabilities.

Leveraging the "SenseNova" large model, SenseTime achieved scaling (large-scale) implementation across sectors like office, finance, marketing, and industry. On the consumer side, it also shattered market perceptions of being "B-end only."

Take its Raccoon Family products in the Pan office (pan-office) sector, which have served over 15 million individual users and thousands of enterprise clients, with monthly active users growing sevenfold in 2025 and achieving tens of millions in annual recurring revenue.

In the industrial sector, SenseTime built an AI intelligence system for CATL, accurately predicting power loads through large models and dynamically generating optimal energy scheduling strategies, saving clients 7% in electricity costs. Its railway large model, developed in collaboration with China Railway First Survey and Design Institute, integrated knowledge across 28 railway engineering disciplines, enabling full-process intelligent design for railway projects.

Its "Kapi" series of AI-native applications launched last year have amassed over 10 million users, with "Kapi Camera" topping AppStore charts in multiple countries and "Kapi Bookkeeping" achieving a 70% T+1 retention rate, ranking among industry leaders. These C-end products have opened new growth avenues for the company.

However, phased transformation results do not mean SenseTime has successfully navigated the AI cycle.

Behind the impressive financial data, a series of long-term development concerns remain unresolved, while the deteriorating competitive landscape poses significant challenges for SenseTime's second half.

In fact, SenseTime's significant loss reduction in 2025 owed much to one-time gains. Financial reports show that "other net gains" exceeded RMB 1.9 billion in 2025, including RMB 1.313 billion from selling subsidiaries. Excluding this one-time gain, the improvement in losses from core business operations would be markedly narrower.

Meanwhile, the company's loss reduction also stemmed from cost compression, with R&D spending declining for the first time in three years in 2025. For an AI company where technology is the core competitiveness (competitive edge), reduced R&D investment may improve short-term profits but could erode innovation capabilities and technological barriers in the long run.

More alarmingly, growth in its core engine is slowing. Financial data shows a clear downward trend in revenue growth for SenseTime's generative AI business over three years. According to third-party data, China's AIGC core market size grew by 70.8% in 2025, with SenseTime's 51% growth lagging behind the industry average.

Meanwhile, IDC data shows SenseTime's market share in China's AI public cloud services declined from 16% in 2023 to 13.8% in 2024, dropping from second to third place domestically, surpassed by Baidu Smart Cloud and Alibaba Cloud.

03

The Endgame for Independent AI Companies: Finding an Irreplaceable Niche

Behind the slowing growth, the AI industry in the large model era has shifted from single-point competition among startups to a full-industry-chain arms race among internet giants.

In 2025, Alibaba announced plans to invest over RMB 380 billion over three years in cloud and AI hardware infrastructure, while Tencent's R&D spending reached RMB 85.8 billion for the year. Baidu's AI business revenue exceeded RMB 40 billion in 2025, accounting for 40% of Q4 revenue.

Internet giants not only have more abundant funds for computing power and model R&D but also richer application scenarios and traffic entry points, enabling deep integration of large model technologies with core businesses.

In contrast, independent AI companies often lack similar high-frequency, high-stickiness core scenarios, making sustained breakthroughs in the C-end market difficult while facing direct competition from giants in B-end businesses, with growth space continuously squeezed.

Another unignorable challenge is the loss of core talent. In recent years, SenseTime has seen a continuous outflow of key personnel: Former president Zhang Wen left to found Biren Technology, former R&D director Cao Xudong founded Momenta, former vice president Yan Junjie founded MiniMax, and former core members Liu Yu, Yu Fengwei, and Song Guanglu co-founded VivixAI—some of whose businesses now directly compete with SenseTime.

As of early April 2026, SenseTime's market value stood at less than HKD 77 billion, having shrunk by over 80% from its listing peak of HKD 370 billion. Meanwhile, MiniMax, founded by former SenseTime vice president Yan Junjie, is valued at over HKD 320 billion—nearly four times that of SenseTime.

Behind this valuation gap lies a fundamental shift in the capital market's valuation logic for AI companies, with markets now prioritizing ecological niche positioning in next-generation AI scenarios over short-term financial loss reduction.

SenseTime's transformation and challenges essentially reflect the existential dilemma faced by all independent AI companies. While a cohort of independent AI companies like OpenAI, Anthropic, and Cohere have emerged globally, China's internet giants have pursued more aggressive AI layouts with larger investments and deeper scenario integration, intensifying competitive pressures on domestic independent AI companies.

For SenseTime to prevail in this cyclical game, the key is to find its own irreplaceable ecological niche. For instance, its nearly decade-long accumulation in computer vision gives it natural advantages in scenarios requiring deep visual understanding, such as industrial quality inspection, intelligent driving, spatial intelligence, and embodied AI.

These scenarios are precisely those that internet giants cannot fully cover. The real economy's scenarios are sufficiently fragmented and diversified for no single giant to dominate all areas. Vertical-sector depth in technological accumulation and industry knowledge represents the deepest moat for independent AI companies.

SenseTime's "1+X" strategy-built ecosystem also offers new possibilities for competition in vertical sectors. By "having the mothership provide technological foundations while subsidiaries compete in vertical sectors," SenseTime can sustain core technology investments while leveraging subsidiary flexibility and market-driven incentives to establish advantages in vertical sectors, forming a synergistic AI ecosystem.

The AI industry's cycles have always been driven by the interplay of technological innovation and commercial implementation. From the computer vision era to the generative AI era, the industry's technological paradigm has shifted, but the essence of business remains unchanged.

For SenseTime, the past 12 years have seen it capture the first wave of AI industry dividends to become a visual-era leader. The next decade will require it to prove it can find its own path to survival in the multimodal era, achieving the leap from a technology company to a sustainably commercial entity.

Original article by Xinmou