SenseTime at the Crossroads: Amputating to Survive, Reducing Losses Shows Results, Core Business Growth Slows

Financial News: Zhang Xiaodi, reporting from Shanghai

From March 27 to 29, 2026, at the Shanghai Global Developer Pioneer Conference (GDPS) on-site, audiences gathered in front of SenseTime’s booth to queue up and try the “one-click shrimp” feature. Staff demonstrated how to quickly call up an AI agent—part of the OpenClaw ecosystem, commonly known as “Lobster”—through SenseTime’s input method AudioClaw to complete tasks such as meeting recording and content rewriting.

This is SenseTime’s second consecutive year of deep involvement in this developer conference. Unlike previous showcases that focused more on the scale of underlying large-model parameters, this time SenseTime’s emphasis clearly tilted toward the application layer: it launched the AI speech open platform SenseAudio, introduced the lightweight AI agent entry point AudioClaw, and hosted three technical workshops.

These moves point to a clear intention—to draw closer to the developer ecosystem and capture the entry points to the AI application layer. Behind this market action is SenseTime’s predicament of painful transformation: according to its 2025 financial report, the company’s net loss narrowed by 58.6% to 1.782 billion yuan, and in the second half of the year it achieved positive EBITDA (earnings before interest, taxes, depreciation, and amortization) for the first time, reaching 380 million yuan; however, at the same time, the growth rate of its core generative AI business plunged from 199.9% in 2023 to 51%, while its computing power operating costs surged by 163.5%.

After cutting off an arm to survive, the combination of reducing losses that actually works and growth that has stalled forms the core tension SenseTime faces right now.

How Much the Loss Reduction Really Helps: Asset Sales and Layoffs Make a Clear Contribution

The 2025 financial report is one of SenseTime’s closest-to-profit results in recent years. Full-year total revenue reached 5.015 billion yuan, up 33%; net loss was 1.782 billion yuan, narrowing by 58.6% from 4.307 billion yuan in 2024. Operating cash flow turned positive for the first time, and the cash turnover cycle shortened from 228 days to 129 days.

However, these improvements in financial figures do not all come from strong growth in the core business.

Since 2023, SenseTime has been continuously adjusting its business structure: first, it separated generative AI into an independent business segment; by the end of 2024, it established the “1+X” strategy, with the Group focusing on two core areas—generative AI and visual AI—while businesses such as “smart vehicles” and “robots” were split off and operated independently.

In the second half of 2025, “X” businesses such as smart vehicles (Jueying) and AI GPU chips were successively spun off. The financial report shows that in 2025 SenseTime’s “other gains, net” exceeded 1.9 billion yuan, of which profit from the sale of subsidiaries was 1.313 billion yuan. If this one-time gain is excluded, the improvement in losses from the main business would narrow significantly.

Over the same period, the company’s total employee headcount fell from 4,672 in June 2024 to 3,206 in June 2025—down by 1,466 people in a year, a decline of 31%. Sales expenses decreased by 20% year over year, and administrative expenses decreased by 3.1% year over year.

This “making cuts” approach to reduce losses has sparked market discussion about the quality of profitability. Shen Meng, a director at Xiangsong Capital, told Financial News that although SenseTime’s performance has improved on paper, fundamentally it cannot dispel investors’ concerns. The improvement comes from non-recurring gains, and is not sustainable.

Liu Zhigeng, a researcher at the SuGang Management Accounting and Auditing Institute, told Financial News that this is a typical “cut costs and improve efficiency” strategy. In the short term, it is a necessary and effective survival measure, but in the long term it is hard to support high-quality development.

Liu Zhigeng said that selling non-core assets, divesting loss-making businesses, and laying off staff on a large scale—along with cutting sales and administrative expenses—has bought the company a transformation window. But this improvement is “unsustainable.” If the gains from one-time asset sales are excluded, the core business’s ability to generate cash still remains weak.

Meanwhile, the business reorganization has also brought new challenges. On the one hand, after business spin-offs, the Group’s sources of revenue become more concentrated, making it more dependent on the performance of its generative AI business. On the other hand, a drop in R&D spending and the loss of core technical personnel may affect long-term technological reserves.

From the financial data, SenseTime’s R&D spending in 2025 saw its first decline in nearly three years. In response, Liu Zhigeng said that layoffs and reductions in R&D spending can reduce costs in the short term, but they may damage the company’s innovation foundation and long-term competitiveness—no different from “quenching thirst by drinking poison.”

Shen Meng believes that the logic behind SenseTime’s main business inherently sets a ceiling for growth, and when combined with its “cut-down” strategy, it is even harder for investors to believe SenseTime has found a path to stable, sustained, and rapid growth.

Regarding the decline in R&D spending, SenseTime’s management at the earnings briefing explained layoffs as “improving workforce efficiency,” and described asset sales as “realizing ecological dividend.”

A noteworthy phenomenon is that SenseTime is becoming a “talent export base” in the AI sector. Besides MiniMax, after former president Zhang Wen left SenseTime, he founded Bairen Technology (the first domestic GPU stock in Hong Kong). Former research and development director Cao Xudong founded Momenta (a provider of China’s intelligent driving solutions). Core members Liu Yu, Yu Fengwei, and Song Guanglu jointly founded Vivix AI (a real-time interactive multimodal AIGC company). Some of their businesses have already formed direct competition with SenseTime.

Core Business Growth Slows Dramatically: Generative AI Growth Is Cut in Half

Generative AI is SenseTime’s core growth engine after its transformation. The financial report shows that its revenue share increased from 34.8% in 2023 to 72.4% in 2025, becoming the company’s absolute pillar. But this engine’s speed is now slowing down.

According to financial report data, the revenue growth rate of SenseTime’s generative AI business fell from 199.9% in 2023 to 103.1% in 2024, and further down to 51% in 2025. Over three years, the growth rate has clearly declined.

It is worth noting that this slowdown is not unique to SenseTime. Third-party data shows that in 2025, the growth rate of China’s core AIGC market size was 70.8%, and SenseTime’s 51% growth is about 20 percentage points below the industry average. Looking globally, after the generative AI market experienced explosive growth of more than 300% in 2024, it is now entering a transition period from “technology trials” to “commercial rollout.” MIT research indicates that 95% of enterprise generative AI projects have not yet generated measurable profit returns. OpenAI also shut down the highly watched Sora video product in March 2026, which the industry has interpreted as the AI industry “returning from technological romance to commercial rationality.” Against this backdrop, SenseTime’s slowdown reflects both its own strategic adjustments and the industry’s stage-specific characteristics.

At the same time, SenseTime’s share in the generative AI market has declined. According to the IDC report “China Artificial Intelligence Public Cloud Service Market Share 2024,” SenseTime’s market share dropped from 16% in 2023 to 13.8% in 2024. Its ranking fell from second in China to third, overtaken by Baidu Intelligent Cloud and Alibaba Cloud.

Another risk worth paying attention to is customer concentration. The 2025 financial report shows that the revenue share from the company’s largest customer is as high as 19%. This means that losing a single major customer could have a significant impact on revenue.

At the 2026 developer conference, SenseTime rolled out a developer-facing speech platform and an intelligent agent entry point, attempting to rebuild its connection with the market by going deeper into the developer ecosystem. AudioClaw’s “one-click shrimp” feature and the SenseAudio open platform lower the threshold for developers to use AI capabilities and attract more developers into its ecosystem. But whether this strategy can reverse the market’s valuation perception of SenseTime remains uncertain.

Gap in Market Value: From AI Leader to a Low Spot

In contrast to the effectiveness of reducing losses, SenseTime’s market value continues to face pressure.

As of April 1, 2026, SenseTime’s market capitalization was less than 77 billion Hong Kong dollars. Yet shortly after it was listed in early 2022, its market capitalization once approached 370 billion Hong Kong dollars. Since then, it has been choppy downward. In April 2024, it fell to around 50 billion Hong Kong dollars, nearly 90% wiped out compared with its high point. Although it returned to the trillion-HKD threshold in September 2025 on the back of optimism from AI policy, it is still currently at a low level.

Meanwhile, MiniMax, founded by former SenseTime vice president Yan Junjie, already has a market capitalization of over 320 billion Hong Kong dollars. Zhipu, founded by a Tsinghua team, also has a market capitalization of over 310 billion Hong Kong dollars. This means SenseTime’s market value is less than one quarter of these two up-and-coming players.

Behind this market value gap is a generational shift in how the capital market evaluates AI companies. Several interviewed institutional investors said the market currently values “ecosystem positioning” capabilities more—that is, whether a company can define standards in the next generation of AI application scenarios—rather than near-term results from reducing losses.

However, in early 2026, the stock prices of MiniMax and Zhipu also saw sharp rallies followed by significant pullbacks.

In Shen Meng’s view, SenseTime, MiniMax, and Zhipu all have performance ceilings because, fundamentally, their businesses are a form of system integration. Whether it is software integration or large-model and agent integration, the purpose is to build systems for customers and solve problems. In system integration, the bidding is essentially on price. To break through a performance ceiling, you need differentiation, more R&D investment, a longer cycle, uncertain results, and higher risk—making it difficult to translate into commercial revenue.

Liu Zhigeng said that SenseTime’s long-term depressed market valuation primarily comes from the market’s deep doubts about its “growth quality” and “future profit model.” First, its profit model is unclear: although the revenue share from generative AI has reached 72.4%, the company is still in a stage of “growing revenue but not growing profit,” and it is severely dependent on large customers, making its business model weak in risk resistance. Second, it carries heavy historical baggage: it has posted losses for 10 consecutive years, with cumulative losses of over 50 billion yuan; and the company’s accounts receivable with more than 3 years aging, accounting for 56.6% of the total, is a potential “profit black hole.”

In addition, SenseTime is facing a worsening competitive landscape. Liu Zhigeng said that the capital market’s enthusiasm has shifted from the “AI four little dragons” to more imaginative general large model companies, making SenseTime’s “story” less fresh. At the same time, SenseTime is also going through the pains of strategic transformation—from visual AI to generative AI. While that is an inevitable trend, the shrinking of the old business and the huge investment in the new business have put pressure on gross margins.

The financial report shows that in 2025 SenseTime’s computing power operating costs surged by 163.5%, and its gross margin fell from 44% year over year to 41%. This means that for every additional 1 yuan of revenue, the company needs to pay higher computing power costs—contrasting sharply with the “marginal cost approaching zero” characteristic of internet platform economies.