On August 17th, the veteran star AI company SenseTime finally saw a long-awaited strong surge, with its share price spiking over 13% during trading before settling more than 8% higher at the time of writing, lifting its total market capitalization back to around HK$64.6 billion. The catalyst for this capital market celebration was a "positive profit alert" released by the company on the evening of the 16th, indicating that it expects to successfully achieve a turnaround to profitability in the first half of 2026. This would also mark the first time SenseTime has recorded consolidated profitability since its Hong Kong listing in 2022.

However, even as the market cheers this historic milestone, SenseTime's stock price remains a vast chasm away from its peak era. Looking back to late 2021 and early 2022, shortly after its listing, SenseTime was highly favored by capital, with its share price once approaching HK$9.70 per share and its total market value surging past HK$270 billion. Today, despite the single-day rally, the stock hovers around HK$1.53, and its market value has shed over 80% from the peak. From being the most capital-courted star during the "AI Dragon" era to enduring a dark period of multiple storms, this former leader is now trying to catch up with the times through a full-scale pivot to generative AI.

Why did it fall behind while AI was at its peak?

Back in 2018, if you asked anyone in the tech circle which company was the strongest in China's AI sector, the answer would likely have included SenseTime. Indeed, during those years, SenseTime was the absolute trendsetter. Facial recognition, smart cities, autonomous driving—almost every hot AI-related track featured its presence. Alongside Megvii, YITU, and Cloudwalk, it was known as one of the "AI Dragons," a frequent headline in tech media, raising rounds of funding and seeing its valuation climb steadily. At that time, SenseTime was like the top student who always aced exams; everyone believed its future was limitless.

But fast forward to 2023, and the narrative shifted dramatically. When ChatGPT, DeepSeek, Claude, and Gemini ignited the globe, when AI became the hottest concept in capital markets, and when OpenAI and Anthropic became the focal points of the tech world, SenseTime gradually faded from the public eye. Why did this happen? The answer may be brutally simple: SenseTime mastered AI 1.0 but missed the boat on AI 2.0.

To understand this, one must first clarify the fundamental difference between the two eras of AI. In the era when SenseTime was founded, AI's primary job was "recognition"—show it a face, and it tells you who it is; give it a street photo, and it flags which car ran a red light; feed it surveillance footage, and it counts passersby. This type of technology is formally known as computer vision, essentially teaching machines to "see" and understand the world. SenseTime excelled here, profiting handsomely by offering facial recognition and smart city solutions to businesses. But the advent of ChatGPT rewrote the rulebook entirely. The new generation of AI is no longer content with just seeing; it also begins to create—writing articles, generating videos, even coding. It's as if SenseTime spent a decade perfecting a "recognition fist" style, only to become the world champion just as the martial arts world suddenly embraced a new "generation sword" technique that everyone is rushing to learn. The old skills aren't useless, but they are no longer the main event.

If it were just a missed technical turn, SenseTime might have had a chance to stage a comeback. What proved truly devastating was the cascade of bad news from 2019 to 2023, as if it were all planned. First, starting in 2019, the US placed SenseTime on sanctions lists. Then, in late 2022, ChatGPT emerged, upending the industry's entire technological paradigm. The computer vision track that SenseTime relied on suddenly shifted from being the future direction to a legacy business, while its efforts in large models lagged a beat. To make matters worse, its former cash cow—the smart city business—began to shrink, and revenue growth stalled. The year 2023 was SenseTime's darkest hour. That year, founder Tang Xiao'ou passed away, a loss mourned industry-wide and a heavy blow to both internal morale and capital market confidence. Almost simultaneously, short-seller Grizzly Research issued a report questioning the company, further depressing the stock. The cumulative toll was stark: in 2021, the smart city business accounted for 45.6% of SenseTime's total revenue; by 2022, it dropped to 28.8%; and in 2023, it fell further to around 12%. SenseTime even proactively reduced its reliance on the smart city segment. Meanwhile, its market value plummeted from a peak of roughly HK$300 billion to just tens of billions of Hong Kong dollars. Headcount was slashed from 6,114 employees to 2,472, a reduction of nearly 60% over three and a half years—essentially six out of every ten employees were let go. Almost every conceivable crisis unfolded over these past few years. This is less a story of falling behind and more a tale of a star AI company being hit by multiple storms simultaneously at a turning point in the era.

Can this company, once reliant on "face-swiping" for revenue, still make a comeback?

If you think SenseTime has completely vanished and will never reappear, you might be overlooking a very interesting fact: SenseTime has transformed from a facial recognition company into a technology enterprise driven by both generative AI and visual AI. In plain terms, SenseTime has deliberately "killed" its former self. Judging from the latest earnings guidance, the transformation is already showing initial results: for the first half of 2026, it expects to record a profit of approximately RMB 500 million to RMB 700 million, compared to a loss of about RMB 1.489 billion in the same period last year. The net loss is projected to narrow by 60% to 70% year-over-year. Notably, SenseTime explicitly stated in the announcement: "This will be the company's first consolidated profit since its listing."

This is a significant signal. Over the past few years, the biggest external skepticism about SenseTime has been whether it can actually make money. As one of the most prominent stars in China's AI industry, from 2018 to 2021, SenseTime told a story of technology changing the world. At the time, it seemed widely accepted that whoever had the most advanced algorithms, the most AI papers, and the highest facial recognition accuracy would become the winner of the next era. But it later became clear that technical leadership does not necessarily translate into commercial success. Previously, SenseTime's core businesses were smart city, smart business, and smart living. More bluntly, it involved selling AI solutions to local governments and enterprises—such as facial recognition access control, urban security systems, traffic identification, and campus management. This business model has a natural flaw: once a project is completed, the revenue stream ends. This makes it more akin to a technical service provider than an internet platform. An internet platform can generate ongoing revenue from users, but a project-based company must constantly hunt for new orders. This is why SenseTime struggled to escape losses for many years.

The emergence of generative AI has changed all that. Previously, SenseTime was selling projects; now, it's selling models. A single large model can serve multiple industries simultaneously, including finance, education, office work, robotics, and content creation. It is no longer a "one-time delivery" business but more like a sustainable subscription. The greatest advantage of this model is its theoretical scalability. In terms of business structure, generative AI has already become SenseTime's new growth engine. For example, in full-year 2025, the generative AI business generated revenue of RMB 3.63 billion, accounting for 72.4% of total revenue. The other main business, visual AI, brought in RMB 1.08 billion, representing 21.6% of revenue. In other words, the SenseTime that once earned its keep through facial recognition now derives more than 70% of its revenue from large models. At the same time, SenseTime has begun to actively control costs. Training large models is essentially a "money-burning" game. In previous years, SenseTime was tagged with the label of a "money incinerator." From its founding to its listing, cumulative R&D investment exceeded RMB 10 billion, and sustained losses were a primary concern for investors. But over the past two years, SenseTime has noticeably adjusted its strategy. On one hand, it emphasizes training efficiency; on the other, it strives to reduce its dependence on high-end computing power by optimizing model architecture and adapting to domestic chips, thereby lowering training costs. From an operational perspective, this strategy is indeed beginning to yield results.

But the issue is that profitability does not necessarily mean a full comeback. So, has SenseTime turned itself around? Probably not yet, because it faces an even bigger question: Do independent AI companies still have a chance? The current competition in large models has essentially become a battle of ecosystems. Look at the players sitting at the center of the table: OpenAI is backed by Microsoft, Google has search, Android, and cloud computing, Alibaba owns Alibaba Cloud and its e-commerce ecosystem, Tencent has WeChat, ByteDance has Douyin... These companies share a common trait: they don't just have models; they also have computing power, users, traffic, and application entry points. The model is merely one component; what truly determines the winner is the entire ecosystem. And this is precisely where SenseTime is weakest. In fact, this is not just SenseTime's problem—it's a challenge faced by all independent AI companies globally. In the past, SenseTime needed to prove that AI technology could become a viable business. Today, it must prove that, in an era dominated by tech giants, an AI company without a super traffic gateway can build a new commercial moat based on its model capabilities and industry deployment strengths.