Financial Report Analysis: Will SenseTime in 2025 Be the Forgotten Top Student?

Over the past three years, the capital markets have repeatedly asked AI companies the same question: technology can change the world, but when—exactly—can it make money?

In 2025, SenseTime delivered a solid answer—its performance reached a historical high, losses narrowed significantly, and cash flow turned positive. This report signals that AI companies have, for the first time, crossed a critical threshold: moving from “burning money to tell a story” to “self-sustaining cash generation.” In the past few years, this has been the hardest question the industry has had to answer.

However, there’s an intriguing scene: the moment SenseTime proved with real cash that “AI can be profitable,” the spotlight from capital shifted to chasing new stars that still record massive losses, yet boast explosive user growth and a “next-generation narrative.”

The “New Six Dragons,” represented by Zhipu AI and MiniMax, are rewriting market rules with astonishing valuation premiums. SenseTime’s inflection point reflects a brutal truth from the early days of the AI industry: in an era where bubbles and belief coexist, what capital rewards is never today’s “profitability,” but rather tomorrow’s “imagination.”

When AI starts making money: what SenseTime crossed isn’t just the profitability line

From a financial metrics standpoint, SenseTime’s 2025 performance is nearly flawless.

Revenue exceeded 5 billion yuan, setting a record high and growing by more than 30% year over year; its generative AI business surged by 51%, becoming an absolute growth engine. More importantly, full-year net losses narrowed by nearly 60%. In the second half of the year, EBITDA turned positive for the first time, and operating cash flow also achieved a net positive inflow for the first time since listing.

Behind these figures lies the real change—not simply “how much it earned,” but something more important:

SenseTime’s unit economics model has begun to hold.

In essence, past AI companies were a kind of “technology investment vehicle.” Their revenue relied heavily on project-based delivery, with long cycles, slow collections, and limited replicability—causing “growth on the books” and “cash capability” to stay disconnected for a long time. A common industry concern was that revenue was growing, but cash was being consumed.

SenseTime’s shift hits this structural problem head-on. Its accounts receivable collections reached a new high, and cash flow turned positive—meaning its revenue began to genuinely convert into cash, rather than remaining confined to the statements. The emergence of this “cash-generating capability” marks the first time AI companies have had the ability to run in a self-sustaining loop.

An even deeper change shows up in its business structure. The surge in generative AI revenue has gradually helped SenseTime break away from the earlier project-based model centered on visual AI, moving toward “platform-style revenue” based on model services. This means revenue no longer depends on one-time delivery; instead, it is built on continuous usage and continuous payment.

From this perspective, SenseTime isn’t “approaching profitability”—it has completed a critical leap: transforming from a technology company into an AI infrastructure provider with a real commercial closed loop.

This step is extremely critical—and also exceedingly rare—in the AI industry. But the problem is that the capital markets did not respond to this “correct answer” with matching feedback.

The path starts to diverge: the AI Six Dragons head toward different endgames

Roll the clock back two years: “China’s AI Four Dragons” was still the mainstream market view of Chinese AI companies. But by 2025, this landscape has been completely broken apart, and a new competitive structure is taking shape—older AI platforms represented by SenseTime, together with new-model forces represented by Zhipu AI and MiniMax, forming a new “AI Six Dragons.”

But the real difference isn’t in the number—it’s in the switching of valuation logic.

Zhipu AI follows a typical “open-source + ToB” path. With a Tsinghua background, its GLM model enjoys high prestige in the developer community, and through an open-source strategy it quickly accumulates ecosystem influence. Even though it is still in a state of massive losses, the market is willing to pay a high premium for its “future platformization capabilities.” Its expected IPO valuation far exceeds its current scale of revenue.

MiniMax represents the other extreme path. By entering from the C-end with applications such as Talkie, it rapidly accumulates global users. Its user base has surpassed 200 million, and in video generation and multimodal fields it ranks in the global top tier. The core support for its valuation isn’t profitability, but the compounded effect of “user scale + a global narrative.”

In comparison, SenseTime’s logic looks “overly rational.” It emphasizes loss reduction, cash flow, and efficiency—an unmistakable “value restoration” narrative.

The issue is that in an industry still in its explosive early stage, capital often does not prioritize rewarding this kind of rationality.

What the market prefers is the steepness of the growth curve, not the steadiness of the income statement. This also directly explains SenseTime’s stock price “falling behind.”

On one hand, it carries a clear label from the “old era.” As a company listed in 2021, SenseTime has long been categorized as a “computer vision company,” and that perception is difficult to completely reverse in the short term.

Even if it has fully shifted to large models, the market still tends to treat it as “the previous generation of AI companies.” By contrast, from the outset, Zhipu and MiniMax have been defined as “generative AI native players,” which naturally grants them greater valuation flexibility.

On the other hand, SenseTime’s share structure also suppresses its stock performance. Early investors’ lock-up releases and sell-downs subject it to ongoing selling pressure, while newly listed companies have the dual advantage of “clean share structure + scarcity premium,” making it easier to attract capital inflows to concentrate.

More importantly, there is a gap in “product perception.” MiniMax has built strong user recognition through C-end applications; Zhipu builds a developer community through open source; and SenseTime’s advantages are still mainly concentrated in the B-end and G-end. These “invisible capabilities” often fail to translate into valuation premiums in the capital markets.

Ultimately, the problem with SenseTime isn’t fundamentals—it’s where it sits. It stands in an awkward middle ground: neither is it a high-risk growth target anymore, nor has it yet become a stable cash cow.

At such a stage, the market is often most likely to “ignore” it.

The real battlefield has just begun: the next fight in multimodal, AI agents, and globalization

SenseTime’s 2025 is about proving it can “survive,” but the next battlefield will be a contest over “how fast it can run.”

The arrival of a profitability inflection point is only an entry ticket into higher-dimensional competition—it’s not the final victory. The decisive showdown is only just beginning. The focus will shift to breakthroughs in multimodal capabilities, the large-scale rollout of AI agents, and expansion into global markets.

Multimodal has become the new technical dividing line. The NEO architecture proposed by SenseTime tries to unify language and vision, integrating understanding and generation into a single system. Implied behind this is a redefinition of the “general intelligence path”—AI is no longer just answering questions; it begins to understand the world.

Meanwhile, AI agents are becoming the new application entry point. From office work and marketing to financial decision-making, AI is evolving from a tool into an “executor.” Whoever can capture this entry point may be the one to become the next-generation operating-system-level platform.

At this level, SenseTime’s advantages still exist. Its “computing power + model + data” closed-loop capability gives it the potential to build infrastructure—this is also where it comes closest to Palantir. And in the enterprise-level AI services field, its path has a certain alignment with C3.ai.

But challenges are equally obvious. Global competition is accelerating. Middle East compute capacity, domestic chip ecosystems, and international market expansion will all become key variables that determine the outcome. And in the C-end and ecosystem layers, SenseTime still needs to make up shortfalls in order to truly enter “platform-level competition.”

In the future, SenseTime must find a new balance between “being practical” and “being visionary.” It must maintain its current cash-generating ability to ensure sustained R&D investment. At the same time, it needs to learn to tell stories like the new forces—building phenomenon-level C-end products, reshaping the open-source ecosystem, and tearing off the “old era” label.

If SenseTime cannot become again the “most trustworthy future,” then it may only ever be an excellent software company—and will not become a technology giant that defines the era.

When everyone is talking about AI’s infinite possibilities, SenseTime must use a growth speed that goes beyond financial data to prove to the market that profitability and dreams are not mutually exclusive.

Otherwise, in this “win by speed” race track, early leaders are easily turned into martyrs.

Author: Sangyu

Source: Hong Kong Stock Research Society