Revenue has declined, but profits have returned! The behind-the-scenes story of Topsta turning losses into profits: using "subtraction" to achieve profitability. Can Embodied Intelligence become a new engine? | Financial Report Anomaly Perspective

This paper (chinatimes.net.cn) reporter Hu Mengran, Shenzhen photo report

Recently, Guangdong TOSUDA (rights) Technology Co., Ltd. (hereinafter referred to as “TOSUDA,” 300607.SZ), an industrial robotics company, released its 2025 annual report. The data show that the company achieved full-year operating revenue of RMB 2.51 billion, a year-on-year decrease of 12.59%; however, net profit attributable to shareholders of listed companies was RMB 73.8725 million, up 130.12% year on year, successfully turning losses into profits. This “reduced revenue but increased profit” performance report reflects that this industrial automation enterprise with nearly 20 years of history is undergoing a profound structural adjustment.

Under a strategy of proactively shrinking low-margin project-based businesses and concentrating resources on industrial robots and embodied intelligence, TOSUDA is trying to answer a question: after “doing subtraction” brings short-term profitability, where will the next wave of growth come from?

What did the business contraction bring?

Looking through the financial report, the most conspicuous change at TOSUDA in 2025 is the restructuring of its business composition. The company proactively compressed its smart energy and environmental management system business; revenue from this segment fell 25.55% year on year to RMB 915 million. This segment was once an important component of TOSUDA’s revenue, but its gross margin was negative and it consumed a lot of capital. In its performance report, the company said this business had “basically completed divestiture,” and that subsequent revenue would decline further.

The price of contraction is a drop in revenue scale, from RMB 2.87B in 2024 down to RMB 2.51 billion. Meanwhile, the annual revenue contribution share of product-based businesses increased by 6.67 percentage points, and the gross profit contribution share reached 60%, becoming the main pillar of the company’s profits. Gross margins for core product lines such as industrial robots and injection molding equipment all improved; the overall gross margin rose from 14.59% to 28.25%.

Zhang Xiaorong, director of the Research Institute for Advanced Technology, said in an interview with reporters from the Huaxia Times that this strategy of “discarding low-margin to generate profit” has a clear financial improvement effect in the short term. “Although revenue is lower, by cutting off businesses that don’t make money, the company directly turned losses into profits, making its finances healthier. In the long run, putting both money and effort into the main business makes the company more competitive. This approach—first sacrificing scale to protect profits—is very pragmatic; the short-term pains for manufacturers are a wise choice for long-term development.”

Zhou Di, a senior engineer at Rongrong Technology and an expert from the Ministry of Science and Technology’s National Science and Technology Library, added to reporters from the Huaxia Times that, in the long run, focusing resources on core businesses such as industrial robots helps build technological and market barriers. “Although short-term scale faces pressure, development is more focused and more sustainable.”

However, an unavoidable question is: after the smart energy business is basically divested, can other businesses support future growth?

TOSUDA told reporters from the Huaxia Times that future growth will be carried out around the “new decade, three steps” strategy: the first step is to “thicken the fundamental base,” strengthening right-angle coordinate robots and injection molding supporting equipment; the second step is to “strengthen major equipment,” deepening the layout of CNC machine tools, injection molding machines, and more; the third step is to “solidify embodied intelligence,” building an intelligent ecosystem across the full domain.

The performance report shows that in 2025, the company’s robot base unit sales reached 10,437 units, up about 13.7% year on year; but the overall revenue of the industrial robots and automation application systems segment fell 9.24% year on year to RMB 685 million. This means that the increase in unit sales has not fully offset the revenue pressure caused by price and/or structural changes. How to convert sales advantages into sustained growth in revenue and profit remains a challenge for the company.

Regarding the reasons for reduced revenue, TOSUDA explained that its automation application systems business previously focused on leading 3C customers; orders and revenue scale from customers in other industries declined. The company has become more focused on R&D and deployment for “robot +” applications, improving standardized production capacity and lowering the proportion of business tied to customized projects. However, as the depth and breadth of cooperation with leading 3C customers increases, the order scale for related business continues to grow; at the end of 2025, orders on hand increased 116.64% year on year. In the industrial robots segment, the company said its product competitiveness has been continuously improving; its large-customer strategy has achieved results; process and application advantages have become more prominent; and operating revenue increased year on year. Among them, its self-produced multi-joint robots rose 25.32% year on year, right-angle coordinate robots rose 7.35% year on year, and the company’s robot products shipped about 12,000 units throughout the year.

The “industrial route” of embodied intelligence

At a time when embodied intelligence has become the hottest track in the market, TOSUDA did not choose the grand narrative of “general-purpose humanoid robots,” but instead entered from the injection molding scenarios it understands best. In 2025, the company launched humanoid robots “Xiaotuo” for injection molding workshops, as well as quadruped robots “Xingzi,” AI flexible sorting workstations, and other products.

This strategy has been viewed externally as “making a hammer by looking at the nail.” In interviews, TOSUDA also pointed out that the company will analyze and deconstruct the injection molding process steps to form general process packages adapted to different industrial scenarios, using the general characteristics of injection molding scenarios to expand horizontally into more industries. Currently, Xiaotuo’s main application scenarios include material picking, palletizing, and packaging in warehousing and logistics, as well as autonomous loading/unloading and sorting in production.

How is this “scenario-first” approach different? In Zhang Xiaorong’s view, embodied intelligence as a whole is still “an air castle” and lacks truly grounded scenarios. “TOSUDA is entering from the injection molding workshop, which is more down-to-earth. It gets deployed faster—directly using existing customers to conduct testing, quickly enabling it to be used in factories. Commercialization is more stable: factories have real needs and are willing to pay, so the company can earn money quickly.”

Zhou Di also noted that rapid validation through existing customers makes costs more controllable and payback faster—“profit certainty is far higher than a general-purpose route.” However, from injection molding to broader industrial scenarios, and even to commercial, service, and home settings, the technical obstacles to cross-scenario reuse cannot be ignored.

TOSUDA admitted that the main challenges lie in high data costs, weak adaptability, and insufficient generalization capability. To address this, the company and Zhipu Huazhang jointly established Matrix Zhituo to develop low-cost, highly adaptable portable gripper data collection solutions, laying a data foundation for embodied intelligence model training. This is also a key part of the company’s construction of a closed-loop business model of “scenario + product + data + AI.”

In competing with quadruped robots “Xingzi,” facing companies that already have layouts such as Unitree Technology and Yunshenchu, TOSUDA’s strategy is to treat it as an extension of its robot product matrix, creating synergy with the “Toxingji” embodied intelligence product series, industrial robots, and AI workstations, to provide differentiated end-to-end solutions rather than simply competing on individual SKUs.

After profitability, the long-distance race has just begun

In Guangdong Province where TOSUDA is located, the robotics industry is thriving. Data show that in 2025, Guangdong produced 336.3k industrial robots, accounting for 43.5% of the national total, ranking first nationwide for six consecutive years; service robot output was 15.1821 million units, accounting for 81.7% nationwide. As the largest robotics-industry province in China, Guangdong has already formed an end-to-end industrial chain layout covering software, hardware, and core components.

郑磊, Chief Economist of SomiYun Technology Group, analyzed for reporters from the Huaxia Times that this industrial cluster provides TOSUDA with three layers of support: first, extremely fast supply-chain response—within the Greater Bay Area there is a complete robotics supply chain, with “iteration speed 10 times that of Silicon Valley and costs only 1/10”; second, leading scenario richness—Guangdong has all 31 categories of manufacturing industries, and TOSUDA’s 15k existing customers are concentrated in the Pearl River Delta, making scenario testing and data collection costs very low; third, highly efficient collaboration across academia, research, and industry—Shenzhen’s “robotics valley” brings together institutions such as Southern University of Science and Technology and the Chinese Academy of Sciences, forming a “basic research—results transformation” closed loop.

However, large-scale deployment of embodied intelligence in industrial fields still faces key bottlenecks; the biggest obstacle at present is the “data dilemma.” Zheng Lei believes there is insufficient supply of high-quality data, including high collection costs, difficulty extracting implicit knowledge, lack of abnormal data, and “data governance sovereignty” barriers in factories. Zhang Xiaorong said that hardware and algorithms are still bottlenecks—“hardware is expensive, the hands are not dexterous enough, and the algorithms are not smart enough; once you switch to another scenario, it won’t work well, and the cost to adapt the factory environment is high.” Zhou Di added that insufficient technology maturity, relatively high whole-machine costs, and an incomplete industrial scenario adaptation ecosystem are also limiting factors.

For the application outlook over the next 3 to 5 years, Zheng Lei believes industrial embodied intelligence will first scale up in structured industrial scenarios such as handling, sorting, pick-and-place (loading/unloading), and logistics warehousing—environments where conditions are controllable, tasks are standardized, and ROI is clear. 2025 is seen as the “deployment year,” and “semi-autonomous + partial swarm collaboration” will become an important breakthrough. Zhang Xiaorong provided more specific scenario expectations: injection molding, loading/unloading in 3C electronics, automobile assembly, logistics handling, and the like—“these areas have simple processes and high repeatability, replacement of manual labor demands is large, and they are easier to promote.”

TOSUDA’s 2025 annual report outlines a typical transformation case of a traditional industrial automation company at the intersection of the industrial cycle and capital enthusiasm. By proactively shrinking low-efficiency businesses, the company achieved a turnaround from losses to profits financially. By betting on embodied intelligence, it is trying to secure a place in the next wave of technological trends.

But the challenges are also clear: endogenous growth in the product segment has not yet been fully unleashed; the migration of embodied intelligence from injection molding scenarios to a broader market still needs to overcome data and technical thresholds; and competitors—whether startups or peers—are accelerating their deployments. TOSUDA has chosen a “not sexy” but solid path: coming from industry, and going into industry.

Whether this path can succeed will depend on the speed of deploying its “scenario + product + data + AI” closed loop, as well as its ability to convert the existing customer base accumulated over the past 20 years into the first paid users of embodied intelligence. This long-distance race has just begun; whether the profitability gained through “subtraction” can ultimately be transformed into “addition”-style growth will be answered by time.

责任编辑：徐芸茜 主编：公培佳

海量资讯、精准解读，尽在新浪财经APP