--- title: "SenseTime Redistributes Computing Power: Can Industrial AI Handle the Profitability Task After Receiving 5 Billion in Revenue?" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/42978006.md" description: "At the online institutional research meeting on July 27, SenseTime's management signaled that investment in general large model resources will place greater emphasis on efficiency, with new resources also tilting towards industrial AI agents and automotive intelligence; the smart city business will abandon competition for low-price projects; R&D expenses will continue to be used to eliminate inefficient links; the Rixin Industrial AI suite has already launched pilot programs in manufacturing enterprises in the Yangtze River Delta, with the goal of extracting replicable industry products. Notably, the operational signals released by SenseTime are quite rare, and this is not occurring during a phase of revenue deceleration. As of now..." datetime: "2026-07-28T04:11:40.000Z" locales: - [en](https://longbridge.com/en/topics/42978006.md) - [zh-CN](https://longbridge.com/zh-CN/topics/42978006.md) - [zh-HK](https://longbridge.com/zh-HK/topics/42978006.md) author: "[港股研究社](https://longbridge.com/en/profiles/3199113.md)" --- # SenseTime Redistributes Computing Power: Can Industrial AI Handle the Profitability Task After Receiving 5 Billion in Revenue? At the online institutional research meeting on July 27, SenseTime's management signaled that resource investment in general large models will place greater emphasis on efficiency, with new resources also tilting towards industrial AI agents and automotive intelligence; the smart city business will abandon competition for low-price projects; R&D expenses will continue to be used to eliminate inefficient links; the Rixin Industrial AI suite has already launched pilot programs in manufacturing enterprises in the Yangtze River Delta, with the goal of extracting industry products that can be replicated. Notably, the operational signals released by SenseTime are not only quite rare, but they did not occur during a stage of revenue deceleration. As of now, SenseTime's latest complete financial 口径 is still the 2025 annual report. Data shows that the company's full-year revenue was 5.015 billion yuan, a year-on-year increase of 32.9%; net loss was 1.782 billion yuan, narrowing by 58.6% year-on-year; full-year EBITDA loss shrank to 471 million yuan, with adjusted EBITDA turning positive in the second half of 2025 to 376 million yuan. Trade receivables collection reached 4.871 billion yuan, and the cash conversion cycle shortened from 228 days in 2024 to 129 days. The simultaneous improvement in revenue, loss reduction, and receivables collection has provided operational space for SenseTime to rearrange R&D resources. Based on SenseTime's actual operational development situation, the focus of observing the information released on July 27 lies in growth quality. SenseTime has already accumulated large models, visual algorithms, computing power platforms, and industry customers. The next stage requires converting these capabilities into shorter delivery cycles, higher product reuse rates, and clearer cash returns. **Computing power scale no longer automatically translates into competitiveness; R&D budgets begin to obey commercial closed loops** In 2025, SenseTime's growth was still driven by generative AI. Data shows that this sector's revenue was 3.63 billion yuan, a year-on-year increase of 51%, accounting for about 72% of the group's revenue; computer vision revenue was 1.083 billion yuan, a year-on-year increase of 3.4%; "X Business" revenue was 302 million yuan, a year-on-year decrease of 5.9%. It is evident that SenseTime's revenue structure is highly concentrated in large models and related infrastructure. Model capability remains SenseTime's business foundation, and its resource adjustments are closer to internal rebalancing. The question is that although it achieved high growth, there are still heavy cost constraints behind it. Data shows that in 2025, SenseTime's R&D expenses were 3.775 billion yuan, a year-on-year decrease of 8.6%, a scale equivalent to about three-quarters of the full-year revenue; total R&D, management, and sales expenses were 5.57 billion yuan, a year-on-year decrease of 10.9%. The gross profit margin during the same period was 41.0%, lower than 42.9% in 2024; capital expenditures reached 3.488 billion yuan, accounting for 69.6% of revenue, and cash reserves at the end of the year were 14.247 billion yuan. SenseTime has the financial conditions to continue investing, but the method of investment needs stronger output constraints. Management expert Peter Drucker distinguished between efficiency and effectiveness: efficiency is about doing things right, while effectiveness is about choosing the right things. For SenseTime, compressing repetitive training, low-return projects, and price competition belongs to efficiency governance; embedding multimodal models into high-professionalism scenarios such as industry and automobiles is the rational business choice. This type of adjustment is also consistent with the cost structure of the large model industry. General models require continuous investment in training, inference, servers, and cloud service fees, but performance improvements may not necessarily translate synchronously into customer budgets. As model capabilities gradually become a public foundation, enterprise competitiveness comes more from unit inference costs, product delivery speed, industry data accumulation, and customer renewals. On July 15, SenseTime further upgraded the Token Plan, emphasizing the entry of domestic chips into the industrial-grade large model service chain and enhancing the calling ability for complex agent tasks. It is evident that the underlying models have not exited, and the company is trying to reduce the production cost of unit intelligence, forming a tighter revenue loop among models, computing power, and applications. The trade-offs in the smart city business should also be placed within this framework. Data shows that SenseTime's trade receivables collection in 2025 increased by 5.4% year-on-year, financial asset and contract asset impairment losses decreased from 781 million yuan to 287 million yuan, and the cash conversion cycle shortened significantly. Low-price projects are usually accompanied by long delivery times, heavy customization, and collection pressure. Scale expansion easily occupies R&D and working capital. SenseTime's continued abandonment of low-quality orders helps protect cash flow quality, but may also cause phased revenue disturbances. For an AI company still in a loss-making state, a decrease in revenue does not equate to marginal deterioration. After the exit of low-margin, long-account-period projects, as long as generative AI, industrial agents, and overseas visual businesses can make up for the scale, the company's overall cash flow quality is expected to continue to improve. Subsequent financial reports need to observe the impact of smart city contraction on revenue, and whether expense reductions can outpace business scale adjustments. **The threshold for industrial AI is shifting from model performance to production systems; SenseTime must cross the "pilot economy"** The tilt of industrial AI towards SenseTime's resources is in sync with policy and manufacturing demand. The "Special Action Implementation Opinions on 'Artificial Intelligence + Manufacturing'" issued by eight departments proposed that by 2027, promote the deep application of 3 to 5 general large models in the manufacturing industry, launch 1,000 high-level industrial agents, build 100 high-quality industrial datasets, promote 500 typical application scenarios, and select 1,000 benchmark enterprises. The policy also explicitly requires AI to enter the entire process of R&D design, pilot verification, production manufacturing, marketing services, and operation management. Policies have already provided scenario and budget directions, but corporate procurement still looks at quantifiable returns. Currently, the judgment standards at the industrial site are very specific: how much the defect rate decreases, whether downtime can be reduced, whether scheduling efficiency improves, and whether energy costs can be lowered. Models must also connect to equipment, sensors, industrial software, and enterprise databases, and meet real-time, reliability, and security requirements. Manufacturing customers rarely pay continuously for a single model demonstration; project acceptance still depends on stable operation and traceable results. There is a significant difference between industrial AI and office agents. Errors in office scenarios can usually be manually reviewed; misjudgments in production line scenarios may cause raw material waste, equipment downtime, or even safety accidents. Therefore, industrial enterprises are willing to pay for reliability and will set longer testing, verification, and procurement cycles. SenseTime's foundation in industrial AI mainly comes from visual AI. The company already has industrial quality control products covering defect detection, safety management, and intelligent operation and maintenance; CV2.0 achieved profitability for the first time in 2025 and recorded positive cash flow for two consecutive years. Visual recognition is suitable for cutting into quality inspection, patrol inspection, and equipment management, while multimodal models can further process process documents, on-site images, equipment logs, and expert experience. SenseTime's advantage lies in the synergy of multimodal, machine vision, and computing power platforms, while its shortcoming is that manufacturing process knowledge is scattered across different enterprises and production lines. Industrial software companies are familiar with processes, automation manufacturers master equipment interfaces, and cloud vendors possess customer channels. SenseTime needs to jointly define products with industrial leaders, equipment manufacturers, and software vendors to reduce the pressure of bearing heavy deliveries alone. The scaling of industrial AI is completed jointly by ecological collaboration, standard interfaces, and customer repurchase; a single model leaderboard cannot replace this process. SenseTime's choice of manufacturing enterprises for pilots in the Yangtze River Delta has good industrial conditions: the region is densely populated with auto parts, electronic information, equipment manufacturing, and new energy industries; the digital foundation of enterprises is relatively mature, and similar factories are easier to form replication chains. However, the gold content of the Yangtze River Delta pilot needs to be confirmed by replication efficiency. Whether the first batch of customers can turn into paid contracts, whether the same set of agents can be deployed across factories, whether the implementation cycle can shorten quarter by quarter, and whether the proportion of software and continuous service revenue can increase—these indicators are more important than the number of pilots. If revenue growth relies on increasing personnel, the gross profit margin will be difficult to sustain, and order visibility will also be weak. Conversely, if the Rixin Industrial AI suite can precipitate unified data governance, model calling, task orchestration, and delivery tools, SenseTime will have the opportunity to transform one-time delivery into a composite income of "platform fees, software subscriptions, model calls, and continuous maintenance." **SenseTime's next answer sheet must be written simultaneously on factory production lines and cash flow statements** Currently, industrial AI is undertaking the upgrade of the revenue structure, smart cars continue to expand technological extensions, and the smart city business has entered the stage of active project screening. Thus, SenseTime's operational focus has shifted from technical scale expansion to the turnover efficiency of technical assets, which is a common topic in the commercialization stage of domestic AI companies. Smart cars provide another commercialization path for SenseTime. In 2025, SenseTime's edge-side chip and intelligent driving sectors completed financing and became independent operations off the consolidated balance sheet, reducing the group's continuous investment pressure, and related businesses obtained independent financing and industrial cooperation space. During the 2026 Beijing Auto Show, SenseTime's Jueying released a cockpit-driving integrated agent product system; public information shows that Jueying has cooperated with over 30 automakers, covering nearly 200 car models, with cumulative shipments approaching 5.5 million units, and plans to launch Robotaxi trial operations with T3 Travel within the year. Going off the consolidated balance sheet will weaken the direct contribution of the automotive business to the group's consolidated revenue, but it can improve capital allocation. SenseTime can retain technological synergy and equity returns, while the smart car team independently finances based on mass production orders. Subsequent observation points include new designated points, mass-produced model sales, per-vehicle value, customer concentration, and cash consumption of the independent company. The automotive business has a long vehicle cycle; once a product enters mass production, it can form several years of revenue visibility; however, the pressure is also clear: automakers continue to suppress supplier quotes, hardware platforms switch rapidly, and intelligent driving R&D investment is highly front-loaded. Shipments need to be further converted into per-vehicle revenue, gross profit, and cash collection to reflect the operational value of the business. SenseTime has already crossed two thresholds: the recovery of revenue growth and the significant narrowing of losses. Industrial AI determines whether the business can form standard products, smart cars test whether ecosystem enterprises can grow independently, and the contraction of smart cities tests the management's trade-off on revenue quality. Among them, positive signals come from expense reductions, improved receivables, CV business profitability, and the turnaround of H2 EBITDA; the part not yet completed is the conversion of industrial pilots into scale revenue, and the continuous positivity of operating cash flow. The AI industry lacks neither technological peaks nor scarcity in the ability to embed technology into daily production and charge for it long-term. SenseTime has begun to push resources to the industrial site, and the direction already possesses industrial foundations and policy support. In the next few financial reporting cycles, revenue structure, R&D expense ratio, industrial customer repurchase, and cash flow continuity will give clearer answers. These indicators will determine whether SenseTime has the opportunity to complete the identity upgrade from algorithm supplier to industrial intelligent platform. ### Related Stocks - [00020.HK](https://longbridge.com/en/quote/00020.HK.md) - [80020.HK](https://longbridge.com/en/quote/80020.HK.md)