In the quagmire, SenseTime bets on RWA, aiming to turn hundreds of billions of Computing Power into gold.

Recently, SenseTime announced the completion of a new share placement of approximately HKD 2.5 billion. What has attracted market attention is not the scale of the financing, but its clear purpose for the funds - the announcement clearly stated that about 20% (nearly HKD 500 million) will be used to promote the development of its RWA (Real World Assets) and stablecoin-related business.

The news caused an uproar in the industry. This move posed a more pressing question for the entire sector: "After the 'Battle of a Hundred Models', when the traditional SaaS charging model can hardly support the enormous R&D investment, how should AI companies truly make a profit?"

For SenseTime, which has accumulated losses of hundreds of billions and is currently in the core battlefield, is this ultimately a helpless move under financial pressure, or a far-reaching layout aimed at revitalizing core assets? SenseTime's RWA strategy may be the breakthrough answer to this question. It reveals a brand-new path: deeply embedding AI capabilities into finance, allowing the company to transform from a "technology service provider" into an "asset creator." This is not only SenseTime's self-rescue but may also be the common chess game for AI giants in the future.

From the core contradictions in the financial report, RWA is a financial breakthrough under the situation of "well-received but not profitable."

Why does SenseTime embrace RWA? SenseTime's financial statements provide us with an answer: they clearly present a harsh reality of being "well-received but not well-sold."

Despite the fact that the generative AI business revenue reached 120 million RMB in 2023, with a year-on-year growth of as much as 200%, making it the fastest-growing new business segment in the company's ten-year history, the net loss attributable to the parent company still reached 6.495 billion RMB in the same year.

From 2018 to 2023, SenseTime's cumulative losses have reached 50.324 billion yuan. If we add the 2.477 billion yuan loss in the first half of 2024, the total cumulative loss has risen to 52.8 billion yuan. This continuous and massive loss stands in stark contrast to the company's significant technological advancements and market share in the generative AI field.

The root of this contradiction lies in the high R&D investment and relatively slow commercialization progress. From 2018 to the first half of 2024, SenseTime's cumulative investment in R&D expenses has reached 18.205 billion RMB. Although the explosive growth of generative AI business and the surge in model usage indicate the powerful potential of the technology, the traditional SaaS charging model (model as a service) is "still very far" from achieving profitability in core business. Faced with a market value of HKD 39.359 billion, which is just a fraction compared to the peak market value of HKD 320 billion, the management must seek a new model that can directly and quickly monetize the huge technology investment to improve cash flow and reduce losses.

RWA provides a potential path for SenseTime to break through the limitations of traditional SaaS models by "realizing" expected returns, enabling rapid recovery of technical investments and revitalization of assets.

SenseCore: From Cost Center to Tokenizable "Heavenly Strategy"

SenseTime's SenseCore AI** large device is its core technological moat and also its largest cost center. **Since around 2018, SenseTime has invested billions annually to build its own computing power infrastructure, including the AIDC intelligent computing center in Lingang, Shanghai, and continues to expand computing nodes in places like Guangzhou and Shenzhen, forming a domestically scarce large model infrastructure. As of 2023, the total computing power scale of SenseTime's large device has exceeded 12,000 petaFLOPS, with the number of GPUs online reaching as high as 45,000 cards. By 2024, its computing power has increased to 23,000 Petaflops, a year-on-year growth of 92%.

Such a massive investment in computing power has led to remarkable improvements in model training and inference efficiency. For example, SenseCore can achieve a 600% increase in large model inference efficiency, with incremental training costs reduced by 90%. However, the construction and maintenance costs of these heavy assets are enormous, and they are not fully utilized, resulting in a significant cost burden.

RWA provides the "chosen strategy" to solve this core contradiction: transforming the huge computing power costs into tradable, financeable, and profitable financial assets through "computing power tokenization." Computing power assets (such as GPU hardware) are considered ideal RWA anchoring assets due to the "rigid demand" of the AI industry and credible "digital genes." Through Web3 technology, SenseTime can digitize physical devices (computing power) to achieve instant monetization of expected returns. This model not only activates the massive fixed assets accumulated on SenseCore but also provides new financing channels for its continuous computing power expansion and technological iteration, fundamentally reshaping the valuation logic of its heavy asset operations.

From visual AI to multimodal, the unique "ticket" to enter RWA after the technology has settled.

SenseTime's strategic transformation from visual AI to generative AI has accumulated a unique technological "ticket" for its entry into the RWA field.

Before its listing in 2021, SenseTime was the largest computer vision software company in China, with businesses covering smart commerce, smart cities, smart cars, and smart living. Over the past decade, SenseTime's deep accumulation in perception intelligence and decision intelligence, as well as its reserve of massive multimodal data, has strengthened its foundational model's understanding of the physical world and multimodal capabilities. For example, SenseTime's SenseNova large model system has reached a leading domestic level in several aspects, including foundational models, multimodality, programming and tool invocation, and lossless context with millions of words.

The value anchoring and liquidity management of RWA often require precise assessment of complex assets, which relies on the integrated analysis of various heterogeneous data such as satellite remote sensing data, IoT sensor data, and text information. SenseTime's extensive applications and data accumulation in remote sensing interpretation, smart city management (such as urban services and emergency response), smart vehicles (SenseAuto), smart mobile terminals (SenseME), and the financial sector (in collaboration with China Merchants Bank, Haitong Securities, etc.) enable it to process and understand these complex multimodal data.

AI technology can automatically identify and assess assets that are traditionally difficult to value through multi-source data fusion and dynamic evaluation, generating accurate dynamic valuation models. In addition, AI also plays a key role in cross-border data compliance review, achieving real-time compliance checks through intelligent identification, dynamic monitoring, and risk assessment, further reducing the compliance risks of RWA projects. These technical capabilities enable SenseTime to empower the issuance, pricing, liquidity management, and compliance of RWA from a technical perspective, providing a solid technical foundation for its deep participation in the RWA market.

In summary, SenseTime's embrace of RWA is an inevitable strategic choice for it to seek monetization of core assets (SenseCore computing power) and reshape valuation logic under financial pressure. The explosive growth in the generative AI field and the contradiction of ongoing losses, along with the operational pressure of SenseCore as a heavy asset, compel it to explore new profit models that can directly convert technological investments into financial assets. At the same time, SenseTime's long-standing deep technological accumulation in visual AI and multimodal data processing provides unique technical advantages for digitizing, valuing, and circulating complex real assets in the RWA field. Therefore, the layout of RWA is not just an embellishment for SenseTime, but a financial self-rescue concerning its future survival and development.

The "AI + RWA" value framework as the core competitiveness of enterprises.

SenseTime's true moat lies not only in its outstanding AI technology but in the complete closed loop it has built to engineer, productize, and ultimately monetize cutting-edge AI capabilities. This unique "AI + RWA" strategic approach gives it the potential to play a triple role as a "value discoverer, risk pricer, and asset generator" in the field of Real World Assets (RWA), thereby breaking through the traditional business model of technology services and opening up new space for value growth.

Regulatory uncertainty is also a potential challenge. Hong Kong's policies are relatively open, but there are still differences compared to the strict delineation of asset nature and securities boundaries practiced by the United States. Enterprises and investors need to follow the long-term game between regulatory attitudes and market outcomes.

Activate dormant physical assets

The first step of SenseTime's strategy is to use its multimodal AI capabilities to conduct precise value assessments of physical assets that are difficult for the traditional financial system to penetrate. Its "Riri Xin SenseNova" large model system, with its powerful ability to process cross-modal information such as language, images, and time series data, provides a unique solution for this.

A typical application scenario is the asset valuation of palm oil plantations in emerging markets such as Southeast Asia. These assets often struggle to obtain traditional credit support due to their remote geographical locations and lack of financial data. SenseTime's AI valuation model can break through this bottleneck by transforming the future income rights of the plantations into investable RWA products. This model will integrate a complex data matrix: utilizing SenseEarth's intelligent remote sensing capabilities to analyze vegetation health and climate patterns through high-resolution satellite imagery to predict yields; combining IoT sensors deployed in the field to monitor key production factors such as soil and environmental conditions in real-time; and leveraging SenseChat's natural language processing capabilities to deeply analyze global trade reports and market futures prices.

In this way, "Daily Renewal" can transform vast amounts of unstructured data into dynamic, credible valuation models that reflect the intrinsic value and potential risks of assets in real-time. This not only opens up innovative financing channels for agricultural assets that have traditionally been difficult to standardize and liquidate, but more importantly, it demonstrates to the market the core capability of AI as a "value discoverer"—allowing dormant data to speak and making invisible asset values manifest.

Create a native category of digital assets

After proving its value discovery capabilities, SenseTime's strategy naturally evolved to the second stage: shifting from assessing existing assets to creating entirely new digital-native assets. Its industry-leading AIGC tools, such as the text-to-image model "SenseMirage" and the multimodal short video creation tool "Seko", laid a solid foundation for this.

A forward-looking business idea is that SenseTime can build a "SenseDrama AI Creation Platform" to empower creators to generate AIGC short dramas at low cost and high efficiency. The platform not only provides creation tools but also incorporates AI evaluation models to sift through a vast array of works to identify high-quality content with significant commercial potential. Subsequently, the future revenue rights of these selected short dramas (such as advertising revenue sharing, copyright sales, etc.) will be packaged and tokenized, becoming a new type of RWA product to be offered to global investors.

The strategic significance of this move lies in its unprecedented deep integration of content creation with the financial market. SenseTime's role has shifted from a technology service provider to a "generator" and "enabler" of digital assets. The company no longer relies solely on the slow software-as-a-service (SaaS) charging model, but is able to directly benefit from the value growth of digital content assets generated by its technology. This not only realizes the direct capitalization of technological investment but also provides a new financing paradigm for the entire creative industry.

Revitalize computing power infrastructure

The third step of the strategy, and the most profound one, is to turn our gaze inward and financialize our most core and heavy asset - the SenseCore AI large device. With a massive computing power cluster of tens of thousands of GPUs, it is both the cornerstone of SenseTime's technological leadership and a heavy asset on its financial statements. Tokenizing it is a strategic move to address this challenge.

By issuing the "SenseCore Compute Token (SCT)", SenseTime can transform standardized computing power usage rights (such as "one hour of A100 GPU inference power") into a financial product that can circulate on compliant digital asset trading platforms. The pricing of SCT will be dynamically adjusted by an algorithm based on factors such as market supply and demand, electricity costs, and equipment depreciation. For SenseTime, pre-selling SCT enables the monetization of future computing power services in advance, obtaining valuable cash flow to support research and development as well as expansion. For the market, it significantly lowers the threshold for small and medium-sized enterprises to access top-tier AI computing power, promoting the inclusiveness of AI technology.

This initiative marks the transformation of SenseTime's business model from a cost center to a profit center. It not only revitalizes the massive fixed asset investments but, more importantly, achieves the most efficient allocation of computing power resources through a market-oriented trading mechanism, directly converting the company's engineering capabilities into highly liquid financial assets.

It can be seen that SenseTime's moat is not a single technological node, but a complete value chain of "engineering-productization-financialization" that it has carefully built. The engineering capabilities of SenseCore construct the underlying infrastructure, the productization capabilities of the SenseNova family turn technology into applications, while the financialization capabilities of "AI+RWA" are the key to realizing value leaps.

This strategic closed loop enables SenseTime to perfectly play the triple core roles of the RWA era:

Value Discoverer: Using AI to penetrate the information fog to price the dormant assets of the physical world.

Asset Generator: Utilize AIGC to create native assets in the digital world and financialize their rights to earnings.

Risk pricing agent: Provides precise risk assessment for the above two types of assets through real-time data and dynamic models.

Through this interconnected strategic framework, SenseTime is not only expected to address current financial challenges but also has the potential to reshape its valuation logic, evolving from a leading AI software company into an indispensable asset value creation and trading platform in the next generation of the digital economy.

What is the outlook? A simple execution path and risk scenario.

Although this path has broad prospects, its success depends on whether the company can clearly recognize and navigate a series of profound internal challenges and external compliance requirements. This is not only a business expansion but also a comprehensive test of strategic determination, organizational capability, and market trust.

Core challenge: Strategic balance between focus and opportunity

The core challenge that SenseTime currently faces stems from the intrinsic tension between its identity as an AI leader and the financial attributes of RWA.

The company is fully committed to the core battlefield of the "Hundred Model War," and its "Daily New" large model system is in a critical window of rapid iteration, where any deviation of resources could lead to high opportunity costs. Against this backdrop, the investment of capital and elite teams into the entirely new field of RWA must carefully weigh its potential dilution effect on the main business. This is not only a matter of fund allocation but also a pull on organizational energy.

The deeper challenge lies in the expansion of capability boundaries. SenseTime's pride in its AI research strength belongs to a different field compared to the financial, legal, and compliance expertise required by RWA. Building an RWA product necessitates compliance officers who are well-versed in the regulatory framework of the Hong Kong Securities and Futures Commission, experts proficient in the design of financial product structures, and experienced licensed personnel. This lack of "financial gene" cannot be easily compensated for in the short term through technological advantages; it constitutes the most realistic gap between technological concepts and compliant products.

Ultimately, all challenges will converge on a fundamental issue: trust. When AI models are used as the core for pricing hundreds of millions of dollars in assets, their reliability becomes the ultimate concern of the market.

The "hallucination" risks of AI and the "black box" nature of the decision-making process are inherently contrary to the principles of transparency, interpretability, and auditability in the financial sector. Therefore, SenseTime must not only prove the advanced nature of its AI technology but also establish a mechanism that allows the market, investors, and regulatory bodies to trust the fairness and robustness of its "AI oracle."

This requires the co-evolution of technology and systems, for example, combining its own AI model with decentralized oracles like Chainlink, introducing third-party independent audits, and actively embracing the regulatory sandbox of the Hong Kong Monetary Authority for pilot testing, gradually building market confidence through transparent practices.

Realistic path: cautiously leveraging the Hong Kong ecosystem to start.

Given the reality of SenseTime being "technologically strong, capital-tight, and without financial licenses," attempting to independently apply for financial licenses such as Virtual Asset Service Provider (VASP) is undoubtedly a long and uncertain road. A more pragmatic and risk-controlled initial path is to choose to become a "technology enabler" of Hong Kong's mature financial ecosystem, rather than a direct financial participant.

The most feasible strategy is to establish deep cooperation with local licensed institutions, such as forming joint ventures or establishing strategic partnerships. By collaborating with financial institutions or virtual asset platforms that hold relevant licenses, SenseTime can cleverly circumvent licensing barriers, leveraging the compliance systems and expertise of its partners to compensate for its own shortcomings in financial operations. The collaboration between Ant Group and Hong Kong Victory Securities has already provided a successful example of this model.

In this process, the role of strategic investor Infini Capital is crucial. As an investment institution focused on Web3, its value goes far beyond financial support. It should be seen as the "pathfinder" and "catalyst" for Sensetime's entry into the Hong Kong Web3 and financial ecosystem, able to introduce key partners and provide deep industry insights for its strategy. This collaborative model allows Sensetime to focus on its core strengths—providing top-notch AI technology solutions, such as asset valuation models, risk assessment engines, and smart contract audit tools—thus enabling it to penetrate the core links of the RWA value chain while avoiding direct regulatory friction.

The evolution from enabler to infrastructure

Based on the above analysis, SenseTime's RWA strategy should be gradual, verifying its maximum value with minimal risk.

The first step should focus on launching a benchmark project for "AI-powered RWA Asset Valuation-as-a-Service." SenseTime should collaborate with a local licensed asset management or evaluation agency to provide its AI dynamic valuation and risk monitoring model for asset classes with high data availability and existing compliance practices, such as renewable energy facilities, logistics assets, or carbon credits. In this model, SenseTime acts purely as a technology service provider, delivering API interfaces or software services to assist partners in improving the efficiency and accuracy of asset valuation. The brilliance of this strategy lies in the fact that it can fully showcase the unique value of its multimodal AI in a real scenario with controllable risks, while also gradually building market trust in the reliability of its AI models, laying the foundation for deeper future involvement.

With a long-term vision, SenseTime should strive to become an "AI infrastructure provider" in the RWA field. This positioning is consistent with the company's core strategy of "large devices - large models - applications" and is a natural extension of its "AI-as-a-Service" concept in the fintech sector. Just as Amazon AWS became the underlying infrastructure of the internet era, SenseTime's ultimate goal is to provide underlying AI valuation, risk control, compliance review, and intelligent matching services for the entire RWA ecosystem.

This evolutionary path began with asset assessment services, gradually expanding to AI-driven compliance and audit services, and finally to smart trading and liquidity optimization solutions, ultimately forming a dedicated AI Agent platform for RWA. This path not only has vast market potential but, more importantly, it allows SenseTime to maximize the value of its technology reuse while remaining rooted in the fundamentals of its technology company, constructing an insurmountable moat in the next generation of digital financial ecosystems with a low regulatory risk model.

Author: Zhao Qirui Editor: Zhao Yidan