The global capital expenditure cycle for artificial intelligence infrastructure is creating a vast new market for the property insurance industry, fundamentally reshaping the risk pool for insurable assets.
A new capital expenditure cycle for artificial intelligence is underway. Goldman Sachs reports that AI investment focus is shifting from model development to broader economic applications. As computing power, electricity, and data centers are rapidly built, AI is accelerating into manufacturing, energy, logistics, defense, life sciences, and robotics. Goldman Sachs projects global AI capital spending on computing, data centers, and electricity will reach approximately $7.6 trillion between 2026 and 2031.
China is rapidly entering this AI infrastructure buildout cycle. By early 2026, daily token calls for Chinese AI models exceeded 140 trillion, a roughly 1,000-fold increase from two years prior. The nation has over 500 intelligent computing centers in operation, construction, or planning, with a smart computing power capacity of 1.882 million PFLOPS (FP16), ranking second globally. During the 15th Five-Year Plan period, China's computing infrastructure investment is expected to be around 2 trillion yuan, rapidly forming a new digital infrastructure system encompassing intelligent computing centers, GPU clusters, and computing networks.
As AI enters this capital expenditure era, a previously non-existent asset pool of AI infrastructure is quickly forming. This multi-trillion-dollar investment creates not only new industrial opportunities but also a massive accumulation of high-value assets. For the insurance industry, a new risk pool is emerging, yet a corresponding risk protection system is still under development.
AI Enters a Capital Expenditure Era, Forming a New Asset Pool
Is China forming a sufficiently large AI infrastructure asset pool? And is this pool large enough to alter the insurance industry's long-standing risk structure? The answer is becoming clear.
As large model applications expand, the scale of AI model calls is growing rapidly. Data shows daily token calls for domestic AI models surged roughly 1,000-fold in two years, from 140 billion in early 2024 to 140 trillion by March 2026. AI is transitioning from a technical experiment for a few tech companies to large-scale commercial applications in finance, manufacturing, energy, and healthcare. Each time a technology moves from experimentation to industrialization, an infrastructure investment cycle begins.
In recent years, competition in the AI industry centered on model capabilities, with ever-larger parameters and continuous algorithmic breakthroughs. However, as generative AI enters a stage of scaled application, the competitive landscape is shifting from a "model competition" to an "infrastructure competition."
The key resources supporting AI are moving from software capabilities to a more asset-heavy infrastructure system. Training large models, running inference services, and enabling enterprise AI applications all require continuous, massive computing resources. Computing power is becoming a new type of infrastructure, akin to electricity, communications, and transportation in the AI era. By March 2026, China's smart computing power reached 1.882 million PFLOPS (FP16), a 2.5-fold increase year-on-year. China's computing market size is expected to reach 835.1 billion yuan in 2025 and potentially exceed one trillion yuan in 2026. Concurrently, the number of intelligent computing centers in operation, construction, or planning has surpassed 500, with an increasing number of ten-thousand-card GPU clusters.
Unlike the past internet industry, which expanded rapidly through software, platforms, and traffic, AI development is distinctly capital-intensive. A large intelligent computing center is not a simple stack of servers; it is a complex infrastructure system comprising buildings, power supply systems, high-performance GPU servers, cooling systems, network equipment, and security facilities.
Globally, AI infrastructure investment is reaching unprecedented levels. Research from Swiss Re indicates that AI development is driving a new capital expenditure cycle, creating a large volume of previously non-existent insurable assets. The construction cost of some large data centers has already reached billions of dollars. After deploying high-value GPU equipment, total investment can increase further, with mega-projects potentially exceeding $20 billion.
This means AI infrastructure is forming a new, high-value asset pool. From an insurance industry perspective, this shift is significant. For decades, property insurance has centered on risk assets from the industrial age, such as factories, production equipment, commercial buildings, and energy/transportation facilities. The AI era is creating new risk exposures: intelligent computing centers, GPU clusters, computing networks, and the digital infrastructure systems supporting AI operations. As new, single-asset investments reach tens or even hundreds of billions of dollars, this creates not only tech industry opportunities but also a new market for risk protection.
Super Computing Assets Create Super Risks
The large-scale construction of AI infrastructure is creating a new, high-value asset pool. However, for the insurance industry, what truly matters is not just the rapid growth in asset scale, but the profound change in the risk profile behind these assets. There is a fundamental difference between AI infrastructure and traditional industrial assets: its value lies not just in the physical equipment itself, but in its ability to continuously provide computing power.
For decades, property insurance has focused on "physical asset loss." A factory fire triggers claims for buildings, equipment, and inventory. An energy facility accident involves equipment damage and repair costs. The risk logic of traditional property insurance is essentially built around fixed asset value. But for AI infrastructure, the risk is changing. The most critical asset of an intelligent computing center is not just the GPU servers deployed, but the collective computing capacity these devices create. A failure in a key system can cause massive economic loss, even without severe physical damage.
This means the question for the AI-era insurance industry is shifting from "what if the equipment is damaged?" to "what if the computing production system stops running?" AI infrastructure may face multiple risks simultaneously: asset loss, business interruption, and a fusion of physical and digital risks. An intelligent computing center consists of high-value GPU servers, power systems, cooling equipment, and network facilities. A failure in any critical component can impact the entire operation.
Compared to traditional industrial assets, AI infrastructure has higher technical complexity and systemic interconnectedness. The risk is not a single equipment failure but can involve the entire infrastructure system. Furthermore, for enterprises relying on computing services, the true loss is not just the repair cost, but the business impact from service interruption: delayed model training, halted AI applications, disrupted customer services, and contract performance risks.
Moreover, AI-era intelligent computing centers carry not just servers and data, but also core enterprise AI applications, model training tasks, and numerous commercial services. Risks from cyberattacks, data breaches, ransomware, and supply chain security are becoming major factors affecting the operation of computing infrastructure. For large centers, the impact of a cyber risk can far exceed traditional enterprise information security incidents. A single attack could halt computing services, delay model training, and force many enterprise clients to suspend their business. Therefore, the future insurance demand around AI infrastructure will not fall into a single product category. It will require an integrated approach covering property damage, business interruption, cybersecurity, third-party liability, and supply chain risks.
The rapid development of AI infrastructure is also challenging the insurance industry's long-established underwriting and pricing capabilities. When a single intelligent computing center's investment reaches tens or even hundreds of billions of dollars, the traditional risk diversification mechanisms face new challenges. Insurers must answer not just "should we underwrite this?" but more importantly, "how do we accurately assess the risk?" How to quantify the risk of a large computing center? How to combine and price physical, cyber, and business interruption risks? How to control overall industry exposure when multiple large projects face simultaneous risks?
These questions mean that insurance for AI infrastructure cannot simply copy traditional property insurance models. It requires building a new risk assessment system. Insurers need to combine engineering risk management, cybersecurity capabilities, data analysis models, and reinsurance mechanisms to enhance their ability to identify and manage risks for complex digital infrastructure.
From Risk Protection to Risk Management: Reconstructing the Insurance Ecosystem for Computing Power
AI infrastructure is creating a new, high-value asset pool, but the real challenge for the insurance industry is not just the emergence of market demand, but understanding and managing a new type of risk that didn't exist before. The traditional insurance system is built on long-term historical risk data, using past loss data to judge probability, severity, and pricing. For intelligent computing centers, GPU clusters, and computing networks, many risks are still in their early stages. The insurance industry is facing a new risk domain characterized by "high value, low historical data."
This means computing power insurance first faces a challenge in risk identification and pricing. While a large intelligent computing center has asset attributes similar to traditional infrastructure, its risk structure is far more complex. On one hand, physical risks like equipment failure, power supply, and cooling are interconnected. On the other hand, digital risks like cyberattacks, data security, and business interruption are further integrated with traditional property risks. Therefore, traditional risk assessment models based on single-asset loss are struggling to adapt to the complex risks of AI infrastructure.
For insurers, the development direction for computing power insurance is not simply adding a new product, but driving a change in the risk management model. In the past, insurance primarily served as a financial compensation mechanism after an accident. However, for a multi-billion dollar intelligent computing center, enterprises need risk management capabilities throughout the entire lifecycle of construction, operation, and service provision. Insurers need to move risk management from post-event compensation to pre-event identification, continuous monitoring, incident response, and loss control.
This trend is already evident in the cybersecurity insurance field. For example, the "prevention + insurance" integrated model pioneered by Yuanbao Technology addresses digital infrastructure risks in smart cities, cloud computing, and big data. Traditional information security can no longer cover all needs. Insurance institutions need to combine professional technological capabilities: identify risks before underwriting, continuously monitor risks during operations, and provide incident investigation, loss assessment, and recovery support after an incident to improve overall risk governance.
In the AI era, insurance faces not just a new business opportunity, but a new class of risk assets. The future insurance system around AI infrastructure will not be limited to a single product line. It will need to integrate multi-dimensional protection including property damage, engineering risk, business interruption, cybersecurity, and third-party liability. Simultaneously, the insurance industry needs to gradually build risk data systems and pricing models for AI infrastructure. Those who can first accumulate operational data, risk cases, and loss experience for computing infrastructure will be better positioned to gain pricing power in this emerging risk market.
From Risk Protection to Long-Term Capital: Insurance Unlocks a New Role in the AI Era
The development of AI infrastructure is not only creating new insurance demand but also presenting new asset allocation opportunities. For decades, insurance funds, with their long-term liability characteristics and stable funding sources, have consistently participated in long-term asset investments in transportation, energy, and traditional infrastructure. As the AI industry enters its infrastructure construction phase, new assets like intelligent computing centers, data centers, and computing networks may become important targets for long-term capital.
AI infrastructure is characterized by long investment cycles, large capital requirements, and strong industrial spillover effects. These traits align well with the long-term, value-oriented investment nature of insurance funds. In recent years, insurance funds have become a significant source of long-term capital for the AI ecosystem. On one hand, they participate in the development of upstream and downstream companies in the AI industry chain through equity investments and industry funds, providing long-term capital support for AI infrastructure construction and industrial ecosystem improvement.
On the other hand, as the scale of new assets like computing centers and data infrastructure grows rapidly, the insurance industry is also developing innovative products like computing power insurance, data center insurance, and cybersecurity insurance to provide risk protection for AI infrastructure construction. Insurance funds are gradually expanding from traditional financial investors into long-term capital partners for the development of the AI industry.
In this sense, AI brings the insurance industry not just a new product opportunity, but development opportunities at different stages on both the asset and liability sides. Insurance funds have become important participants in AI industry chain investment, supporting AI infrastructure development with long-term capital. Concurrently, the new risk protection around AI infrastructure is still in the exploration phase. The insurance industry needs to further enhance its risk identification and protection capabilities. As computing power becomes a critical means of production in the digital economy, the ability of the insurance industry to establish a risk management framework that matches the development of AI infrastructure will become a key indicator of its role in serving new productive forces.