Uweb In-Depth Report: Global AIDC Transformation Cycle and Valuation Reconstruction Logic of Listed Companies

Uweb Research: Analysis of the Effects of Global Listed Companies Transforming to AIDC (Intelligent Computing Centers)

Co-produced by: Uweb × The Hong Kong Polytechnic University Business School TGG Stablecoin and RWA Innovation Center

1. Global AIDC Enters a Super Construction Cycle

AIDC is a physical expansion that has been locked in by orders and power, with demand certainty significantly higher than that of general technology themes, which is the premise for judging whether listed companies can truly benefit.

1.1 Growth Drivers Have Shifted from Internet Traffic to AI; Whoever Holds AI Capacity Holds Incremental Growth

The essence of AIDC growth is a physical expansion measured in GW, where AI has transformed from a supporting role to the sole main engine of incremental growth. JLL's 2026 Global Data Center Outlook predicts that the 103GW of 2025 will double to 200GW by 2030, with AI loads expected to account for approximately 50% of data center capacity by 2030, up from about 25% in 2025. According to McKinsey's estimates, global data center total demand will increase from 82GW in 2025 (44GW for AI and 38GW for non-AI) to 219GW in 2030 (156GW for AI and 64GW for non-AI), with AI capacity growing 3.5 times over five years and accounting for about 70% of total demand by 2030.

In the next five years, nearly all incremental data center capacity will come from AI, which means the growth logic of the industry has shifted from traffic-driven in the internet era to AI training and inference-driven. For listed companies, whether they can secure AI-ready capacity and orders directly determines whether they capture incremental or existing growth.

At the same time, trillions of dollars in investment mean that the entire industry chain will be re-priced. According to McKinsey's April 2025 report, by 2030, global data centers are expected to require $6.7 trillion to meet computing capacity demands. Data centers capable of handling AI workloads are expected to require $5.2 trillion in capital expenditure, while traditional IT application data centers are expected to need $1.5 trillion. Overall, by 2030, this amounts to nearly $7 trillion in capital expenditure, which is astonishing regardless of the measurement standard used.

From a market size perspective, the AIDC industry has entered a phase of rapid growth. In 2024, China's AIDC market size is expected to be approximately 49.4 billion yuan, with Kezhi Consulting predicting it will grow to 196.3 billion yuan by 2027, corresponding to a compound annual growth rate of about 58%. Structurally, the computing power leasing market is expected to grow from 41.5 billion yuan to 152.8 billion yuan, continuously increasing its share and becoming an important component of industry growth; the intelligent computing infrastructure market is expected to grow from 7.9 billion yuan to 43.5 billion yuan, maintaining steady expansion. Against the backdrop of rapidly growing AI computing power demand, the service delivery model is accelerating its penetration, and the AIDC industry is evolving from a focus on infrastructure construction to a model of "computing power services + infrastructure collaborative development."

In June 2026, Li Chao, Deputy Director of the Policy Research Office and Spokesperson of the National Development and Reform Commission, stated at a press conference that during the 14th Five-Year Plan period, more attention will be paid to supply-demand matching and strengthening the coordinated planning and construction of computing networks with new power grids and new generation communication networks. In terms of hard investment, more effective computing-power and electricity collaboration models will be explored to strengthen computing with electricity and promote electricity with computing; innovation in computing network integration will be strengthened, and the expansion of direct connection lines between national hubs will be appropriately promoted to further reduce network transmission delays. In terms of soft construction, monitoring and market-oriented scheduling of computing power resources will be strengthened, and the construction of a nationwide integrated computing power network that is interconnected, user-friendly, green, and safe will be accelerated. The NDRC also announced that the complete list of the 200 billion yuan equipment renewal projects will be officially issued by the end of June 2026. This funding is an important part of the "Two New" policy, which has previously issued 185.1 billion yuan in two batches, benefiting over 11,000 projects and driving total social investment of over 840 billion yuan.

1.2 AIDC Demand Has Been Verified by Real Money

The difference between AIDC and most technology concepts lies in the fact that its demand is not predicted but is already reflected in the established facts of cloud vendor expenditures and Nvidia's revenues, elevating the industry's certainty from narrative to orders. The Dell'Oro Group reports that the four major U.S. hyperscale vendors—Amazon, Google, Meta, and Microsoft—are entering 2026 with nearly $600 billion in data center capital expenditures, with global data center capital expenditures expected to approach $1 trillion for the year. Nvidia's revenue for the 2026 fiscal year is projected to be $215.94 billion, a 65% increase from the 2025 fiscal year, with data center business revenue for the year reaching $193.7 billion, a 68% year-on-year increase, accounting for over 90% of the company's total revenue. Revenue for the first quarter of the 2027 fiscal year is approximately $75.2 billion, with a quarter-on-quarter increase of about 21% and a year-on-year increase of about 92%.

On one end, there is a commitment of nearly $600 billion in spending from cloud vendors, and on the other end, Nvidia's 92% growth rate in data center revenue mutually confirms each other, indicating that the money on the demand side is already being spent and goods are being delivered. This means that for downstream AIDC operators, the question is whether they can catch this wave of demand.

1.3 Energy is a Real Physical Bottleneck

With capital not being a constraint, power and grid connectivity have become the real bottlenecks for AIDC expansion, shifting the competitive edge from who has money to who has electricity and land. In terms of electricity demand, the International Energy Agency (IEA) reports that global data center electricity consumption will more than double to approximately 945 terawatt-hours by 2030, approaching Japan's total electricity consumption; AI-specific data centers will see electricity consumption grow more than fourfold, with U.S. data centers expected to account for nearly half of the country's electricity increase before 2030. Therefore, electricity has become the real constraint for AIDC implementation, explaining why companies that first secure electricity and land are gaining an advantage in this round of transformation.

SemiAnalysis predicts that the compound annual growth rate of the intelligent computing center infrastructure market revenue will exceed 30% from 2024 to 2032. To meet the continuously exploding demand for AI computing power, operators and large cloud service providers are actively seeking areas rich in land and electricity resources to expand their businesses, further promoting the intensive and large-scale development of data centers. Although the growth rate of general-purpose data centers is not as high as that of intelligent computing centers, they still maintain stable growth under the overall drive of the intelligent computing ecosystem, reflecting the strong radiation force of intelligent computing demand on AIDC computing power scale from core to periphery.

2. Classification of Listed Companies' Transformations and Regional Differences

2.1 The Gold Content of Transformations Among Listed Companies in Mainland China, Hong Kong, and the U.S. Varies

Although all three regions discuss AIDC transformation, the gold content is systematically different.

• U.S. listed companies primarily transform from Bitcoin mining enterprises to AI hosting, as mining companies already possess electricity, facilities, and cooling, making the transition to AI a reuse of existing computing power facilities;
• Mainland Chinese listed companies often start from scratch in industries unrelated to IT, such as monosodium glutamate, papermaking, steel, lottery, children's clothing, and furniture;
• Hong Kong listed companies are positioned in the middle, primarily focusing on real estate cross-industry and upgrading native IDC.

2.2 Overview of Four-Level Classification

Based on the degree of transformation realization, this report categorizes listed companies in mainland China, Hong Kong, and the U.S. into four levels of transformation targets, separately listing native IDC and industry references.

1. Transformation effects have already emerged: AIDC revenue has been consolidated and accounts for a significant proportion or has already turned profitable, represented by companies like Hengrun Co., Ltd., Zhongbei Communication, and Meili Cloud.

2. Existing facilities are upgraded accordingly: Rapidly entering AI hosting by leveraging existing electricity and facilities, represented by companies like IREN, TeraWulf, Core Scientific, Hut 8, Applied Digital, Cipher, and Galaxy.

3. Cultivation period for layout: Computing power business is still in its early stages, with limited revenue share or related assets still in progress, represented by companies like Lianhua Health, Hangang Co., Ltd., Gaoxin Development, and Guangdong-Hong Kong-Macao Intelligent Computing.

4. Cautious adjustment period: Some enterprises have adjusted or postponed their computing power layout after evaluation, reflecting the normal trial and error of the early stage of the industry.

The fifth category includes native IDC and industry references. In addition to the aforementioned four levels of transformation targets, the unified reference table also includes a number of native computing power operators, cloud vendors, and AI platform companies for comparison. Most of them do not belong to specialized cross-industry transformations but are natural upgrades and computing power deployments of their main businesses, included to provide a more complete reference coordinate system.

Runze Technology, Aofei Data, Data Port, Guanghuan New Network, Baoxin Software, Kehua Data, and Wangguo Data are all data center operators that are upgrading from traditional IDC to high-power liquid-cooled AIDC capabilities, rather than cross-industry, included to anchor the real performance and listing levels of the industry's first tier; CoreWeave, Nebius, and other native AI clouds also fall into this category.

Alibaba, Tencent, Microsoft/Google/Amazon/Meta, Oracle, and the three major telecom operators primarily build computing power to serve their own cloud and AI businesses, representing natural deployments of existing businesses rather than transformations. They are the financial backers on the demand side and the main force behind AIDC capital expenditures, included to observe upstream investment intensity. Companies like SenseTime and Fourth Paradigm extend from AI algorithms and platforms to computing power operations, positioned between software and infrastructure, representing an extension down the business chain.

The purpose of placing these categories alongside transformation targets is to provide a complete reference, facilitating the distinction between natural upgrades, cross-industry layouts, and early cultivation paths, rather than suggesting that they all belong to transformation.

The participants of mainland Chinese companies in AIDC are diverse, ranging from specialized operators to cross-industry enterprises, with significant differences in the pace of realization. From an industrial research perspective, the key measure is whether AIDC revenue is consolidated, its proportion, and the quality of profits. On one end where results are evident, Hengrun Co., relying on dual engines of wind power and computing power, achieved a net profit of 83.48 million yuan in 2025, turning losses into profits, with its computing subsidiary Shanghai Runliuchi's revenue growing by 743.60% year-on-year. In the first quarter of 2026, the net profit reached 65.14 million yuan, a year-on-year increase of 117.90%.

The paper company Meili Cloud reported a cloud business revenue of 324 million yuan in 2025, accounting for 94.65% of total revenue, with a gross margin of 45.58%; its quarterly report showed a year-on-year net profit increase of 102.61% and positive operating cash flow.

Another typical case is represented by Zhongbei Communication, which had a computing order reserve of 2.87 billion yuan in the first quarter of 2026, with rapid growth in intelligent computing revenue. However, due to depreciation, financial costs, and impairment, profits were under pressure during the period. This reflects a common pattern during periods of heavy asset expansion, where profits often lag behind revenues; the faster the construction pace, the greater the pressure on current profits.

Some cross-industry companies still have a low proportion of computing revenue in their overall revenue, or the relevant asset injections and collaborations are still ongoing. Their computing business has not yet made a significant contribution to overall performance, reflecting more of the market's expectations of the concept, and future realization still needs to be observed, which is a common characteristic of the early stage of the industry.

Some early participants have also experienced adjustments at the cooperation or project level, reflecting the uncertainties in the implementation of cross-industry layouts.

2.4 Hong Kong-listed Companies: Focus on Realizing Orders from Native Leaders, Diverse Paths for New Entrants

The value of AIDC for Hong Kong-listed companies is concentrated on the realization of orders from native IDC leaders. GDS Holdings signed a record 200MW in new contracts in the first quarter of 2026, with a total order backlog of 1.8GW by the end of the first quarter, and a target of over 500MW for new AI orders in 2026. On the cross-industry side, Guangdong-Hong Kong-Macau Holdings changed its name to Guangdong-Hong Kong-Macau Intelligent Computing in 2026, entering the computing power sector through the acquisition of Tiandun Data, with AI computing revenue accounting for 61.5% of its revenue in 2025.

GDS Holdings' 1.8GW order backlog represents verifiable real demand and is a typical example of stable realization; Guangdong-Hong Kong-Macau Intelligent Computing represents a new entrant path that quickly enters through acquisitions and cooperation with state-owned assets. Although the proportion of computing revenue is high, the business has been established for a short time, and its sustainability still needs to be verified over time. Together, they reflect the diverse landscape of participants in Hong Kong-listed companies, from mature leaders to new entrants.

2.5 U.S.-listed Companies: Mining Enterprises are the Most Mature in Transformation, Contracts First, Revenue Follows

U.S.-listed mining companies are the most mature in transformation among the three regions, as they follow a path of signing long-term contracts first, then increasing capacity, with revenue realized subsequently. The valuation anchor is based on the contracts in hand rather than current profits. IREN signed a $9.7 billion GB300 AI cloud contract with Microsoft; AWS signed a 300MW, 15-year hosting agreement with Cipher. Core Scientific's collaboration with CoreWeave expanded to approximately 590MW, with a total of about $10.2 billion in take-or-pay hosting contracts over 12 years.

The long-term take-or-pay contracts signed with investment-grade counterparties like Microsoft, AWS, and CoreWeave essentially convert the mining companies' power facilities into predictable long-term cash flows. This is why U.S.-listed mining companies have relatively high intrinsic value; they sell capacity that has already locked in buyers. A quantitative proof is that the revenue per megawatt from AI contracts is about three times that of traditional mining.

3. Judgment: Before and After Transformation to AIDC

3.1 Valuation Anchor Migration

Before transforming to AIDC, the valuation anchor of companies was their original main business, with MSG and paper looking at consumption and capacity cycles, steel looking at bulk cycles, and lottery printing looking at licenses. After transforming to AIDC, for companies that have realized transformation well, the valuation anchor has shifted to verifiable long-term computing contracts and power capacity. GDS Holdings is revalued based on its 1.8GW order backlog, while U.S.-listed mining companies are revalued based on backlog and ARR, rather than current earnings per share.

3.2 The Real Difference Lies in Contracts and Power

The real distinction between before and after transformation is whether the company has secured verifiable long-term contracts and power indicators, rather than whether it has publicly announced computing power. Some companies have experienced significant stock price fluctuations after announcing their computing power layouts due to low revenue proportions or slower-than-expected project advancement; in contrast, companies with power, land, and long-term contracts have received relatively sustained valuation support. Power is the real bottleneck under IEA standards, which also explains why the path of securing power and land first, then discussing AI stories, is easier to realize.

3.3 Profit Lag is a Common Characteristic

Whether it is Zhongbei Communication in mainland China or CoreWeave and IREN in the U.S., periods of heavy asset expansion show rising revenues, pressured profits, or even losses. The capital market's tolerance for this is based on the visibility of future cash flows provided by long-term contracts. Once long-term contracts do not materialize or power indicators are lacking, the concept premium will quickly evaporate.

The capital market highly recognizes the AIDC concept, but the recognized objects are contracts and power. For example, GDS Holdings, CoreWeave, and a number of mining companies have received significant revaluation based on their backlog, with mining stocks rising by about 70 percentage points in 2026; Oracle completed a market capitalization revaluation from database software to AI cloud based on its hundreds of billions of dollars in RPO backlog.

4. Risks and Sustainability

This chapter outlines the main risks in the AIDC investment cycle from an industrial research perspective, for reference when assessing the sustainability of industry prosperity. The current prosperity is based on the assumption of sustained high growth in AI training and inference demand. If the commercialization of large models does not meet expectations, or if improvements in chip and algorithm efficiency significantly reduce unit computing power demand, the industry may face phase-specific capacity digestion pressures.

4.1 Mismatch of Depreciation and Technological Iteration

One of the core financial risks of AIDC comes from the mismatch between GPU depreciation cycles and accounting depreciation assumptions. According to public industry discussions, some computing power operators depreciate GPUs over approximately six years, while engineering and legal estimates of the actual usable life of GPUs are often between three to four years, with some analysts believing it to be only two to three years. As NVIDIA iterates from Blackwell to Vera Rubin, the residual value and rental levels of the previous generation of computing power may decline faster than the depreciation schedule; if the actual lifespan is shorter than the accounting assumptions, the true returns for operators will be overestimated. This is a variable that needs to be carefully verified when assessing the profit quality of heavy asset operators, with a degree of confidence.

4.2 Customer Concentration and Contract Stability

Long-term contracts (take-or-pay) provide predictable cash flows for project financing, but they also bring customer concentration risks. OpenAI's external computing power procurement commitments include approximately $22 billion to CoreWeave, about $300 billion to Oracle, and about $38 billion to Amazon; many Neocloud and transformed mining companies' revenues are similarly highly dependent on a few investment-grade counterparties. The stability of long-term contracts may be tested in changing market environments; once individual large customers adjust their demand pace, the revenue visibility of related operators may fluctuate.

4.3 Leverage and Circular Financing

AIDC is capital-intensive, and the scale of debt financing is rapidly expanding. CoreWeave completed a financing of approximately $8.5 billion in 2026 and received an investment-grade rating, being one of the first investment-grade financings supported by HPC infrastructure, indicating the credit market's initial recognition of this model. At the same time, there is discussion in the market about circular financing; NVIDIA holds about 7% of CoreWeave's equity and has committed up to $100 billion in investments to OpenAI, with some funds ultimately flowing back upstream through GPU procurement, which is similar to the supplier financing of telecom equipment manufacturers at the end of the last century.

5. Financing Innovation: REITs and Asset Securitization

AIDC is a typical heavy asset industry, and relying solely on self-owned funds and bank loans is insufficient to support exponential expansion, making financing models a key variable determining the speed of operators' expansion. Asset securitization (REITs, ABS, CMBS) allows operators to offload and recoup funds from mature assets that have been built and put into operation, reinvesting them into new construction, forming a cycle of "construction - securitization - reinvestment," promoting the industry's evolution from a heavy asset model to a light asset operation.

At the same time, the China Securities Regulatory Commission approved the first two public REITs for data centers on June 18, 2025, namely the Southern GDS Data Center REIT and the Southern Runze Technology REIT. The underlying assets of the former are GDS's data center projects located in Kunshan, while the latter is the ICFZ A-18 data center project located at the national hub node in the Beijing-Tianjin-Hebei region. This marks the first time that data centers have been included in public REITs in the domestic capital market, which is significant for the asset valuation and expansion capabilities of third-party IDC leaders; the first-mover advantage of Runze and GDS in REITs is also an important support for their relatively stronger expansion capabilities compared to peers.

The issuance scale of debt securitization (ABS/CMBS) in the U.S. AIDC is rapidly growing, and some operators have also achieved investment-grade rated financing supported by HPC infrastructure. According to JPMorgan's forecast, the annual securitization issuance scale of U.S. data centers is expected to reach $30 billion to $40 billion in 2026 and 2027. The maturity of financing channels is one of the important reasons for the differences in expansion pace between operators in China and the U.S.

Beyond asset-side securitization, the financialization of computing power is further extending to the output side. The Chicago Mercantile Exchange and Silicon Data announced in May 2026 the launch of the world's first GPU computing power futures, anchoring the daily GPU leasing price benchmark, allowing operators and computing power buyers to hedge against fluctuations in computing power prices, similar to hedging oil and electricity prices; in China, the approach is steadily advancing with indices leading, spot pilots, and policy guidance. By the end of 2025, the China Securities Index Company released the Intelligent Computing Supply Index, and in April 2026, the Ministry of Industry and Information Technology proposed exploring businesses such as computing power banks and computing power supermarkets. CITIC Securities predicts that computing power futures may land within the year. While REITs and ABS financialize completed assets, computing power futures financialize the computing power revenue itself, forming two layers of parallel advancement in the financialization of computing power along the asset and output sides; once pricing and hedging tools on the output side mature, they will further reduce the uncertainty of operators' revenues and enhance the valuation and issuance capabilities of asset-side securitization.

However, the realization of computing power futures is still constrained by standardization, as computing power lacks a unified reference price due to differences in chip models, computing precision, and network architecture, and pricing transparency and delivery mechanisms have yet to be resolved. Its more accurate positioning is as an emerging tool expected to land within the year, still awaiting preconditions.

6. Conclusion

According to JLL's 2026 Global Data Center Outlook, the proportion of AI workloads in total data center demand is expected to rise from about 25% in 2025 to about 50% by 2030, with a structural turning point anticipated around 2027, when inference workloads will surpass training and become the main driver of AI computing power demand. Current demand remains primarily focused on large-scale centralized training.

The requirements for infrastructure differ between training and inference. Training tends to be concentrated, ultra-large scale, and has extremely high single-point power density; inference, on the other hand, needs to be close to users and low-latency, which makes it more decentralized and drives regional deployments and edge computing, such as the growth of micro data centers and edge colocation. This means that the competitive factors for AIDC will gradually shift from single-point scale to networked layout and latency coverage.

For operators, the migration of inference benefits players with multiple regional nodes that are close to core economic areas and users; for the device chain, the requirements for energy efficiency and heat dissipation in inference will further elevate the penetration of liquid cooling and high-efficiency chips. From an industrial research perspective, this is a key variable in determining the next phase of AIDC investment direction, shifting from building large clusters to deploying networks.

This report and related research are intended solely for AIDC industry research and transformation path analysis and do not constitute any investment advice or trading recommendations for securities, nor do they predict the future performance, valuation, or secondary market performance of any company. The secondary market performance information listed in the charts of this report is for industry observation reference only and may be subject to delays or discrepancies. Readers should verify and make independent judgments, and any decisions made based on this table are not related to the author.

Disclaimer: This content is provided for general branding and informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online events, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets or to use any services. Crypto assets are highly volatile and may result in loss. WEEX services and online events may not be available in all regions and are subject to applicable laws, regulations, and eligibility requirements. You are responsible for ensuring that your use of WEEX services complies with local laws and for carefully assessing the risks before participating in any crypto-related activities.