China's listed companies have rushed to position themselves as providers of AI computing power, but their 2026 interim results show that businesses carrying the same “compute” label can have very different economics.


A review of ten companies associated with compute leasing or computing services — including Xiechuang Data, Dongyangguang, Range Technology, Hongxin Electronics, J-Win Intelligent, Yunsai Zhilian, GBA Digital Intelligence, Wangsu Science & Technology, Lettall Electronic and Maxs Technology — reveals sharp differences in revenue mix, asset intensity, order conversion and profitability.


The comparison also shows why headline contract values and installed capacity should not be treated as equivalent to recognized revenue.


Scale Does Not Always Mean Compute Exposure

Xiechuang Data reported the largest overall revenue among the group, but a substantial share of its business still comes from hardware products, storage and server-related manufacturing or trading. Dongyangguang also reported a large revenue base, while AI-related activity accounted for only a small portion of its total business.


By contrast, several smaller companies reported much higher exposure to computing services. Public disclosures cited in the comparison indicate that AI accounted for more than 90% of revenue at GBA Digital Intelligence, while compute services represented roughly 92% at Maxs Technology. Range Technology's AIDC business contributed more than half of its revenue, and compute-related services made up a majority of Lettall Electronic's relevant business mix.


These differences matter because equipment sales, systems integration, data-center operations and GPU leasing produce different margins, cash-flow profiles and levels of recurring revenue.


Orders, Capacity and Revenue Are Different Metrics

The compute-leasing market frequently uses several incompatible measures of scale. Some companies disclose computing capacity in FP16 petaflops, while others report data-center capacity in megawatts or numbers of racks. Contract announcements may refer to framework agreements, intended orders, delivered capacity or completed settlement.


Hongxin Electronics, for example, reported RMB 9.744 billion in cumulative signed computing-resource service orders in its interim disclosure, but RMB 1.813 billion had been settled. The gap illustrates the time required for procurement, delivery, acceptance and billing before an order becomes reported revenue.


Range Technology, meanwhile, represents a more conventional data-center model. Its AIDC operations are backed by heavy infrastructure investment and long delivery cycles, creating relatively visible operating assets but also requiring sustained capital expenditure.


The distinction is important for overseas observers assessing China's AI infrastructure buildout: announced orders indicate demand, but they do not by themselves demonstrate utilization, collection of cash or profitability.


From Renting Hardware to Selling Tokens

Several companies are trying to move beyond rack- or card-based leasing toward token-based services and model-as-a-service platforms.


Hongxin Electronics is developing what it calls a “Token Factory” alongside AI infrastructure projects. J-Win Intelligent is combining an intelligent-computing center with token services and cloud inference. GBA Digital Intelligence is building a quantitative computing platform that charges by card usage and token consumption.


This transition reflects a broader change in the market. Selling raw capacity leaves providers exposed to GPU prices, utilization and financing costs. Token-based delivery could allow them to package infrastructure, model access and software services into a higher-level product. However, most of these initiatives remain early, and their economics have yet to be demonstrated over a full reporting cycle.


A Capital-Intensive Expansion

The sector's main constraint is financing. Data centers, GPUs and power infrastructure require large upfront expenditure, while customer acceptance and settlement can take months. Companies with high leverage may achieve rapid expansion, but rising interest expenses can absorb much of the operating gain.


The interim reports therefore suggest that the most useful indicators are not headline capacity alone, but utilization, recognized service revenue, operating cash flow, financing cost and the proportion of recurring business.


China's compute-leasing sector is clearly expanding, but it is not a single business model. Some participants are primarily manufacturers or equipment distributors, others operate heavy data-center assets, and a smaller group is attempting to become token-based computing platforms. The second half of 2026 will test which companies can turn announced capacity and signed orders into durable revenue.


This article is based on public company disclosures and is intended for industry analysis only. It does not constitute investment advice.