The AI computing power sector has welcomed another company rushing toward an IPO on the Hong Kong Stock Exchange.

On September 29, DataCanvas submitted its application to list on the Hong Kong Stock Exchange.

As an established AI company founded in 2013, DataCanvas has pivoted swiftly in the AI era.

From 2023 to 2025, DataCanvas's revenue surged from RMB 106 million to RMB 1.098 billion, growing over 9x in just two years. In the first half of 2026, revenue rose by 92% year-on-year to RMB 588 million.

According to CIC data, based on China's intelligent computing service revenue in 2025, DataCanvas ranks among the top 10 domestic providers and is the only one classified as an 'intelligent computing new cloud' enterprise.

DataCanvas's rapid scale-up is attributed to its 'aggressive' model:

First, it quickly expands its controllable computing power scale through intelligent computing center construction projects and long-term leasing partnerships. Then, it attempts to convert this scale into sustained revenue and profits through software, orchestration, and cloud services.

This model has enabled DataCanvas to achieve rapid growth in revenue and computing power scale in a short time, but the cost has been significant: the company has locked in computing power costs for the coming years.

The most visible impact is reflected in lease liabilities.

At the end of 2024, DataCanvas's lease liabilities stood at RMB 187 million. By late July 2026, this figure had soared to RMB 3.087 billion, increasing over 15x in less than two years.

In the first half of 2026 alone, the company's lease liability interest reached RMB 32.46 million.

On one hand, revenue and gross margins are growing rapidly; on the other, long-term computing power leasing liabilities continue to swell. Today, Silicon-Based Insights will analyze DataCanvas, a somewhat unique AI computing power company, through its prospectus.

From Selling AI Software to Building Intelligent Computing Centers

Compared to AI computing power companies like Jiliu Technology, DataCanvas is a veritable 'veteran.'

Founded in 2013, the company initially focused on data science platforms, providing data analysis and machine learning tools primarily to enterprise clients in finance and other sectors. In simple terms, it helped enterprises leverage AI to utilize their data.

This type of business required high software expertise, but revenue growth was often constrained by factors such as enterprise IT budgets and project delivery cycles. For a long time, DataCanvas's business scale remained modest.

The real turning point came in 2023.

With the explosion of large models, the AI industry's most scarce resource shifted from algorithms and data to computing power. DataCanvas saw a new opportunity.

In 2023, the company began transitioning toward intelligent computing infrastructure, undertaking intelligent computing center construction projects. In April 2024, it launched the Alaya NeW intelligent computing operating system and further introduced intelligent computing cloud services in September 2024.

In just two to three years, the company's revenue structure underwent a fundamental transformation.

The prospectus shows that in 2023, DataCanvas's original AI software services (AIFS) generated RMB 81.29 million in revenue, accounting for 76.5% of total revenue.

By 2025, this business segment's revenue had dwindled to RMB 18.58 million, accounting for just 1.7%. In the first half of 2026, it further shrank to approximately RMB 500,000.

Meanwhile, the intelligent computing infrastructure solutions business rapidly emerged as a new revenue pillar.

From 2023 to 2025, revenue from this segment reached RMB 24.94 million, RMB 334 million, and RMB 1.002 billion, respectively. In just two years, revenue grew over 39x, accounting for 91.2% of total revenue.

According to CIC data, based on China's intelligent computing service revenue in 2025, DataCanvas ranks among the top 10 domestic providers and is the only one classified as an 'intelligent computing new cloud' enterprise.

In other words, DataCanvas has transformed from an AI software-focused company to a bona fide AI computing power company in nearly three years.

From a business perspective, DataCanvas's new operations primarily consist of two segments: intelligent computing infrastructure solutions and intelligent computing cloud services.

The first segment is straightforward: helping clients build intelligent computing centers.

For example, if a local government or state-owned enterprise wants to construct a large-scale intelligent computing center, DataCanvas can handle the entire process, from architectural design and server procurement to network deployment, system integration, and equipment commissioning.

What clients ultimately receive is a complete computing power facility capable of running AI training and inference tasks.

DataCanvas's second business segment is intelligent computing cloud services.

By integrating controllable computing power resources, the company provides enterprises with services such as model training, fine-tuning, and inference. Clients can use computing power on demand without building their own intelligent computing centers.

Interestingly, these two businesses are not entirely independent—they even interconnect.

DataCanvas can first build an intelligent computing center for a client and then lease back the computing power from them. This is the biggest difference between DataCanvas and other AI computing power integrators.

As of late June 2026, DataCanvas's controllable computing power reached 26,824 PFLOPS, with 84 intelligent computing cloud enterprise clients and a computing resource utilization rate of 79.6%.

This construction-and-operation hybrid business model has also led to a unique phenomenon at DataCanvas: some major clients are both customers and suppliers.

The prospectus shows that in 2025, two of DataCanvas's top five clients also appeared on its list of top five suppliers.

Specifically, Client Group J purchased approximately RMB 231 million worth of intelligent computing infrastructure solutions from DataCanvas, accounting for 21% of the company's annual revenue. Meanwhile, DataCanvas procured approximately RMB 235 million worth of AIDC leasing services from this enterprise, accounting for 14.4% of total annual procurement.

A similar situation existed with Client Group K. In 2025, DataCanvas sold approximately RMB 121 million worth of intelligent computing infrastructure solutions to this group, accounting for 11% of annual revenue. At the same time, it purchased approximately RMB 515 million worth of servers from them, accounting for 31.5% of total annual procurement.

In other words, these two enterprises alone contributed approximately RMB 352 million in revenue to DataCanvas in 2025, accounting for 32% of total annual revenue. Meanwhile, DataCanvas's total procurement from them reached approximately RMB 750 million.

This unique model forms the basis for understanding DataCanvas's value.

Computing Power Leasing Pressures Behind High Growth

This 'build-and-sell' model has undoubtedly helped DataCanvas secure more orders.

In 2024, DataCanvas delivered five intelligent computing center projects, generating RMB 334 million in related revenue, averaging approximately RMB 66.8 million per project.

In 2025, the number of delivered projects increased to six, but revenue surged to RMB 1.002 billion, with average project revenue reaching RMB 167 million.

The number of projects increased by just one, yet average project revenue grew by approximately 1.5x.

In the first half of 2026, the company delivered three more projects, achieving RMB 434 million in revenue.

As project scales expanded, DataCanvas's profitability also improved. In 2024, the gross margin for its intelligent computing infrastructure solutions business was just 12.6%. By the first half of 2026, this figure had risen to 41.1%.

For comparison, Jiliu Technology, another AI computing power infrastructure enterprise, reported a gross margin of just 16.8% for its AI computing power cluster products in 2025.

According to the company, the gross margin improvement was driven by two main factors: reduced low-margin hardware procurement and an increasing proportion of larger, more complex, and higher-margin projects.

However, behind the impressive revenue and gross margin figures, DataCanvas faces another challenge.

First, the company is highly concentrated in both its supply chain and client base.

From 2023 to 2025, revenue from DataCanvas's top five clients accounted for 54.8%, 85.7%, and 88.0%, respectively, and still reached 80.8% in the first half of 2026.

The supply side is similarly concentrated. In 2025, procurement from DataCanvas's top five suppliers accounted for 84.8% of total annual procurement; this figure remained at 74.5% in the first half of 2026.

This means that while DataCanvas can rapidly scale its revenue through large projects, its reliance on a few upstream and downstream partners has deepened.

Especially since intelligent computing center projects often involve hundreds of millions of yuan, any changes in major clients' procurement plans, project acceptances, or payments could significantly impact the company's revenue and cash flow.

Beyond client and supplier concentration, DataCanvas faces an even more critical issue: cost pressures from long-term computing power leasing.

Under the leasing model, the company does not need to rely entirely on self-built data centers, but the trade-off is that it must bear the agreed-upon costs in leasing contracts even if downstream demand falls short of expectations.

The prospectus shows that at the end of 2023, DataCanvas's lease liabilities stood at RMB 11.7 million. By the end of 2024, they had increased to RMB 187 million, and further to RMB 957 million by the end of 2025.

By late June 2026, this figure had risen to RMB 1.48 billion.

Subsequently, the company signed new long-term computing power arrangements. As of late July 2026, lease liabilities had reached RMB 3.087 billion, more than doubling in just one month.

Corresponding lease liability interest also surged from RMB 360,000 in 2023 to RMB 35.88 million in 2025, and reached RMB 32.46 million in just the first half of 2026.

The issue is that these pre-locked computing power resources have not yet been fully converted into profits. In 2025, DataCanvas's intelligent computing cloud services generated RMB 77.61 million in revenue but incurred a gross loss of RMB 72.19 million, with a gross margin as low as -93.0%.

By the first half of 2026, with growing computing power demand and improved resource utilization, the intelligent computing cloud business finally achieved positive gross profit, though the gross margin remained at just 2.5%.

On the bright side, the company's overall losses are narrowing significantly. In 2024, DataCanvas reported an adjusted net loss of RMB 206 million, which narrowed to RMB 91.3 million in 2025 and further to RMB 14.88 million in the first half of 2026.

Conclusion

Over the past three years, DataCanvas has seized the opportunity presented by the large model boom to rapidly transform from an AI software company into an intelligent computing infrastructure service provider.

Behind this swift transformation lies DataCanvas's aggressive expansion strategy: first rapidly expanding its controllable computing power scale through intelligent computing center construction projects and long-term leasing partnerships, then converting this scale into sustained revenue and profits.

The benefits of this approach are clear. DataCanvas does not need to rely entirely on self-built data centers to rapidly scale its computing power, saving substantial upfront construction time and capital investment.

But the trade-offs are significant. As computing power scale expands, the company has pre-locked leasing costs for the coming years, giving its business model a pronounced operational leverage.

When computing power demand is strong, pre-locked resources can serve more clients. As resource utilization improves, fixed costs are continuously diluted, and profits have the potential to release rapidly.

Conversely, if the computing power supply-demand balance shifts, GPU leasing prices decline, or hardware iterates quickly, long-term leasing contracts could become a heavy cost burden.

Over the past three years, DataCanvas has proven its ability to rapidly scale amid the large model wave. Next, it must demonstrate that it can sustain (chixu - continuously) convert these pre-locked computing power resources into profits and cash flow.