In 2023, AI software was DataCanvas’s business. By the first half of 2026, it was almost absent from its sales.
The change is unusually stark. AI foundational software services generated 76.5% of the Chinese company’s revenue in 2023. Three years later, their contribution had fallen to 0.1%.
This was more than an old division being overshadowed by a faster-growing one. Software-service revenue itself contracted—from RMB81.3 million in 2023 to less than RMB500,000 in the first six months of 2026. Meanwhile, total annual revenue climbed from RMB106.2 million in 2023 to RMB1.10 billion in 2025, a compound annual growth rate of 221.5%.
DataCanvas had built a different business.
The Beijing company, known in China as 九章云极, began redirecting its resources toward computing infrastructure in early 2023. It subsequently added cloud services, turning from a supplier of software tools into a company that helps assemble computing infrastructure and sells access to its capacity.
On September 29, 2026, it filed for a Hong Kong listing. The application opens a window onto that reinvention: a company founded more than a decade ago, approaching the public market with its original revenue engine effectively switched off.
Before the Servers
DataCanvas was established in 2013, long before generative AI made computing infrastructure a mainstream business story.
Its early proposition was to help organizations work with data and build machine-learning models. A visual modelling platform followed in 2014, with subsequent products extending across model development and deployment.
These tools addressed an awkward gap inside enterprises. Having data did not mean a company could readily turn it into a working model. Building a model did not mean it could be reliably incorporated into everyday operations.
DataCanvas offered software to bridge those steps. Its products helped customers prepare data, construct analytical workflows and put models to work.
But building the technology was not the same as finding a business.
In 2022, founder and chairman Fang Lei recalled that the team had spent its early years concentrating on code while underestimating the market. The founders were experienced engineers; understanding what customers would actually pay for took longer.
Banks became an important entry point. They had substantial data, established technology budgets and reasons to improve how information informed decisions. Working with such customers gave DataCanvas a commercial setting for its tools—and a clearer understanding of what it meant to sell them.
The company offered enterprises the means to develop and use machine learning. Customers still had to provide the computing environments in which those tools would operate.
That boundary would become the next business opportunity.
The Decision to Change the Business
By 2023, large AI models were making the machinery behind software much harder to ignore.
Customers needed substantial computing resources, coordinated across processors, networks and storage. They had to obtain that capacity, make the components work together and keep workloads running.
For DataCanvas, this opened a choice: continue selling software into those environments, or take responsibility for more of the infrastructure itself.
It chose the latter.
The company began providing computing infrastructure solutions in 2023 and signed its first project contract that September. These were larger assignments than delivering a standalone software platform. They involved designing the computing environment, procuring and integrating hardware, and deploying systems to support customers’ workloads.
DataCanvas redirected resources toward the new operation while progressively withdrawing from its original software-service business.
The commercial consequences appeared quickly. Infrastructure solutions generated RMB24.9 million in 2023. By 2025, they brought in just over RMB1 billion and accounted for 91.2% of total revenue.
The old business moved in the opposite direction. Its revenue fell in both 2024 and 2025.
This was the decisive break in DataCanvas’s history. Selling tools had given it a role inside customers’ technology operations. Delivering infrastructure made it responsible for more of the environment those operations depended on.
The company had moved closer to the machinery—and taken on a larger share of the work.
Build It, Then Keep It Running
The next step came when DataCanvas launched its cloud service in September 2024.
An infrastructure project helps a customer establish computing capacity. A cloud service lets customers access capacity without arranging an entire installation themselves.
For DataCanvas, that meant moving beyond project delivery toward an ongoing operating relationship.
The first cloud-service revenue was tiny: RMB196,000 in 2024. It rose to RMB77.6 million in 2025 and reached RMB153.8 million in the first half of 2026. By then, cloud services supplied 26.1% of revenue, alongside the 73.8% generated by infrastructure solutions.
Its cloud-service enterprise customer count increased from two in 2024 to 75 in 2025 and 84 in the first half of 2026.
That progression gives the reinvention two distinct stages. Infrastructure projects first established the scale of the new business. Cloud services then began changing how the company maintained and monetized its relationship with customers.
A completed installation could produce project revenue. Providing access to computing capacity created a reason for customers to keep buying.
The software company had become an infrastructure provider. It was now also becoming an operator.
What Survived the Reinvention
DataCanvas did not stop writing software when software services stopped driving its sales.
It introduced Alaya NeW OS in April 2024, placing a software control layer at the centre of its newer business. The system coordinates computing resources and the workloads using them. Its subsequent cloud platforms build on that foundation.
This is the revealing tension in the transformation. DataCanvas was withdrawing a way of selling software while retaining the engineering that made its new services work.
Previously, the platform was something a customer bought. Increasingly, it became the mechanism through which DataCanvas delivered another product: usable computing capacity.
The company also developed its own computing measurement unit, DCU, which it markets in China as “one degree of computing power,” echoing the everyday expression for a unit of electricity.
The idea is to make computing easier to measure and purchase across different resources, with charges linked to consumption. It expresses the company’s new ambition in practical terms: customers should be able to buy computing as a service without assembling every ingredient needed to produce it.
For DataCanvas, this offered a different answer to a question it had encountered early in its life: what should an enterprise buy from the company?
Its first answer was software that helped customers develop and deploy models. Its newer answer included the resources on which those models run.
The Price of Starting Again
Replacing a business can produce a spectacular growth chart. It also changes what growth costs.
DataCanvas reported an overall gross margin of 40.6% in 2023, falling to 11.7% in 2024 as infrastructure solutions became the dominant source of sales. The margin recovered to 17.4% in 2025 and 31% in the first half of 2026.
The initial decline shows the distance between the two businesses. An infrastructure contract can contain substantial hardware procurement and deployment work. A much larger order can therefore bring a much larger cost of delivery.
Operating computing services adds another set of demands. Capacity must be available when customers need it, supported by systems that allocate resources and keep workloads functioning. Winning an order is only part of the job.
DataCanvas combines arrangements with infrastructure owners and partners with investment of its own. Moving into computing delivery nevertheless brings equipment costs, capacity commitments and the continuing task of turning available resources into paying usage.
Profitability remains unfinished. Its adjusted net loss narrowed from RMB194.3 million in 2023 to RMB91.3 million in 2025, then to RMB14.9 million in the first half of 2026.
Its reported first-half loss was much larger, at RMB1.09 billion, principally because of RMB1.07 billion in fair-value changes in financial liabilities. That accounting charge explains most of the gap between the reported and adjusted figures.
DataCanvas has demonstrated that it can replace its old revenue. It is still establishing how much it can earn from the replacement.
Taking the New Company Public
The Hong Kong IPO arrives three years into the transformation.
DataCanvas plans to use the proceeds to strengthen research and development, expand computing resources for its data-centre projects, develop sales channels and support working capital. Hardware procurement forms part of the proposed capacity expansion.
Those plans extend the decision made in 2023. The company is seeking capital to deepen a business that has already reshaped its operations and revenue.
For prospective shareholders, the task is to understand what that business has become. DataCanvas carries the history and technical capabilities of a software company. Most of its revenue now comes from infrastructure delivery, while cloud operations account for a growing share.
Its founding identity no longer describes how it makes most of its money.
What makes DataCanvas distinctive is the extent of the choice it has already made. It spent a decade building tools for enterprises to use AI, then deliberately stepped away from selling those tools as its main business. In their place, it built the infrastructure and services needed to keep those systems running.
The revenue transformation is largely complete. The Hong Kong listing brings the next question into public view: whether the business DataCanvas chose to become can turn its reinvention into durable earnings.
