For years, Chinese IT-services companies were judged mainly by their ability to implement enterprise software, integrate systems, and support large digital-transformation programs. iSoftStone’s latest results suggest that the center of gravity is moving. The Beijing-based company says AI-related business generated 11.435 billion yuan in the first half of 2026, representing 61.46% of its total revenue and growing 46.61% from a year earlier.
The headline figure needs to be read carefully. It is the company’s own definition of AI-related revenue, and it covers much more than a single model or chatbot product. In its August 25 earnings release, iSoftStone describes a strategy spanning AI infrastructure, computing products, industry applications, intelligent terminals, and token services. That breadth is precisely what makes the disclosure noteworthy: it offers a view of how a large Chinese systems integrator is trying to turn the AI boom into a full operating model.
The company reported total first-half revenue of 18.607 billion yuan, up 17.91% year on year. Its computing-products and intelligent-electronics business brought in 9.479 billion yuan, a 40.32% increase and more than half of revenue. Yet the story is not one of uncomplicated growth. National Business Daily reported that iSoftStone’s net loss attributable to shareholders widened to 265 million yuan in the first half, even as second-quarter profitability improved. The combination captures the basic tension of China’s AI infrastructure buildout: revenue can rise quickly while the cost of hardware, services, and capacity investment remains substantial.
From IT Integration to an AI Factory Model
iSoftStone has framed its approach around an “AI Factory,” a term that can sound abstract until it is broken into its commercial parts. The company is attempting to link data-center and computing capacity with AI servers, deployment services, industry agents, and the continuing supply of inference tokens. In other words, it is positioning itself as a provider that can help an enterprise acquire computing resources, connect them to models, deploy them into a workflow, and pay for usage over time.
Its release says the company has lit what it calls the Beijing No. 1 Token Factory and that a public computing and token platform in the Greater Bay Area is now operating. These are company-reported milestones, not independent measures of demand or profitability. Still, they show why the token has become an organizing unit in China’s AI infrastructure market. Rather than selling only a machine or a software license, companies increasingly seek to sell access to model inference measured by usage.
That approach connects with a larger trend in which computing capacity, energy, and service delivery are being bundled into a commercial offering. EastFrontier’s report on Inner Mongolia’s effort to link green computing with AI service exports showed how regional infrastructure projects are being tied to the business of providing AI capacity beyond a single local customer. iSoftStone is operating at the enterprise-services layer of the same transition. It is not merely building a data center; it is trying to make capacity usable by clients that need models, tools, integration, and support.
The company says its business spans financial services, manufacturing, agriculture, energy, transport, and other sectors. Such claims should be understood as a description of its target market and project portfolio, not a guarantee that every industry deployment has reached the same level of maturity. But they reflect an important change in the market. Chinese AI adoption is increasingly about putting models into established business systems rather than asking users to visit a stand-alone chatbot.
Servers, Tokens, and the Cost of Growth
The financial results illustrate why that shift is hard to execute. The computing-products and intelligent-electronics division is becoming a larger driver of revenue, helped by demand for servers, workstations, and AI-capable terminals. But hardware-led growth can carry low margins, inventory exposure, and high working-capital needs. National Business Daily also noted the company’s higher impairment losses and financial expenses, underscoring the costs that can accompany an expansion into equipment and computing infrastructure.
This is a crucial distinction for readers tracking China’s AI industry. A large percentage of “AI revenue” does not automatically mean a company has found a high-margin software business. In iSoftStone’s case, the figure includes infrastructure, computing products, and enterprise delivery alongside software and agents. It demonstrates scale, but it does not by itself reveal which part of the stack will produce the strongest long-term returns.
The company’s model also depends on the reliability of the underlying computing ecosystem. As EastFrontier noted in its recent analysis of AI chip packaging and thermal design, AI infrastructure is constrained by much more than the availability of a processor. Memory, packaging, cooling, networking, power, software compatibility, and service delivery all affect what customers can actually deploy. An enterprise-services company that promises an AI Factory has to coordinate across that chain.
That coordination may become an advantage if customers prefer a single provider that can assemble a complete system. It may also become a burden if iSoftStone has to carry expensive assets while competing against cloud providers, chip vendors, telecom operators, and specialist AI firms.
What the Earnings Report Signals for China’s AI Market
iSoftStone’s disclosure is significant because it shows AI moving from a discrete product category toward the core accounting structure of a large technology-services company. When more than 60% of revenue is classified as AI-related, the question is no longer whether AI matters to the company. The more useful question is how much of its business it can turn into repeatable, profitable services rather than project-by-project delivery.
The company’s first-half loss makes that question more urgent. Growth in AI servers, token capacity, and enterprise deployment may strengthen market position, but it requires investment before returns are certain. The second-quarter improvement reported by National Business Daily is encouraging, yet it does not resolve the economics of maintaining infrastructure and delivering customized projects at scale.
For China’s wider AI market, that pattern is familiar. The country has no shortage of companies that can build models, sell hardware, or demonstrate agents. The harder task is building businesses that make those capabilities dependable inside banks, factories, farms, and public systems. iSoftStone’s results show one way that established service providers are trying to close that gap. Its next reports will show whether an AI Factory can become not just a strategic label, but a durable business.