On September 28, Xunce Technology (3317.HK) held its first extraordinary general meeting of 2026, during which all three special resolutions were unanimously passed with approval rates exceeding 99%: investing no more than 12 billion yuan to build an AI inference and computing center based on computing power and data, applying for a syndicated loan of up to 10 billion yuan with corresponding guarantees, and amending the company's articles of association to expand its business scope.

All three resolutions point in the same direction: computing power. The smooth passage at this shareholders' meeting resembles more of a collective choice by institutions: institutional shareholders and major long-term holders who have long been heavily invested in Xunce Technology cast their votes of confidence through voting rights for a computing power layout built on high-quality orders with 100% resource utilization, paving the way for the company to deepen its core business and advance its full-chain closed loop of AI to B.

High-Quality Orders First, Asset Quality Recognized

Xunce Technology has already accumulated a large number of high-quality clients across eleven high-value industries including finance, telecommunications, power, energy, high-end manufacturing, and biomedicine. As these clients move from "data tokenization" to "AI inference implementation," demand for privatized, compliance-oriented computing power naturally emerges — TokenCloud was born precisely for these real orders.

Completely different from the traditional computing power leasing path, the first principle of Xunce Technology's computing power business is: orders first, construction follows. This model allows TokenCloud servers to seamlessly connect and operate once deployed, with utilization rates approaching 100%, essentially eliminating idle computing resources waiting to be rented out.

This layout logic is equally recognized by institutions. More importantly, the structural characteristics of the orders stand out. Xunce's clients are enterprise-level To B customers — finance, telecommunications, energy, and other high-barrier industries. Such orders naturally possess the characteristics of "high cost, long cycle, high replacement cost": once customers deeply embed AI into their core business processes, replacement costs become extremely high, forming a strong lock-in effect. This stands in sharp contrast to the short-term flexible pricing and high customer churn rates commonly seen in the scattered rental market.

At a time when GPU cloud service price hikes are spreading, the value of high-quality long-term orders will only be further amplified. Overseas player Nebius has announced two consecutive rounds of price increases this year, with B300 cumulative increases reaching 56%; domestic AI computing power demand growth is approximately 3.3 times the supply growth rate. When "having cards means having pricing power" becomes an industry consensus, the batch of TokenCloud orders in Xunce's hands — characterized by "high cost, long duration, favorable terms" — constitutes a more solid and lucrative revenue foundation than ordinary computing power leasing.

High-quality, stable, long-cycle orders are themselves the best credit endorsement. Very few AI to B companies in the industry can secure a 5-year, ten-billion-level, low-interest syndicated credit facility, which directly reflects financial institutions' high recognition of Xunce Technology's underlying asset quality and business growth prospects. Low-interest funds purchase computing power assets, high utilization generates high returns, and high-quality orders further feed back into credit — Xunce's capital chain has formed a positive cycle of continuously declining costs.

Deep Binding with Domestic GPU Leaders, Stable Supply Chain

The hidden barrier of the computing power business lies not only in front-end orders but also in the back-end supply chain. The essence of the global computing power shortage is a supply bottleneck. Against the backdrop of restricted imports of high-end GPUs, domestic computing power has become a确定性 direction. And "whoever can secure stable, priority GPU supply" holds the entry ticket to this game.

Xunce Technology's positioning is remarkably forward-looking. Since June this year, the company has formally signed strategic cooperation agreements with three leading domestic GPU manufacturers — Muxi, Iluvatar CoreX, and Biren Technology — a landmark event marking the first time a domestic data infrastructure enterprise has systematically and comprehensively partnered with the core supply chain of domestic computing power. Each partnership has its own focus: with Muxi, the focus is on full-stack software compatibility in intelligent manufacturing and technology finance; with Iluvatar CoreX, the focus is on "heterogeneous computing power networks" and "embodied intelligence"; with Biren Technology, the collaboration involves joint research and development of specialized all-in-one machines for urban management and intelligent computing clusters.

This "three-chip advance" strategic layout enables Xunce Technology to form deep strategic cooperation with leading domestic GPU manufacturers, jointly developing training and inference chips and platform ecosystems for vertical industries and enterprise AI. In a new cycle where bargaining power on the computing power supply side has reversed, a stable supply source is itself a moat.

TokenOS and TokenCloud Software-Hardware Synergy, Full-Chain Closed Loop Builds Barriers

Xunce Technology's full-chain product matrix has clear division of labor: TokenOS focuses on "data capabilities," converting enterprise multi-source heterogeneous data into standardized scenario Tokens in real time, solving the problem of "data usability"; TokenCloud focuses on "model training, inference, and fusion capabilities," integrating computing power scheduling, model inference optimization, enterprise small model refinement and tuning orchestration, providing hardware service support for Token generation; AIDP (underlying data resource processing service system) and TokenRouters (top-level model achievement service and scenario connection exchange center) work together to connect the full chain of "computing power — data — Token — model — application."

This closed loop is driving a positive cycle flywheel: TokenCloud undertakes TokenOS's privatization inference needs, significantly enhancing TokenOS's customer stickiness and bargaining power, directly driving TokenOS revenue and gross margin improvement; the data capabilities and scenario understanding that TokenOS hones across more industries in turn help TokenCloud attract higher-quality customers and sign better orders.

TokenCloud's significance to Xunce is akin to what Alibaba Cloud is to Taobao: without it, customers would flow to others' inference clouds, and the foundation of TokenOS's stickiness would loosen. This is defense. But TokenCloud is more than defense. When TokenOS's software capabilities and TokenCloud's training and inference computing power are delivered as a packaged "integrated hardware-software" solution, customers are not just purchasing a set of tools but a closed loop — migration costs are higher, replacement willingness is lower, and Xunce's bargaining power rises accordingly. This closed-loop ecosystem is the greatest barrier that other pure computing power players find difficult to replicate.

Conclusion: When the Tide Goes Out, Those Who Remain at the Table Are Players Where "Every Unit Is Profitable"

Xunce Technology's 12 billion investment in computing power appears on the surface to be a heavy-asset expansion, but in essence it is an inevitable closed loop extending from the TokenOS business. In the second half of the AI industry, where the focus shifts from "competing on models" to "competing on implementation," computing power bids farewell to the era of "stacking cards" and enters a new phase of "competing on efficiency." Xunce Technology's path precisely avoids the three biggest pitfalls of pure computing power companies — uncertain orders, low utilization rates, and high capital costs. It drives computing power investment through customer stickiness, exchanges long-cycle orders for low-cost financing, builds differentiated barriers through industry know-how, ensures stable supply through domestic GPU ecosystem binding, and forms a self-reinforcing closed loop through the TokenOS + TokenCloud flywheel, opening up a new round of high-growth space.