New compute player amid the price hike wave: Why can XUNCE (03317) TokenCloud "make money on every unit built"?

Is the wave of GPU cloud service price increases rewriting the negotiating dynamics of the global computing power industry? Recently, Nebius announced a full-line price hike of approximately 20% for its GPU cloud services starting October 1 this is already its second price increase this year, with the cumulative increase for B300 reaching as high as 56%. Customer demand visibility exceeds 24 months, and some are even willing to bid at a premium for Blackwell computing power; CoreWeave's order backlog has surpassed $100 billion, with contracted power scale continuing to climb. Jensen Huang put it bluntly: "The core constraint is not demand, but production capacity." As "having cards means having pricing power" becomes consensus, the capital markets are beginning to re-examine a question: Who is the truly "certain beneficiary" in this computing power bull market? In China, XUNCE Technology (03317), known as the "first Token stock," has given an answer different from overseas Neoclouds. Contrary to the traditional computing power leasing path of "build the data center first, then find customers," XUNCE first secures high-quality customers and high-quality orders, then builds computing power according to those orders. This "orders first, construction follows" approach has placed the TokenCloud one-stop AI model training and inference computing power platform on a distinctive starting line from its very inception. Customer demand-driven, with high-quality orders in hand What does the computing power leasing industry fear most? Not being unable to get cards, but building data centers without customers massive depreciation from idle assets waiting to be rented is the number one killer for all asset-heavy computing power platforms. Overseas Neoclouds can still hedge risks through long-term contracts with major clients, but the large number of computing power players who rushed in following the trend are facing the hidden concern of "idle upon completion." XUNCE's path is exactly the opposite. Before launching TokenCloud, XUNCE's TokenOS operating system had already completed real-time tokenization transformation of enterprises' multi-source heterogeneous data, achieving large-scale deployment across eleven high-value industries including finance, telecommunications, power, energy, high-end manufacturing, and biomedicine. This means that before venturing into computing power, XUNCE already held a large base of real, paying enterprise customers whose AI usage intensity continues to rise. Drawing an analogy to the classic "Taobao/Tmall + Alibaba Cloud" playbook, the e-commerce platform (TokenOS) has accumulated massive high-quality merchants and real transaction demand, while cloud computing (TokenCloud) serves as infrastructure to absorb these demands, with virtually no need to acquire customers from scratch. Every unit of computing power built by TokenCloud corresponds to real orders that have already been signed and paid for. In industry parlance: every unit of construction is profitable, with no idle assets waiting to be rented. More critically, it's about the quality of the orders. At a time when AI computing power scarcity is moving "from narrative to financial statements," pricing power on the supply side has reversed, and long-cycle, high-unit-price premium terms have become scarce resources. XUNCE's customers are enterprise-level To B clients payment terms, budgets, and renewal willingness in high-barrier industries like finance, telecommunications, and energy are far better than the "retail investors" in the spot rental market. The company's gross margin holding steady at a high of 60.1% in the first half, rapid ARR growth, and ARPU jumping from 1.64 million yuan to 5.56 million yuan are all direct footnotes to order quality. As the on-demand prices of Nebius and others become the industry pricing anchor and continue to rise, the value of these "high-cost, long-cycle, favorable-terms" Cloud orders in XUNCE's hands will only be further amplified. Cloud utilization rate approaching 100%, strong capital cost advantage Computing power is a business of "utilization rate + capital cost." The same H100-class computing power card with 90% utilization versus 40% utilization represents vastly different business models; the same loan for purchasing cards, the interest rate level directly determines the break-even point. XUNCE has built moats in both dimensions that are difficult for peers to replicate. First, Cloud utilization rate. The construction logic of traditional computing power leasers is "betting" betting on future demand, betting on technology roadmaps, betting on customer composition. The clusters they build either sit idle waiting or are discounted for promotion. XUNCE, however, plans construction in reverse based on the highest-quality orders, precisely matching computing power scale with real demand. Combined with TokenCloud's own full-chain acceleration capabilities through model distillation and inference optimization, accelerating cards from "waiting to compute" to "continuously computing" its Cloud utilization rate approaches 100%. This means that for assets of the same scale, XUNCE produces far more Tokens and revenue than peers, with unit depreciation costs diluted to the extreme. In 2026, when computing power is in short supply and price increases are the norm, "fully loaded computing power assets" are themselves a money-printing machine. Next, capital cost. The asset-heavy expansion of computing power relies heavily on leverage, and subtle differences in financing rates amplify into enormous divergences in profitability over multi-year cycles. XUNCE's advantage comes from two ends: First, high-quality stable orders are themselves the best credit endorsement predictable cash flows make banks willing to offer extremely low-cost loans, mirroring the logic of overseas Nebius using "deployed GPUs + contract cash flows" as collateral to obtain low-cost financing; Second, the backing and support of shareholder background further lowers the comprehensive cost of capital. Low-interest funds purchasing computing power assets, high utilization generating high returns, high-quality orders feeding back into credit XUNCE's capital chain forms a positive cycle of continuously declining costs. Ecosystem positioning bound to domestic GPUs, and a self-reinforcing positive cycle If orders and capital are the "technique," then supply relationships and the ecosystem flywheel are XUNCE's "way." The essence of the global computing power shortage is a supply bottleneck. Jensen Huang predicts chip sales will double by 2027 but still emphasizes that production capacity cannot keep up; Intel CEO Lip-Bu Tan admits that CPU supply can only meet about 50% of demand. Against the backdrop of restricted imports of high-end GPUs, domestic computing power has become a certain direction, and "whoever can secure stable, priority access to domestic GPU supply" holds the ticket to this game. XUNCE's positioning is remarkably forward-looking: Since June this year, the company has successively established deep strategic partnerships with nearly all leading domestic GPU manufacturers including MetaX, ILUVATAR COREX, and Biren, and has further moved toward joint research and development jointly building training and inference chips and platform ecosystems for vertical industries and enterprise AI. This is not a simple procurement relationship, but a deep binding of "computing power + data": TokenCloud can uniformly manage heterogeneous computing power including GPU, NPU, and FPGA across architectures, achieving intelligent scheduling from hundreds to tens of thousands of cards, paving the way for large-scale deployment of domestic chips in highly sensitive industries such as finance, telecommunications, and energy. For domestic GPU manufacturers, XUNCE is both a stable major customer and the best partner for validating chips' real-world value and refining software ecosystems the stability and priority of supply relationships thereby gain institutional guarantees. And all these advantages ultimately converge into a self-reinforcing positive flywheel: After TokenCloud's enterprise-level customers run on its computing power, they naturally generate data governance and tokenization needs, driving TokenOS's revenue and gross margin improvement; the data capabilities and scenario understanding that TokenOS cultivates across more industries in turn help TokenCloud attract higher-quality customers and sign higher-quality orders, supporting higher-utilization computing power construction. TokenCloud and TokenOS are not two isolated products, but two deeply synergistic sides of the same coin on the AI To B track every positive interaction thickens barriers that competitors find difficult to surmount. The reversal of pricing power in computing power supply is no longer a narrative, but a reality written into Nebius's price increase letters, CoreWeave's hundred-billion-dollar order book, and XUNCE's financial reports. But standing in the same tailwind, players' quality varies enormously: some build cards on concepts and gamble on cycles, while others hold high-quality orders and build to order, with every cent of capital expenditure corresponding to certain returns. What makes XUNCE TokenCloud unique is precisely that it is not a "from scratch" gamble, but a computing power landscape that naturally grew from the high-quality customers and demand accumulated by TokenOS orders first, full-load operation, low-interest capital, stable supply, and a self-rotating flywheel. When the tide goes out, those who remain at the table are always the players where "every unit of construction is profitable."