Orders‑Backed Foundation, Capital‑Driven Momentum, Ecosystem‑Built Moat: Xunce Technology’s TokenCloud Weathers the GPU Price‑Hike Cycle Through Multi‑Dimensional Strengths

Recently, Nebius announced a blanket price increase for its GPU cloud services effective October 1, with an average hike of roughly 20% covering multiple chip models including H100 and H200. This marks the second round of price rises from Nebius since May this year. It is understood that some clients have reserved computing capacity as far ahead as 2028, with demand visibility stretching beyond 24 months. Amid a widening supply-demand gap, data centre operators are gaining stronger pricing power, and the industry landscape is undergoing shifts.

Domestically, Xunce Technology (03317.HK, the “Company”) launched TokenCloud, an all-in-one computing platform for AI model training and inference, in early September. The platform aims to deliver end-to-end workflows spanning data ingestion, computing resource orchestration, model inference optimisation, and refinement and deployment of enterprise small models. The Company recently issued a circular, outlining plans to invest up to RMB 12 billion in phased development of AI inference and computing centres, subject to approval by the extraordinary general meeting.

Why is Xunce Technology proactively building computing infrastructure? First and foremost, the move is driven by client demand. Xunce has built up a large roster of high-quality clients and robust cloud orders. TokenCloud’s development is built around its highest-quality order pipeline, delivering near 100% cloud utilisation with no idle capacity awaiting tenants. TokenCloud represents a natural extension of Xunce’s upstream and downstream capabilities, similar to the “Taobao & Tmall + Ali Cloud” model. Beyond demand secured under customer orders, this platform is also backed by solid technical architecture. Computing‑acceleration features are embedded end-to-end – from request ingestion, computing execution through result delivery – to sustain high utilisation of every accelerator card.

Second, high‑quality orders further reinforce TokenCloud’s operational resilience. Unlike standard pricing for generic cloud leasing, Xunce holds contracts with favourable commercial terms featuring higher contract values and longer tenors. The high quality of these orders stems from its FDE model, which is deeply embedded in clients’ operational scenarios. Its services are tightly integrated with clients’ day‑to‑day operations, creating substantial switching costs. Deep integration between the product suite and client businesses fosters long‑term trusted relationships validated by real‑world deployment. As of end‑June 2026, customer retention exceeded 90%. ARPU rose from RMB 1.64 million in the same period last year to RMB 5.56 million, representing a year‑on‑year increase of approximately 240%.

Third, stable order books also translate into meaningful benefits in capital costs. Xunce recently issued a circular proposing to seek a syndicated loan facility of tens of billions of renminbi. This low‑cost financing is underpinned by the Company’s shareholder profile and lenders’ recognition of the quality of its order book.

In addition, Xunce has formed deep strategic partnerships with domestic GPU vendors including MetaX, Biren and Iluvatar CoreX to jointly develop training‑and‑inference chips and platform ecosystems for vertical sectors and enterprise AI use cases. As full‑stack domestic AI development moves into systematic engineering optimisation, platforms that package heterogeneous computing resources and models into services are emerging as a critical enabler for enterprises deploying domestic AI solutions.

Reliable upstream supply and solid downstream orders provide dual safeguards for TokenCloud. Meanwhile, the positive feedback loop between TokenCloud and TokenOS gives the model its self‑reinforcing properties.

TokenOS focuses on refining data capabilities to unlock usable data; TokenCloud focuses on model training and inference plus data‑model integration to address hardware orchestration challenges. The two platforms deliver deep synergy for AI To B scenarios. Higher adoption of TokenCloud drives growth in TokenOS revenue and gross margins. In turn, ongoing improvements to TokenOS bring higher quality clients and orders to TokenCloud. Together, the two platforms reinforce one another, creating a flywheel effect.

Fuelled by the non‑linear surge in inference‑computing demand driven by AI Agents, infrastructure providers with end‑to‑end delivery capabilities will be among the first to enter a period of accelerating earnings growth. By deepening its footprint in downstream business scenarios and engaging upstream with algorithm and computing ecosystems, Xunce Technology stands poised for strategic re‑rating as it evolves from a “digital infrastructure foundation” to an “AI productivity platform”.