China’s domestic AI compute ecosystem took a significant commercial step this week as Hongxin Electronics, a Shenzhen-listed technology company valued at roughly 174 billion yuan, announced a partnership with the Wuxi National High-Tech District to construct a large-scale AI inference facility built around Huawei’s Ascend 384 super-node computing cluster. The initiative, framed as a “Token Factory,” represents one of the clearest articulations yet of how China intends to commercialize domestic AI infrastructure at scale.
The timing is deliberate. China Telecom rolled out its commercial token subscription service nationwide on May 17, offering personal plans starting at 9.9 yuan per month for 10 million tokens and enterprise plans from 39.9 yuan per month for 15 million tokens. A separate procurement announcement from China Telecom’s Ningxia branch, estimated at 16.451 billion yuan, specifically targeted Token Factory generation capability services, according to Pandaily reporting. The convergence of hardware deployment in Wuxi and nationwide token commercialization by a state-owned carrier reflects coordinated infrastructure buildout rather than isolated corporate initiative.
The Ascend Stack Takes Center Stage
The Ascend 384 super-node cluster is central to this deployment. Huawei’s Ascend 910C chips, part of the broader Ascend 910 family, form the compute foundation for large-scale inference and training workloads within the facility. The architecture is designed to handle the exponentially growing token consumption associated with AI agent applications and model inference demand, which has accelerated sharply across China’s digital economy over the past year.
The “Token Factory” concept first appeared at Nvidia’s GTC 2026 conference as a framework for thinking about AI compute industrialization. That China’s domestic ecosystem has adopted and localized this framing around Huawei hardware, rather than Nvidia, illustrates how the domestic AI supply chain has matured under pressure from US export restrictions. The Wuxi facility, backed by significant government commitment from the National High-Tech District, is explicitly designed to serve inference demand generated by expanding AI agent applications using homegrown chips. The Global Times reported on May 18 that the initiative reflects China’s broader strategy of deploying domestically developed compute infrastructure at commercial scale.
Huawei’s AI chip revenue is projected to jump 60% to $12 billion in 2026, underscoring the scale of demand that projects like the Wuxi Token Factory are designed to serve. The Ascend ecosystem’s growing dominance in state-backed compute infrastructure has also been visible through Tencent Cloud’s decision to deploy DeepSeek V4 on Huawei Ascend 950 chips, signaling that China’s largest cloud providers are building their AI stacks around domestic silicon.
Commercial Token Markets Take Shape
China Telecom’s nationwide token subscription rollout marks a structural shift in how AI compute is being monetized in China. The move from enterprise API access to subscription-style consumer and business plans mirrors the maturation of cloud computing markets a decade earlier, when infrastructure investment eventually produced standardized, mass-market service tiers.
Personal plans at 9.9 yuan per month for 10 million tokens are priced to drive mass adoption. Enterprise plans provide a lower per-token cost at higher volumes, following conventional SaaS pricing logic. This suggests the underlying inference infrastructure has reached sufficient scale and efficiency to support mass-market pricing, which is only possible if the compute layer, the Ascend-based Token Factories, can deliver output at competitive cost.
For Huawei, the Token Factory model represents something strategically significant beyond hardware revenue. By embedding Ascend infrastructure into the commercial token economy, the company is positioning its chip ecosystem not merely as a substitute for Nvidia but as the architectural foundation of China’s AI service layer. The involvement of Hongxin Electronics, which operates across consumer electronics and industrial sectors, indicates that the addressable market for Ascend-based compute is broadening beyond hyperscaler cloud providers.
Geopolitical Underpinning
The Wuxi project and China Telecom’s national token rollout both advance the same strategic objective: reducing China’s AI compute dependence on foreign hardware. US export restrictions have effectively closed off the path to Nvidia’s most advanced data center GPUs. Rather than simply creating a gap, those restrictions have accelerated investment in domestic alternatives, with Huawei’s Ascend ecosystem as the primary beneficiary.
The 16.451 billion yuan procurement estimate for Token Factory services from China Telecom Ningxia alone indicates the scale of capital being deployed. Whether the Ascend 384 super-node clusters can deliver performance comparable to the Nvidia systems they are replacing for inference workloads remains a subject of active evaluation among Chinese developers. The commercial ecosystem is being built regardless, with state capital backstopping the transition. The Token Factory in Wuxi is unlikely to be the last.