The front lines of China’s artificial intelligence race have shifted from a 'parameter arms race' to a high-stakes battle for deployment. At the center of this transition are 'Token Factories'—startups that treat AI inference as a standardized utility, much like a water treatment plant delivering data on demand. Recent funding rounds, including a staggering 2-billion-yuan Series B for SiliconFlow and a 700-million-yuan injection for Infinigence AI, highlight a venture capital frenzy surrounding the Model-as-a-Service (MaaS) sector.
These firms are witnessing explosive operational growth. SiliconFlow reports a ten-fold increase in revenue over the past year, while Infinigence AI has seen its daily token calls surge twenty-fold since the end of 2025. National data suggests that China’s total daily token consumption has leapt from 100 billion at the start of 2024 to over 140 trillion by early 2026. This exponential demand is fueled by the rise of AI agents, multimodal applications, and the deepening digital transformation across Chinese industries.
However, this massive growth is colliding with a margin-crushing reality. China’s internet giants have entered the fray with predatory pricing strategies. Alibaba recently established its own 'Token Foundry' division under CEO Eddie Wu, while Tencent and Xiaomi have slashed API prices by up to 99% to match DeepSeek’s ultra-low-cost benchmarks. For independent startups, this means that even as their volume scales, their profit potential is being hollowed out by a commoditization of intelligence.
Further complicating the outlook is the rapid rise of edge computing. Industry experts predict that up to 50% of AI inference tasks will soon migrate from the cloud to local hardware, such as AI-powered PCs and smartphones. This 'edge diversion' threatens to cap the growth ceiling for cloud-based token providers. Simultaneously, large enterprise clients in sensitive sectors like finance and energy are increasingly building private, on-premise clusters, further eroding the market share of third-party token factories.
Industry analysts warn that while the token economy is a fundamental pillar of the AI age, the current market is exhibiting signs of a bubble. Many startups are chasing scale at the expense of sustainability, relying on venture capital to subsidize compute costs. The eventual survival of these independent players will likely depend on their ability to offer specialized, high-efficiency optimizations that the generic infrastructure of Big Tech cannot match.