China’s state-owned telecommunications giants—China Mobile, China Unicom, and China Telecom—are undergoing a fundamental identity shift. Long relegated to the role of 'dumb pipe' providers for data and voice, these behemoths are now aggressively entering the generative AI market by selling 'Tokens,' the fundamental units of large language model (LLM) consumption. This move follows a growing industry consensus, echoed recently by Nvidia CEO Jensen Huang, that Tokens have become a new form of digital asset and a primary unit of revenue in the intelligence era.
Despite the carriers' infrastructure advantages, the reception from the developer community has been decidedly chilly. Many Chinese developers are voicing frustration over pricing structures that they describe as prohibitively expensive compared to the aggressive price wars led by tech giants like ByteDance, Alibaba, and Baidu. While these internet firms have slashed LLM costs to near-zero to capture market share, the carriers struggle to reconcile their high-overhead infrastructure costs with the brutal reality of the current MaaS (Model-as-a-Service) market.
The tension highlights a deeper strategic rift between the carriers’ ambitions and their technical utility. While China Mobile has attempted to integrate AI into consumer services—such as its '5G New Calling' feature which promises real-time AI translation and avatars—critics argue these offerings feel like 'AI-washing' of legacy telecommunications. For developers, the value proposition of buying Tokens from a carrier remains unclear when specialized AI firms offer superior performance and ecosystem integration at a fraction of the cost.
Ultimately, the 'Big Three' are racing against a ticking clock to monetize their compute power before it becomes commoditized. Their attempt to sell Tokens represents an effort to move up the value chain, but without a significant shift in how they engage with the developer ecosystem, they risk being sidelined once again. The transition from selling gigabytes to selling intelligence requires more than just compute power; it requires a level of agility and developer-centric innovation that has historically eluded state-owned enterprises.