AI Token Bundles Emerge: Will They Become as Affordable as Data Plans?

Recently, China's three major telecom operators—China Telecom, China Mobile, and China Unicom—have simultaneously launched "token bundles." For as low as 9.9 yuan, users can obtain 10 million tokens. Many might wonder: I use large language models for chatting and queries daily without paying, so what do these token bundles have to do with me? In reality, casual AI usage for asking questions, casual conversations, or drafting text consumes very few tokens—so few that AI providers are willing to offer them for free. However, for "heavy" AI users, token costs can add up significantly. For example, generating a one-minute video with an AI tool can consume over 1 million tokens, costing around 50 to 60 yuan. Some users are beginning to "raise AI assistants" like Lobster, where more complex tasks lead to exponentially higher token consumption. The new token bundles from telecom operators aim to make AI tasks more affordable. Generating a one-minute video using these bundles averages just about 1 yuan. This development feels reminiscent of the early days of data plans, raising the question: Will tokens become as commonplace and affordable as data once did? In the past, mobile data was a "luxury." Loading a webpage took ages, and streaming videos was a costly endeavor. With the rollout of 4G and 5G networks, data prices plummeted, and today, affordable plans allow unlimited video streaming without financial strain. Will the AI era witness a similar transformation? To answer this, it's essential to understand the similarities and differences between tokens and data. Both serve as "units of measurement." Data measures bytes transmitted over networks, while tokens measure semantic units processed by large language models. Using AI to write an article or task an AI assistant like Lobster consumes tokens, akin to streaming videos consuming data. The key difference lies in their nature. Data is essentially a "transportation fee"—it moves data from point A to point B without regard to content. Tokens, on the other hand, represent "intellectual effort" and are more like "processing fees," underpinned by computing power, models, and large-scale computations. The same 10 million tokens processed by different AI models can vary drastically in cost and quality. Given this understanding, the trend toward "tokenization" similar to data is highly plausible. Currently, besides the three major telecom operators, leading tech companies are also rolling out token-based services. These new bundles and offerings are set to accelerate the adoption of AI applications among consumers. Concurrently, national efforts to build a "computing power grid" suggest that computing resources may eventually be supplied and managed like utilities such as water and electricity, potentially driving token prices down. However, tokens are unlikely to become entirely like data. Data is highly homogeneous, whereas tokens are inherently tied to "intelligence capabilities." In the future, basic tokens may become very affordable, but premium tokens—offering stronger models, lower latency, or advanced features—may still command a premium. Just as tap water is cheap, mineral water and functional beverages have their own price points. The simultaneous launch of token bundles by major telecom operators signals that tokens are entering the everyday consumer's billing landscape. In the future, they will likely become as affordable and convenient as data, while offering additional intelligent benefits.