Summary of Key Points
Recently, banks and AI companies have been launching co-branded credit cards that allow users to exchange their consumption points for Tokens, which are the “fuel” required to run large AI models. This collaboration seemed like a win-win situation: banks could attract young, high-income AI users, while AI companies could gain new customers at lower costs. However, as Token prices have plummeted (for example, DeepSeek’s Tokens are now available for just 1 yuan per million inputs), the value of credit card points in exchange for Tokens has become increasingly insignificant. There are three main reasons behind this price drop: AI models’ self-optimization, improvements in hardware efficiency, and fierce competition among companies. As for the future of Token prices, some believe they will continue to decline, similar to how the cost of utilities like water and electricity has decreased over time; others worry that rising upstream costs could lead to inflation. For ordinary workers like Programmer Xiao Gao, the biggest challenge is that while their bosses want them to use AI to improve efficiency, they have to bear the cost of Tokens themselves, effectively meaning they are responsible for providing their own “work materials.”
I. Why Banks and AI Companies Collaborate on Co-Branded Cards?
The collaboration between banks and AI companies on co-branded cards serves both parties’ needs:
- Banks’ Motivation: The credit card market is declining. The number of credit cards has dropped from a peak of 807 million in 2022 to 687 million in 2026, showing a continuous decrease for 14 quarters. Banks need new strategies to attract high-quality customers—AI users, such as programmers and creators, who are young, highly educated, and have stable incomes, making them ideal targets. Using Token benefits as an incentive helps banks target this demographic more effectively.
- AI Companies’ Benefits: AI models need revenue to operate, but many users are reluctant to pay directly. By linking credit card benefits with membership (for example, the Kimi co-branded card offers a free monthly subscription), companies can expand their user base while reducing acquisition costs, which is cheaper than running traditional advertising campaigns.
For instance, Programmer Xiao Gao’s Agricultural Bank of China Kimi card allows him to earn 1 point for every 1 yuan spent. He can exchange 1000 points for Tokens and even get discounts to subscribe to services—this initially solved his problem of having to pay for AI-related expenses at work.
II. Why Are Tokens Getting Cheaper?
The sharp drop in Token prices is not accidental; there are three main factors behind it:
1. AI Models Saving Costs: AI models can optimize their code, reducing operating costs. For example, OpenAI rewrote the underlying code of GPT-5.6 Sol, cutting costs by 20%, and Kimi K3 improved GPU performance by 2.5 times, making its training cost comparable to that of GPT.
2. Improving Hardware: GPU utilization has increased from 30% to over 70%, and failure recovery times have been reduced from hours to minutes, significantly improving inference efficiency. Stanford data shows that the inference cost of GPT-3.5-level models has decreased by 280 times in two years.
3. Fierce Competition: AI companies are competing fiercely to attract users by lowering prices. Foreign companies like Anthropic and OpenAI have already reduced their Token prices, and domestic company DeepSeek has set a very low price (1 yuan per million inputs). In June 2026, Chinese AI models accounted for 63.5% of usage, while in the United States, this figure dropped to 35.5%, causing concern among foreign companies.
III. Co-Branded Cards Becoming Less Valuable
With the decrease in Token prices, the benefits of these cards have become less attractive:
Programmer Xiao Gao found that it’s more cost-effective to exchange his points for physical gifts from the Agricultural Bank rather than Tokens. This is because when Tokens were more expensive, exchanging points for them was a good deal; now that Tokens are cheaper, the value of points has diminished.
For example, if 1000 points used to be exchanged for 10 yuan worth of Tokens (enough to write 1000 lines of code), with Tokens now costing only 5 yuan for the same amount, exchanging points is no longer worthwhile. It’s more practical to use those points to buy something tangible like laundry detergent.
IV. The Future of Tokens: Deflation or Inflation?
There are two opposing views on the future of Token prices:
- Deflation: Some believe that AI will become as ubiquitous as utilities, and Token prices will continue to fall. Eventually, users won’t need to be as cautious with their usage, just like they use mobile data today.
- Inflation: Others worry that rising costs for hardware (computing chips, electricity, storage), along with the need for AI companies to make a profit, could lead to price increases or a tiered pricing system where higher-quality AI services will come at a higher cost.
For Xiao Gao, these debates are somewhat irrelevant. What matters to him is that his boss wants him to use AI to improve efficiency, and he has to pay for the Tokens out of his own pocket, adding another financial burden to his already challenging workload.
V. The New Burden on Workers: Providing “AI Fuel” at Work
In the age of AI, workers face a new challenge: while technology has made tasks easier, they now have to provide their own “fuel” in the form of Tokens. For example, although his boss asks him to use AI for simple tasks, he still has to pay for the Tokens required to do so.
The irony is that lower Token prices might seem like a good thing, but it could lead to more demands on workers, such as writing more code. It’s similar to when gas prices drop, and your boss still expects you to drive long distances—the cost ultimately falls on you.
In summary, the co-branded cards between banks and AI companies, which were once seen as a solution, have become less useful due to rapid changes in the AI industry. Technological advancements bring convenience, but the costs are often passed onto the workers. This analysis explains the situation in plain language, from the logic behind the collaboration to the practical challenges faced by ordinary users, making it easy for non-financial professionals to understand.