Summary in One Sentence
This is a very straightforward warning to the current domestic AI industry regarding its inflated growth: while the long-term value of AI-related infrastructure construction is acknowledged, the practice of manipulating token usage volumes to inflate business prosperity is firmly opposed. It sets a simple benchmark for the AI community: the actual productivity created by AI technology must exceed the cost associated with using it; otherwise, the entire industry is merely participating in a bubble with no real value.
---
Detailed and Easy-to-Understand Explanation
1. Let’s clarify what those mysterious “tokens” really are for everyone:
There’s no need to remember academic definitions like “the smallest unit of information processing in large models.” Think of all large models as shared “power banks” that charge based on the amount of information used. For every question you send to a model, it generates a response, and each response consumes a unit of computing power, which is represented by a token. The current pricing for large models ranges from a few cents to several dollars per 1,000 tokens. In other words, the “token bill” is essentially your total cost for using AI services.
A couple of years ago, showing off a high monthly token usage volume (e.g., over 100 billion tokens) was a popular way for AI companies to demonstrate their success, implying a large user base and thriving business. This metric was a key indicator for investors to assess the growth potential of AI companies.
2. Why support infrastructure but oppose false prosperity?
The idea of supporting infrastructure is completely valid—just like the deployment of fiber optics nationwide 20 years ago or 5G infrastructure 10 years ago, building computing centers, developing large models, and providing related services are all beneficial for society and will yield positive returns in the long run. However, the current practice of manipulating token usage to create a false sense of prosperity is detrimental. The real AI demand in the industry could be met by just a few computing centers, but by inflating these numbers, governments and investors end up investing in unnecessary facilities that remain idle, resulting in wasted funds and ultimately burdening taxpayers.
3. “Productivity must exceed token costs” sets a practical standard for AI companies:
Many AI companies talk about “universal artificial intelligence” and “changing the world,” but they rarely consider the basic question: does the money earned from using AI cover the costs? For example, if a small restaurant uses AI to manage customers, calculate sales, and write marketing content, and the AI increases revenue by $1,000 while costing $100 in tokens per month, that’s productivity. In contrast, some companies hire interns to generate meaningless questions and generate a $1 million token bill, claiming 100,000 active users and 500% growth. They use this to raise funds but create no real value, essentially getting something for nothing.
4. This statement exposes a long-hidden issue in the AI community:
It’s been a well-known fact in the AI startup scene that 90% of the reported high token usage volumes are inflated. Some companies use automated scripts to generate large volumes of data, significantly boosting their valuations. This practice benefits those who manipulate numbers but harms those working on practical AI applications, preventing them from getting funding and disrupting the industry’s development.
5. The AI bubble affects everyone:
Many people think that the AI industry’s problems are unrelated to them, but the impact is significant. If everyone believes that manipulating data can lead to easy profits, no one will invest in practical AI solutions. As a result, valuable tools that could simplify tasks or improve healthcare may be delayed for years. Once the bubble bursts, many AI companies will fail, and thousands of engineers will lose their jobs, causing the industry to regress. In the end, ordinary people will suffer the consequences.
In short, this warning highlights the need for real innovation and productivity in the AI industry, rather than relying on misleading metrics to inflate its value.