The investor is ready to make a payment with Tokens.

“How much do you spend on Tokens each month?”

This group of entrepreneurs can hardly avoid this question. The reason is simple. In the era when AI has evolved from conversation to execution, everyone wants to make AI work, but AI runs on Tokens, and Tokens need to be bought with money.

Quietly, Tokens have become the currency of the AI era. Recently, Jensen Huang mentioned that in the future, each engineer at NVIDIA will need an annual Token budget - which has been jokingly said by the outside world that salaries will be paid in Tokens.

Coincidentally, yesterday (March 19th), ZhenFund announced the joint launch of the Token Grant, which will provide a Token fee of 50,000 yuan for each selected project, supporting entrepreneurs to build the next - generation AI - native products from the very beginning.

This has also sparked discussions in the VC circle: Can Tokens be used for some investment payments in the future? After all, Tokens are the most needed ammunition for AI startups at present.

Token costs are too high. Sorry, startups can't afford to burn them

All the bosses are suffering from AI anxiety.

The OpenClaw tsunami has brought a subtle sense of self - doubt: if a company doesn't show a positive attitude towards embracing AI, it will fall into the fear of being eliminated.

Now, fully embracing AI brings even more anxiety - Token costs.

Many AI startups feel this deeply: AI tools such as Claude Code, Cursor, Codex, and even OpenClaw have gradually become standard in the workflow. But there's no free labor in the world. Every word you say to AI, every logical operation it performs, and every line of code it outputs are all priced in Tokens.

“I burned 10,000 dollars' worth of Tokens, but not a single product was launched,” a startup founder posted. Currently, R & D employees treat API calls as product development. They change prompts one day and parameters the next, burning Tokens like crazy.

Even wealthier investors can't help but complain. Chamath Palihapitiya, the head of Social Capital and a former early - stage executive at Facebook, recently publicly said that the costs of the software startup he founded have more than tripled since November 2025. One of the main reasons is that the AI programming tool Cursor consumes a large number of Tokens. According to the current trend, the annual cost on AI will reach 10 million dollars.

“Thanks to VCs for paying the bill for this Token buffet,” Chamath Palihapitiya said, which sounds like self - mockery.

Looking at the situation in China, although the unit price of Tokens is indeed cheaper, the usage is usually calculated in tens of thousands or billions. For startups, this is not a free - of - charge feast they can enjoy without a burden.

Guo Sir, the founder of a one - person company based in Beijing, told the investment community that currently, Token expenditures in his company account for 70% of the costs. Iterating a product once may require 1,000 - 1,500 dollars' worth of Tokens. In his opinion, AI costs must be controlled according to expected returns. Consuming more Tokens should result in a more intelligent product, greater market advantages, and higher returns.

“Otherwise, Token fees may drag down startups.”

We heard a case from an investor: during a roadshow, an AI toy company claimed that a toy priced at 600 yuan consumes about 2 million Tokens per day to prove its popularity. In this case, with a “fair price” of only 5 yuan for 2 million Tokens, how can this AI toy balance Token costs and product revenues after four months?

“Especially for startups with intelligent levels based on the cloud, they will more or less face the common problem of continuous Token consumption. The higher the intelligence level of the product, the more Tokens are used, thus falling into linear cost competition,” an AI investor friend admitted. Currently, he is more concerned about the business models and long - term marginal costs of AI startups.

The currency of the AI era: when Tokens start to replace salaries

In the AI era, business models are no longer the same as before.

For a long time in the past, when entrepreneurs developed a software product, they usually invested a large amount of money in the early stage. As time passed, the costs would gradually be spread out until they almost disappeared. Under this logic, the more products were sold, the more money was made.

Now, with Agents integrated into the workflow, the law of decreasing marginal costs has been overturned - every user interaction consumes Tokens, and every interaction requires a payment. If the traditional subscription - based charging model continues, the most loyal users are exactly the most expensive ones. On the contrary, the cheapest users may also be the ones who are the hardest to retain.

However, AI is a boat that one has to board. Whoever has AI has more productivity. Moreover, everyone hopes to use more intelligent and better models to create more outstanding products, which requires consuming more Tokens.

All these bring a sense of Token economics.

As Jensen Huang, the founder of NVIDIA, put it: “In the new AI world, computing power is equivalent to revenue because without computing power, Tokens cannot be generated, and without Tokens, revenue growth cannot be achieved.” If computing power is compared to a money - printing machine, Tokens are the real currency of the AI era.

Currently, the giants that have sensed the trend are promoting Tokens as infrastructure at the employee level.

It is reported that Alibaba is promoting an internal plan to provide employees with Token quotas to encourage the use of advanced AI models and tools at work. At the same time, Alibaba has established the Alibaba Token Hub business group, led by CEO Wu Yongming himself. The goal is to establish a unified dispatching center centered on creating, delivering, and applying Tokens.

Across the ocean, Jensen Huang also mentioned: “In the future, each engineer in our company will need an annual Token budget. Their basic annual salary may be several hundred thousand dollars. On top of that, I will allocate about half of that amount as a Token quota for them to achieve a 10 - fold efficiency improvement.”

How many Tokens are included in your job offer? It's a magical scene in the AI era.

The Token price - hike wave

When the sea is turbulent, fish become more expensive.

In this March when lobsters are popular, the providers of Tokens have collectively announced price hikes:

· On March 18th, Alibaba Cloud issued an announcement. Due to the global explosion of AI demand, rising supply - chain costs, and a sharp increase in Token call volume, it will raise the prices of some products starting from April 18th, 2026. On the same day, Baidu Smart Cloud announced that it will adjust prices starting from April 18th, 2026. The prices of AI computing - power - related product services will be increased by about 5% - 30%, and the prices of parallel file storage, etc., will be increased by 30%.

· On March 16th, Zhipu launched the base model GLM - 5 - Turbo for intelligent agent tasks such as OpenClaw and simultaneously raised the API price by 20%.

· A little earlier, Tencent Cloud also announced that starting from March 13th, 2026, the Tencent Cloud Intelligent Agent Development Platform will adjust the billing strategy for some models. The input and output prices of the Hunyuan series model Tencent HY2.0 Instruct have increased by more than 400% per thousand Tokens.

The computing - power bill is getting thicker. For startups that rely on Tokens, the pressure is self - evident.

If a company's Token conversion efficiency is not fundamentally different from its competitors, then embracing AI seems to turn into a Token consumption war. In Token economics, the one who survives to the end is not the one who can burn the most money, but the one who can squeeze out irreplaceable value from every dollar's worth of Tokens.

Therefore, humans are still important.

After all, the mission of machines is to exhaust all possible executions within the established rules using computing power. However, it is still a human - exclusive privilege to use sharp business intuition to seize that irreplaceable decision - making moment and take risks for the final failure.

It reminds me of the excellent statement by Naval Ravikant, a top Silicon Valley investor: “In an era of infinite leverage, judgment is the most important skill.” As the AI era advances at full speed, we are experiencing many new experiences that we've never had before.

This article is from the WeChat official account “Investment Community” (ID: pedaily2012), written by Feng Yuchen and reposted by 36Kr with authorization.