360's Zhou Hongyi: Tokens can never be like mobile data with "unlimited usage," AI will only get more expensive the more you use it.

Qihoo 360 cofounder Zhou Hongyi said at the 2026 Zhongguancun Forum annual conference in 2026 that Token can never be used on a monthly unlimited basis like mobile data packages. He pointed out that the essence of AI is “compute power consumption.” The more complex the task, the higher the consumption, and the business model will fully shift to pay-as-you-go Token economics.
(Background: OpenClaw founder gives interview to Bloomberg: the U.S. can learn from China’s adoption of AI for small crayfish)
(Additional background: OpenClaw will set up an independent operation for the “Crayfish Foundation”! Nvidia, ByteDance confirm joining)

After Crayfish (OpenClaw) became a breakout hit in China, many people began to expect local tech companies to offer “unlimited” Tokens, just like mobile data, turning into a monthly subscription model with unlimited usage.

At the 2026 Zhongguancun Forum annual conference held in Beijing from March 25 to 29, Qihoo 360 founder Zhou Hongyi directly gave the answer: “impossible.”

His core argument is that the foundation of traditional internet services is “traffic.” Fiber-optic capacity is nearly unlimited; the more users there are, the lower the marginal costs—only then does the monthly all-you-can-eat business model make sense. But AI runs on “compute power” and “intelligence.” The more complex the task, the higher the consumption; the logic is fundamentally reversed.

Tokens are not traffic

Zhou Hongyi said that Tokens are the unit for measuring the compute power and intelligence consumption of AI. Their per-unit price is relatively fixed, and the more you use, the more you pay. This is completely different from the mobile-data model where marginal costs decline.

The real numbers have already validated this claim. He said that today, a heavy OpenClaw user consumes about 100 million Tokens per month, which costs around 7,000 yuan RMB. Cheetah Mobile CEO Fu Sheng revealed that the monthly fee for a fully equipped OpenClaw package is close to 30,000 yuan.

Even if you assign no tasks, OpenClaw’s built-in “heartbeat mechanism” alone consumes about $20 per day.

In Shenzhen, even an engineer’s API key was stolen, and after being used fraudulently for three days, $12,000 was drained—making it a form of “Token anxiety” for some people.

Zhou Hongyi specifically called out OpenClaw’s strategy. He said that this “software is free, Tokens are paid” model, in essence, is about cultivating users’ willingness to pay. As for Alibaba establishing a Token Hub business group for this purpose, Zhou Hongyi’s view is that Alibaba has identified the path of Token economics.

Another judgment driven by OpenClaw’s heat is the surge in demand for inference compute. Zhou Hongyi predicted that as OpenClaw becomes widespread, demand for inference compute will skyrocket by a hundredfold or even a thousandfold. He also specifically distinguishes between “training compute” and “inference compute”: training is the ticket for big companies to enter, but in the future the real breakout point will be on the inference side.