Token Equity — The Core of the AI Industry Isn't How Smart the Model Is

2G → 3G → 4G: A Replicable Path of Falling Prices

Wushi Caijing (巫师财经, a financial commentary channel on Bilibili) maps the cost evolution of AI tokens onto the roughly fifteen-year trajectory of mobile-data price cuts:

Era	Price Level	What It Unleashed
The 2G era	Data at sky-high prices	The age of pure text — "watching" a live NBA game meant someone typing "Kobe pulls up from three — it's in!" rather than watching video
The 3G era	Sending a photo no longer cost a fortune	The explosion of image-and-text communities — the rise of Renren, Weibo, Tianya, and WeChat official accounts
4G + Wi-Fi	The cost of video fell to the point of indifference	The rise of video platforms — Youku, iQiyi, Tencent Video, and Mango TV, plus Douyin and Kuaishou, with Wushi Caijing itself riding the same dividend
AI tokens (now)	Equivalent to the 2G era	Token costs are still in the "hurts-to-spend" zone — but China is pulling them down

The law this trajectory reveals: what matters for a technology's mass adoption is not how advanced it is in itself, but that it becomes cheap enough to be "wasted."

The Road to Token Equity (The Chinese Model)

China's AI giants, locked in ferocious competition with one another, are pushing token equity forward in three ways:

1. Extreme low pricing / free tiers: DeepSeek charges only a few yuan; the cost of running a Chinese large model is, in one commenter's words, "basically the electricity bill plus a little."
2. Subsidizing users: rather than merely not charging, they hand you money to get you to use their products — because "if you don't use them, it means this giant will be knocked out of the next era."
3. All-out price wars: domestic products such as Doubao (ByteDance) and Qwen (Alibaba) are slashing each other relentlessly, driven by the survival anxiety of the internet giants behind them.

The Result: China Becomes the World's Only Country Where AI Has Truly Gone Mass-Market

The near-zero price of tokens produced a counter-intuitive consequence: AI is no longer confined to programmers and tech workers — it has seeped into every corner of society:

Helping an elderly person recognize benign paroxysmal positional vertigo — no more burning incense and praying to the Buddha.
Spotting the traps in an employment contract for a young worker.
Walking a primary-schooler through the homework one more time.
Keeping company, deep into the night, with someone who has just lost their mother.

These people cannot write a prompt and have no idea what deep learning is. Yet the technology has already reached them — crossing the lines of wealth, geography, and age.

The Cost-Structure Inversion: The Economic Paradox of Free

This is the most fundamental difference between the AI industry and the traditional internet. The response of American AI giants (OpenAI, Google, and the like) has been — mass account bans, withdrawn discounts, tighter weekly quotas, steep price hikes. The underlying logic: "The cost structure is inverted; I would rather not have free users at all."

Even with the cost structure inverted, Chinese AI firms have stuck with free or near-free pricing — which means absorbing enormous hidden costs (R&D + headcount + cash-burn promotions + servers).

Global Token-Call Data (2026)

Rank	Model	Country
1	Qwen 3.6 Plus	China
2	MiMo-V2-Pro	China
3	DeepSeek V3.2	China
7	Claude Sonnet	United States
8	Gemini 2.5 Flash	United States
10	ChatGPT-4o	United States

On OpenRouter, an API aggregation platform, the share of calls going to Chinese models has overtaken that of American models — reaching 2 to 4 times the US level at its peak.

The Battle for the Right to Diffuse — A Score of 95 for 20 Yuan Beating a Score of 100 for 100

The industrial logic of token equity carries a second layer of meaning when viewed from the angle of US–China competition. In his reading of Liang Wenfeng's internal remarks, the commentator Yang Sheng Alex (杨升Alex) points out: what the United States is fighting for is a monopoly over the most advanced models and technical standards; what China is fighting for is the right to diffuse AI.

Open source is the vehicle of the right to diffuse. American firms rely on closed-source models to spin a high-margin flywheel — the more advanced the model, the more dependent the users; the more dependent the users, the more money there is to buy the next batch of chips and extend the lead further. DeepSeek's open-sourcing lets developers around the world use, deploy, and modify Chinese models, directly eroding the pricing power of closed-source American models.

The arithmetic of 95 and 20. If the American model scores 100 on capability at a cost of 100, and the Chinese model scores 95 at a cost of 20 — for the vast majority of enterprises, which is more competitive is not a hard question to answer. Many developing countries do not necessarily need the world's absolute strongest model; what they need is a model that is good enough, far cheaper, deployable at home, and does not require handing sensitive data over to American cloud providers.

Control the commanding heights, or reach the whole world? This is two sides of the same coin as the essay's opening claim that "tokens must be cheap enough that people dare to waste them": token equity is not merely industrial logic — it is also a weapon of competition. If the United States permanently holds the strongest chips and models while China delivers near-equivalent AI at a fraction of the price, who actually wins this contest? The side that controls the technological heights — or the side that truly spreads the technology across the world?