Are we really ready to spend hundreds of dollars each month on seemingly random strings of text and code?

Article author, source: 0x9999in1, ME News

TL;DR

• The illusion of traffic fades, and computing power returns to common sense: Doubao introduces three paid tiers, signaling the end of the classical strategy of "burning cash for daily active users" in domestic large models. The extremely high marginal inference cost of generative AI is forcing giants to return to the fundamentals of business.
• The "user filtration" behind tiered pricing: 68 yuan as a baseline, 200 yuan to identify frequent users, and 500 yuan to target enthusiasts and professional productivity users. This isn't just a price increase—it's a precise segmentation between "casual users" and "efficiency believers."
• Agentic AI is reshaping payment logic: Chinese users are unwilling to pay for pure tools (SaaS), but are willing to pay for "results" and "labor." When AI evolves from an auxiliary tool into an intelligent agent capable of autonomously executing complex workflows, willingness to pay will undergo a qualitative shift.
• The industry elimination race has accelerated: after leading applications set paid benchmarks, smaller models lacking core use cases and relying solely on superficial wrappers will face dual pressure. China’s large model market has officially entered the brutal phase of competing for business闭环.

The App Store page changed silently.

No flood of PR press releases. No emotionally charged launch event under the spotlight.

DouBao, a ByteDance large model app that has consistently ranked among China's top free apps and is widely regarded by users as the top choice for "free AI perks," has revealed its hand—the paid subscription service announcement.

Standard plan: 68 yuan/month. Enhanced plan: 200 yuan/month. Professional plan: 500 yuan/month.

Three price tiers, clearly defined, with annual subscription discounts provided simultaneously.

Blinding? Maybe. Surprising? Absolutely not.

If you have even a passing understanding of the global hash rate arms race over the past two years, you know this day was inevitable. There’s no such thing as a free lunch—and if there seems to be, someone is bleeding profusely. Now, the giants no longer want to, and can no longer afford to, bleed alone.

The free feast of large models is like a dazzling fireworks display. We stand beneath the tree, gazing up in awe at the miracle of technology. But now that the fireworks are over, the venue owner hands us the bill. This isn’t just a problem for DouBao alone—it’s a soul-searching question that determines the very survival of China’s generative AI industry as a whole:

In an internet environment accustomed to the model of “the cost is borne by the pig,” and after countless failed attempts to SaaS-ify the C-end market, can AI subscriptions really work? Are we truly ready to pay hundreds of dollars each month for seemingly randomly generated text and code?

The Physical Ledger of Hashrate: Why Must We Turn the Blade Inward?

Let me ask a question first: Does ByteDance need money?

The answer is clearly no shortage. This super giant has extremely strong cash flow in short-form video, e-commerce, and global expansion.

Why, at this critical moment, risk losing a massive number of users by opening the paywall for Doubao?

Because the ledger of large models isn't the ledger of classical internet—it simply doesn't balance.

Recall the era of mobile internet. When you build an app, what’s the cost of acquiring one more user? Nearly zero. The marginal cost of server bandwidth is infinitely diluted as the user base grows. This is why WeChat and TikTok can be free—they use massive numbers of free users to build an ecosystem, then monetize through advertising, e-commerce, and game partnerships. In this logic, users themselves are the product.

But in the era of generative AI, this entire approach has completely failed.

Why? Because "inference" is too expensive.

Every time you type a question into the chatbox and click send, expensive GPU clusters in data centers thousands of miles away must run at full capacity. Electricity is consumed, chips are worn down, and liquid cooling systems roar. This is real, physical consumption.

The marginal cost of generative AI does not approach zero; instead, it rises exponentially as the complexity of user queries increases.

Asking an AI to write you a good morning greeting might consume negligible computing power. But what if you ask it to read a 100-page English industry report, extract key data, and generate a mind map? The computing cost explodes instantly.

Data doesn't lie. Sequoia Capital once highlighted in a famous industry observation the "$20 billion AI problem"—the entire industry has spent tens of billions of dollars purchasing NVIDIA's computing hardware, yet the industry's annual recurring revenue (ARR) remains vastly out of sync with this investment. Infrastructure development has far outpaced commercial monetization.

The domestic "hundred-model war" has been even more brutal. What once was an API price war—charging fractions of a cent for millions of tokens, or even announcing free APIs—seemed like a fiery boom, but in reality, everyone was gritting their teeth and cutting into their own flesh, betting on who could outlast the others.

Today, capital's patience is returning to reason, and even a landlord’s surplus grain cannot be wasted forever. DouBao’s shift to paid services is not born of greed, but of reverence—for the fundamental logic of business, and for the physical nature of computing power. Turning the blade inward, we cut through the illusion of “free prosperity,” pricing services that truly demand massive computational resources—this is a return to reason.

The underlying logic of tiered pricing: precisely identify "efficiency believers"

Let’s take a closer look at these three pricing tiers. These aren’t numbers pulled out of thin air by a product manager—they’re an extremely precise scalpel.

What does it need to cut through? The vast yet extremely complex structure of China's internet users.

68 yuan is equivalent to two specialty coffees in a first-tier city, or a quarterly VIP membership to a video streaming platform.

At this price point, Doubao targets ordinary white-collar workers, college students, and content creators who have a light but genuine need for AI but are unwilling to spend heavily.

For them, AI is a “convenient plugin.” Help me polish my year-end summary, translate my boss’s obscure instructions into actionable steps, or draft a social media post that sparks interest. For 68 yuan, you’re buying convenience and dignity.

More importantly, this price establishes a fundamental “anti-free-ride barrier.” It instantly filters out无效流量—those who casually seek AI chat for entertainment and send meaningless commands. By removing this commercially worthless traffic that consumes computing power and bandwidth, server resources are reserved for those with genuine needs—this is the first layer of strategic design.

200 yuan, which amounts to over 2,000 yuan per year. This is not an amount to be spent impulsively.

Those willing to spend 200 yuan are surely "efficiency believers."

For these people, AI is no longer a toy—it’s a core productivity tool. They could be e-commerce sellers who need to generate hundreds of product images daily; they could be content creators who need to quickly process vast amounts of information; or they could be junior developers using it to find bugs and write basic code.

At this level, the user’s logic is brutally pragmatic: I pay you $200 per month—can you save me $2,000 in labor costs or help me generate an additional $2,000 in profit?

If it can, 200 yuan is a bargain. If it can’t, even a penny is robbery. The 200-yuan enhanced version is a bet against these most savvy users: betting that my large model’s capabilities match your hunger for efficiency.

The main event is here. 500 yuan per month, 6,000 yuan per year.

What is this concept? Even in the highly willing-to-pay U.S. market, the base pricing for ChatGPT Plus and Claude Pro remains at $20/month. At 500 RMB/month, it far exceeds the average C-end user’s perception of affordability, approaching the pricing threshold of professional financial data terminals or premium SaaS data software.

Who will buy?

The answer is aimed at professional-level industry researchers and individuals striving to reconstruct their entire workflow using "agents."

Imagine an extreme, high-intensity work scenario: you are a professional researcher deeply immersed in the intersection of Web3, blockchain, and AI. You must track global cryptocurrency market trends daily, decipher highly complex regulatory rulings from the SEC and CFTC, analyze fund flows of Bitcoin ETFs, and extract institutional movements from vast amounts of overseas information.

Even with a personal computer equipped with an AMD Ryzen 8745H processor and 24GB of RAM, running local open-source models will still fall short when tackling complex Agentic AI tasks requiring massive concurrent searches, long-text inference, and cross-lingual parsing. Local computing power has physical limits, and sophisticated business intelligence analysis demands cloud-based models with top-tier parameter scales and uninterrupted data throughput.

By this point, the free version of the AI app would have long since started babbling due to a full context window. But what premium增值服务 does the 500-yuan professional version offer? It delivers epic long-text memory, unimpaired complex logical reasoning (e.g., deep integration with frameworks like OpenClaw and Hermes Agent), unrestricted full-power multimodal parsing, and a dedicated high-speed computing channel.

On this level, spending 500 yuan per month isn’t buying software—you’re hiring a tireless, highly intelligent junior analyst. When you only need to give high-level instructions and the AI automatically gathers data, compares regulations, and delivers a well-formatted, logically sound “market report,” an annual cost of 6,000 yuan is an extremely cost-effective investment for professional institutions and top creators.

Breaking the SaaS Curse: The Truth About Chinese Users Paying for AI

After discussing pricing strategies, we must confront the darkest, most demoralizing industry curse: Chinese users simply don’t have a habit of paying for software.

Looking back over the past decade, the Chinese enterprise services (SaaS) market has struggled. Countless teams returned home with Silicon Valley business plans, attempting to transplant overseas models into China. What was the result? Big tech companies used free strategies to dominate the market, while small and medium-sized enterprises preferred piecing together free tools rather than paying hundreds of dollars per month for software subscriptions. Aside from a few entertainment apps that enforced payment through exclusive copyrights, any productivity software that dared to charge typically faced a sharp drop in retention rates.

If you weren’t willing to pay before, why would you pay for large models now?

The breakthrough lies in the fundamental distinction between "Tool" and "Agent".

What is a tool? Traditional SaaS is a tool.

The property of the tool is that it requires you to invest significant personal effort to operate it. You buy a typesetting software, but you still have to align every pixel yourself; you buy a data analytics dashboard, but you still have to drag and drop charts yourself. In the subconscious of Chinese consumers: I’ve already put in the effort and sweat—why should you charge me for a mere software?

But AI large models are undergoing a qualitative transformation. They are no longer merely tools—they are evolving into “black-box subcontractors.”

Under the concept of "Vibe Coding," powered by Agents, you no longer need to give step-by-step instructions—just state your requirements and set your goals. Ten seconds later, the product is ready.

In this process, you did not perform the labor; the AI completed both the mental and physical output on your behalf.

You're no longer buying a hammer; you're paying for labor.

Chinese users are not fond of paying for tools alone, but they are extremely willing to pay for “results” and are highly enthusiastic about paying for “cost reduction and efficiency improvement.” As long as the ROI makes sense and the solution can genuinely replace outsourcing costs, the willingness of Chinese professional users to pay will undergo a dramatic shift. DouBao’s three-tier pricing model is not an attempt to break Chinese users’ consumption habits; quite the opposite, it aligns with a deeper, results-driven business intuition.

Endgame Scenario: The Knockout Stage and the Rise of Super Individuals

Step back and look at the bigger picture. DouBao's move is not isolated—it's a microcosm of China's AI battlefield shifting from trench warfare to brutal urban combat.

Look across the ocean. Despite still burning cash, OpenAI’s substantial cash flow, driven by its subscription model and API revenue, proves that the large model subscription model is not only viable but an inevitable path. Domestically, Kimi has already experimented with pay-per-use and peak-time tipping models; Baidu’s Wenxin Yiyan has also launched a professional subscription tier.

When DouBao, regarded as the "king of traffic," enters the arena with transparent pricing, the rules of the game change completely. This is not just a business strategy adjustment, but an accelerator for industry reshuffling.

1. The "funding drought" crisis for small and medium-sized models: Small and medium-sized AI startups that lack the financial resources to keep burning cash and don’t have sufficient technical depth to convince users to pay willingly will face a cliff-edge survival crisis. Users have been conditioned by giants to pay for services, but this has also made them extremely picky. Dare you charge? Does your model understand multi-step logical reasoning? Can it handle long texts without forgetting? If not, users will immediately walk away.
2. The boundary between B-end and C-end is blurring: the emergence of a $500 monthly fee has already blurred the line between individuals and enterprises. Numerous small studios and independent researchers will no longer purchase bulky and expensive enterprise-grade on-premise solutions, but instead directly buy several high-end C-end accounts to run their entire core business flow via Agent.
3. Amidst the ebb and flow, return to value: the future competition will no longer be about whose model can write the most beautiful acrostic poem, but who can better integrate into the incredibly complex, fragmented, and hyper-competitive real workflows of Chinese internet users.

In the internet industry, there's a saying: "When you don't know what the product is, you are the product."

Over the past two years of free large model indulgence, we freely served as data annotators for the giants, using our wildly varied questions and countless error reports to help train and align their models.

Now, the model has evolved. It has begun wearing a suit, sitting at a desk, and handing you a clearly priced service contract.

In the short term, there will inevitably be complaints about “greediness,” and users accustomed to free services will walk away. But the great wheel of history won’t stop rolling because of complaints. What truly matters are those who stay—the super individuals who put their own money on the line, using top-tier AI as a lever to unlock greater commercial value.

The old era of frenzied growth fueled by endless capital infusion has ended.

A new, hardcore era grounded in real-world adoption and tangible value delivery has arrived, climbing the steps of 68 yuan, 200 yuan, and 500 yuan.

As the superficial glitter on the water fades, see who is swimming naked and who is truly using computational power to measure the future.

Source:

• [1] Sequoia Capital. (2023). AI's $200B Question.
• [2] OpenAI. (2023-2024). ChatGPT Pricing & Subscription Plans.
• [3] Baidu Intelligent Cloud. (2023–2024).Ernie Bot Professional Edition Pricing and Service Details.
• [4] Moonshot AI. (2024).Kimi AI Assistant: Tips and Peak Computing Power Plans.
• [5] U.S. Securities and Exchange Commission (SEC) / Commodity Futures Trading Commission (CFTC). (2024-2026). Public Regulatory Data & Market Reports.