The May Day holiday had not even ended when a seismic signal shook the technology world. Doubao — ByteDance’s nationally dominant AI application — quietly launched three paid subscription tiers on the App Store: a Standard tier at RMB 68 per month (≈ USD 9.44), an Enhanced tier at RMB 200 per month (≈ USD 27.78), and a Professional tier at RMB 500 per month (≈ USD 69.44), with an annual fee reaching as high as RMB 5,088 (≈ USD 706.67). The moment the news broke, #DoubaoGoesPayd instantly shot to the top of Weibo’s trending topics, and the comment sections exploded into argument. Some users declared outright “I’ll uninstall it if they charge,” others lamented “the good days of free AI are over,” while others understood the weight of the cost pressures behind the high price tags. Official response came swiftly: the basic free version will be retained permanently, payment applies only to high-computing-power productivity scenarios, and the plans are still in testing. But everyone understands the reality: for the domestic AI frontrunner sitting on 345 million monthly active users and more than 150 million daily active users, this shift to paid service is no temporary experiment. It is the final straw that collapses the free AI model — and a landmark inflection point at which the entire large model industry pivots from “burning money for traffic” to “verifying value.” The free large models that once blossomed everywhere have finally reached the junction where charging is unavoidable.
It’s Not That They Don’t Want to Stay Free — They Simply Cannot Afford to Keep Burning
Many users are puzzled: AI chat is just typing back and forth — why is it so expensive?
The cost logic of AI large models is the complete inverse of traditional internet products like Douyin or WeChat. With a conventional app, the more users there are, the closer marginal costs come to zero, and traffic monetization becomes a reliable profit engine. Large models are different: the more active users are, the deeper their usage, the higher the platform costs — it is a pure case of “the more you use, the more you lose.”
A set of figures makes the pressure plain: as of March 2026, Doubao’s large model daily average Token invocation volume surpassed 120 trillion, a 1,000-fold surge compared with the time of its 2024 launch, processing more than 13.8 billion token requests per second. Text inference costs alone run to tens of millions of yuan per day, translating to tens of billions of yuan in annual expenditure. Add to this the fact that the paid version’s headlining features — PPT generation, data analysis, film and television production, text-to-video generation — consume hundreds of times more Tokens than ordinary chat, and costs escalate exponentially.
The investments behind this are staggering: in 2025, ByteDance’s capital expenditure exceeded RMB 150 billion (≈ USD 20.83 billion), the vast majority directed at the AI domain; a dedicated RMB 85 billion (≈ USD 11.81 billion) has been reserved for AI chip procurement in 2026; 2025 net profit fell more than 70% year-on-year, and internal voices calling for Doubao’s commercialization have grown increasingly loud.
More brutally still, the entire AI supply chain has already entered a full-chain price increase cycle. Tencent Cloud, Alibaba Cloud, and Baidu Intelligent Cloud have collectively raised computing power prices, with maximum increases exceeding 30%; high-end GPUs and HBM high-bandwidth memory are in short supply with prices doubling, and hardware cost pressures have propagated all the way through to the application layer. Even overseas, OpenAI — with its USD 20 per month Plus membership — continues to lose tens of billions of yuan annually. Doubao’s free model simply had no long-term viability.
Retaining a free version is primarily about holding onto the basic user base and maintaining brand mindshare; charging is about recovering the costs associated with high-value, high-consumption heavy users, and no longer having the platform absorb the bottomless “computing power pit” alone.
Pricing Triggers Controversy: Inflated Prices, or Appropriate Value Alignment?
The three-tier pricing structure Doubao introduced is the core of the public controversy.
By horizontal comparison, Doubao’s pricing is not particularly aggressive. Overseas, ChatGPT Plus charges USD 20 per month, with the Pro version reaching as high as USD 200; Claude and Gemini’s paid tiers are similarly substantial in price. Domestically, competing product Kimi’s basic membership starts at RMB 49 per month (≈ USD 6.81); Zhipu and MiniMax basic memberships also fall in the RMB 30 to RMB 60 range (≈ USD 4.17 to USD 8.33), and Zhipu has raised its API prices three times within the year, with cumulative increases exceeding 60%.
ByteDance’s pricing strategy also carries some subtle calculation: using the RMB 68 (≈ USD 9.44) Standard tier as a low-barrier entry point catering to ordinary high-frequency users; using the RMB 500 (≈ USD 69.44) Professional tier as a price anchor that makes the RMB 200 (≈ USD 27.78) Enhanced tier look compelling by comparison, attempting to cover paying users across different tiers.
But netizens are not buying it, and the central contradiction zeroes in on a mismatch between pricing and product capability.
On one hand, within Doubao’s 345 million monthly active users, students and middle-aged and elderly users make up the majority. They use it for everyday chat and looking things up, and have no real need for paid productivity features. Yet the working professionals, creators, and developers who actually have those needs feel that Doubao’s professional capabilities fall short of competing products — code writing, deep processing of long-form text, and multi-Agent task execution are not even on par with DeepSeek’s free version, making the RMB 200 (≈ USD 27.78) Enhanced tier feel like poor value for money.
On the other hand, domestic users’ willingness to pay for AI is far below that of overseas users. Survey data indicates that more than 40% of users consider an acceptable monthly price to be only RMB 48.3 (≈ USD 6.71), well below the RMB 68 (≈ USD 9.44) starting price. The mindset of “I’ve gotten used to free AI” makes paid conversion an uphill battle. Baidu’s Wenxin Yiyan had previously launched a paid membership that was ultimately forced to revert to free due to insufficient user uptake — a cautionary precedent directly in front of them.
There is also a more practical problem: even if users pay, unlimited usage is unlikely to be feasible. Take Doubao’s text-to-video feature as an example — generating 100 seconds of video costs close to RMB 100 (≈ USD 13.89), meaning an RMB 200 (≈ USD 27.78) monthly subscriber can only generate around 2 minutes of content before the allowance runs out and additional payment is required. What membership provides is in essence a limited computing power usage right, not unlimited access.
Industry Sea Change: The Era of Low-Price Tokens Ends, Tiered Services Become the New Normal
Doubao going paid is never an isolated event — it is the collective pivot of the entire large model industry. The two-year-long Token price war has come to a definitive end; the era of free AI is officially over; and tiered services have become the industry standard.
Cast your mind back to 2024: domestic large models were still mired in a “cabbage price” race to the bottom. When Doubao launched, its API pricing was set at 150 times lower than peers, Alibaba Cloud and Baidu Cloud kept slashing prices, Token prices fell to fractional levels, and vendors competed for users and market share using capital subsidies, playing out a “hundred model battle.”
In just over a year, the industry logic has completely reversed. As AI transitions from simple chat toward the AI Agent era, the Token consumption for a complex task is dozens to hundreds of times that of ordinary conversation. Surging demand compounded by hardware supply chain scarcity has turned computing power from abundant to scarce; cloud vendors collectively raised prices, and vendors are no longer able to sustain undifferentiated free access.
Going forward, domestic AI applications will essentially coalesce into four fixed tiers:
Free basic tier: retaining basic functions such as everyday chat, lightweight Q&A, and simple copywriting, used for customer acquisition and retaining mass users; but entitlements will gradually shrink, with slow response during peak hours and feature restrictions becoming the norm.
Low-cost membership tier: targeting ordinary high-frequency users, with higher invocation allowances, faster response speeds, and optimized model performance, covering everyday light office productivity needs.
Professional paid tier: targeting heavy users such as working professionals, creators, and developers, with access to the full suite of productivity features, focused on high-value scenario services.
Enterprise API service: usage-based billing plus customized packages, with dedicated deployment for enterprise clients, forming the core profit center of industry commercialization.
This also means that purely free, unlimited large models will disappear entirely. What the market will offer is either a combination of “free castrated version plus paid full version,” or direct pay-per-use billing. The days of lying back and getting something for nothing are over for good.
AI Competition Shifts From Traffic to Value
Doubao’s move to paid service looks, on the surface, like asking users for money. In reality, it is the beginning of the AI industry’s return to commercial fundamentals.
Over the past two years, the industry competed on user scale and daily active user metrics — whoever offered the most aggressive free access grabbed the early advantage. Now, in the second half, the competitive core has completely shifted: whoever can control computing power costs, whoever can build truly capable products, and whoever can get users to pay for value — those are the ones who will survive to the end.
For ordinary users, there is no need to reflexively resist paying — the basic free features are sufficient for everyday needs, and if genuine high-efficiency office productivity is required, paying to save time is a worthwhile trade. For vendors, charging is not an endpoint but a starting gun that forces product upgrades. Simply raising prices and cutting back on free entitlements will only drive users away; only by delivering tangible, hard-core productivity capabilities can the case for payment be made to stand.
The era of everyone using AI freely and without limit — that era has truly ended. What comes next is the true beginning of AI’s commercial value being realized. And every one of us, as users, is witnessing this great industry transformation.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.