Author: Think AI, Aaron
The most fantastical moment in the AI industry has appeared.
On one hand, Doubao confirmed on June 1 that it would officially begin charging fees by the end of the month, following its paid testing in mid-May. The pricing is relatively high: the Standard Plan is 68 yuan for a continuous monthly subscription, the Enhanced Plan is 200 yuan, and the Professional Plan is 500 yuan.
At this price, Doubao Pro is clearly higher than ChatGPT Plus at $20 per month and even approaches some premium AI subscription tiers overseas.
On the other side, DeepSeek announced a permanent price reduction at the end of May, making the 75% discount on DeepSeek V4-Pro permanent and lowering API prices to one-quarter of the original rate. DeepSeek’s cached input costs as low as RMB 0.02–0.025 per million tokens, and output costs approximately RMB 2–6 per million tokens, setting a new global low. Regarding DouBao’s pricing, online criticism has been widespread.
"DouBao is dumb and charges money," "I'll uninstall DouBao if it charges"—these phrases have consistently trended recently. Meanwhile, Liang Wenhong has received widespread praise, with DeepSeek hailed as a pride of China’s domestic AI industry. Why have two of China’s leading AI companies announced major moves at the same time? After such drastically different strategies, which company will ultimately come out on top?
Why go against it?
One is raising prices, the other is lowering them—essentially, the two are AI companies pursuing different strategies. Doubao focuses on product experience and primarily targets C-end users. DeepSeek aims to capture the B-end market, emphasizing model invocation. First, let’s look at Doubao: according to QuestMobile data, Doubao currently has the highest daily and monthly active users among AI products in China, with 345 million monthly active users, while DeepSeek has approximately 127 to 130 million.
Doubao's pricing model essentially signals the end of the free model for AI models, with paid subscriptions being the primary revenue stream for AI companies targeting end consumers. ByteDance possesses massive consumer traffic, driven by Douyin and Toutiao, and initially rapidly captured users through a free-plus-subsidy strategy.
However, DouBao’s current daily usage of over 120 trillion tokens incurs massive computational costs, particularly in complex productivity scenarios such as PowerPoint generation, video creation, and data analysis, which consume significant token resources. ByteDance has stated that its AI investment will rise to RMB 200 billion by 2026, with daily spending exceeding RMB 500 million—much of which goes toward foundational resources like computational power. This free model is unsustainable. The enormous investment has significantly reduced ByteDance’s profitability in the first quarter.
Currently, Doubao's paid features target heavy users and advanced functionalities, while basic chat remains free, aiming to balance scale through free access with monetization through premium services. However, users are concerned that the free version of Doubao may be downgraded in intelligence or face usage restrictions in the future. DeepSeek’s price reduction is not merely a price war—it reflects confident strength built upon a well-established competitive moat.
Through architectural innovation, DeepSeek has reduced computational consumption to just 27% of the previous generation when processing long contexts of up to millions of tokens, achieving a technical reduction in unit inference cost. It has also attained a degree of computational autonomy by deeply optimizing the model for domestic computing platforms such as Ascend, reducing reliance on overseas high-end hardware and significantly lowering hardware procurement costs. Furthermore, engineering optimizations play a crucial role: extreme optimizations on the inference side enhance computational utilization, and economies of scale help amortize fixed costs, creating a virtuous cycle where increased usage drives down costs.
This technology-driven cost reduction makes the price cuts sustainable. In the enterprise model invocation segment, DeepSeek has established a certain competitive advantage; by continuing to lower prices and deeply expand into the enterprise market, it has the potential to become the most widely used foundational model domestically. According to OpenRouter data, DeepSeek V4 ranked first globally in total large model invocation volume over the past month. Perhaps one day, DeepSeek could become the “Android” of AI.
Does AI have better profit models?
However, whether it’s Doubao or DeepSeek, both are currently in a phase of heavy spending and losses. Even if Doubao moves toward a paid model, it can only offset the enormous costs of computing power and still struggles to achieve profitability. In comparison, ChatGPT primarily relies on user subscriptions, yet OpenAI continues to incur massive losses.
Anthropic has already achieved profitability, with market estimates projecting its annual recurring revenue (ARR) to reach $47 billion by 2026, prompting new considerations for the entire AI industry. How did Anthropic accomplish this? Primarily, over 80% of its revenue comes from enterprises and developers; its customers offer high customer lifetime value, predictable query patterns, and the company focuses on high-ROI use cases such as code generation and agents.
This offers the market a new perspective. If AI merely chats with users, it can only charge tens or hundreds of dollars in subscription fees. But if AI can help businesses save on labor costs, it can charge software fees. When AI truly integrates into workflows and solves real work problems, that’s where AI begins to generate significant revenue. In other words, merely developing large AI models will lead to increasing competition and ultimately become unsustainable due to cost pressures. Only by establishing a complete business ecosystem—embedding models into real-world environments and applications—can companies fully escape the cycle of losses.
Recent reports indicate that DeepSeek’s Series A funding round reached approximately $7 billion, pushing its valuation to $59 billion, with Liang Wenhong personally investing $20 billion and Tencent contributing $10 billion. Post-funding, DeepSeek will maintain its leading advantage in the B2B sector, and with the addition of Tencent and other richer industrial use cases, its competitive edge will be significantly strengthened. Doubao’s strength lies in its scalable closed-loop ecosystem. Ultimately, the winner in this industry will be the player who successfully transforms AI into tangible ROI—whether through a C-end productivity ecosystem or a B-end agent platform.
AI commercialization is still in its early stages—let’s wait and see.