Author: Think AI, Aaron
The most surreal scene in the AI industry has emerged.
On one side, after testing a paid model in mid-May, Doubao confirmed on June 1st that it will officially begin charging fees by the end of the month. The pricing is quite high: the Standard version costs 68 RMB/month for continuous subscription, the Enhanced version 200 RMB, and the Professional version 500 RMB.
Calculated at these prices, Doubao's Professional version is already significantly higher than ChatGPT Plus's $20/month, even approaching some overseas high-tier AI subscription levels.
On the other side, DeepSeek announced a permanent price cut at the end of May. DeepSeek has made the 75% discount on V4-Pro permanent, reducing API prices to a quarter of the original. DeepSeek's cached hit input costs are as low as 0.02-0.025 RMB per million tokens, and output costs about 2-6 RMB per million tokens, setting a new global low. Regarding Doubao's fees, online criticism is widespread.
“Doubao is dumb and still charges” and “If Doubao charges, I'll uninstall it” have recently trended on hot searches. In contrast, Liang Wenfeng has received widespread praise, with DeepSeek being hailed as a shining example of domestic AI. Why did two leading domestic AI companies announce major moves in the same timeframe? Following these diametrically opposed operations, which company will have the last laugh?
Why the Divergence?
One raises prices, the other lowers them; essentially, the two companies are following different AI strategies. Doubao focuses on product experience, primarily targeting C-end users. DeepSeek strives to capture the B-end market, focusing on model calls. First, look at Doubao. Currently, Doubao has the highest daily active users (DAU) and monthly active users (MAU) among domestic AI products. According to QuestMobile data, Doubao's MAU reaches 345 million, while DeepSeek's is about 127-130 million.
Doubao's move to charge essentially signals the impending end of the free model for AI, and paid subscriptions are currently the main monetization method for AI companies targeting C-end users. ByteDance possesses massive C-end traffic, with引流 from Douyin and Toutiao, and initially used a free + subsidy model to rapidly capture users.
However, Doubao's current daily call volume exceeding 120 trillion+ tokens incurs massive computing power costs, especially in complex productivity scenarios like PPT generation, video creation, and data analysis which consume significant tokens. ByteDance currently states that its AI investment will increase to 200 billion RMB by 2026, with a daily investment exceeding 500 million RMB. A large portion of this is in computing power and other foundational resource investments, making the free model unsustainable. The massive investment has also significantly reduced ByteDance's company profits in the first quarter.
Currently, Doubao's fees target heavy users and premium features, with basic chat remaining free, aiming to balance free scale with value-added monetization. However, users worry whether the free version of Doubao will become 'dumber' or face usage restrictions later. DeepSeek's price cut is more than just a price war; it's a confident stance built upon an established moat.
Through architectural innovation, DeepSeek's V4 series consumes only 27% of the computing power of the previous generation when processing million-token-long contexts, achieving a technical reduction in unit inference cost. To some extent, it has achieved computing power autonomy, with models deeply adapted to domestic computing power like Ascend, reducing reliance on overseas high-end computing power and significantly lowering hardware procurement costs. Furthermore, engineering optimization capabilities play a crucial role. Extreme optimization on the inference side improves computing power utilization. Economies of scale dilute fixed costs, forming a virtuous cycle where 'usage feeds back into cost reduction.'
This technology-driven cost reduction makes the price cuts sustainable. In the enterprise-level model call race, DeepSeek has established a certain moat. By lowering prices, it continues to deeply expand into the enterprise market and is expected to become the most widely used model base domestically. According to Openrouter data, DeepSeek V4 ranked first globally in large model call volume in the past month. Perhaps one day, DeepSeek has the potential to become the 'Android system' for using AI.
Are There Better Monetization Models for AI?
However, whether it's Doubao or DeepSeek, both are currently in a stage of burning money and incurring losses. Even if Doubao moves to charge, it only helps offset the enormous computing power costs and will still struggle to achieve profitability. Compared to ChatGPT, which mostly relies on user subscription fees, OpenAI is still incurring heavy losses.
In contrast, Anthropic has taken the lead in achieving profitability, with market estimates projecting its Annual Recurring Revenue (ARR) to reach $47 billion by 2026, bringing new considerations to the entire AI industry. How did Anthropic do it? Mainly because over 80% of its revenue comes from enterprises and developers. Its customers have high customer value, predictable queries, and the company focuses on high-ROI scenarios like coding/Agents.
This offers a new line of thinking for the market. If AI merely chats with users, it can only charge membership fees of tens or hundreds. But if AI can save manpower for enterprises, it can charge software fees. When AI truly integrates into workflows and solves work problems, that's where AI starts making big money. In other words, solely developing pure AI large language models will not only become increasingly competitive but will ultimately become unsustainable due to cost pressures. Only by forming a complete commercial loop, embedding the model into real ecosystems and application scenarios, can companies completely escape the loss trap.
Latest news indicates DeepSeek has raised about $7 billion in its first funding round, with its valuation climbing to $59 billion. Liang Wenfeng personally contributed 20 billion RMB, Tencent invested 10 billion RMB, among others. Post-funding, it can continue to maintain its leading edge in the B-end market. If it can leverage Tencent and other richer industrial scenarios, it will greatly strengthen its advantage. Doubao excels in scale and closed-loop. The ultimate winner in the industry will be the player that truly converts AI into ROI—whether it's a C-end productivity ecosystem or a B-end agent platform.
AI commercialization is still in its early stages. Let's wait and see.
