Author: Think AI, Aaron

The most magical scene in the AI industry has appeared.

On one side, Doubao confirmed on June 1 that it will officially start charging at the end of the month after testing a fee in mid-May. The charging prices are quite high: the standard version costs 68 yuan for a monthly subscription, the enhanced version is 200 yuan, and the professional version is 500 yuan.

At this price, Doubao's professional version is clearly higher than ChatGPT Plus's $20/month, and even approaches some overseas high-tier AI subscription levels.

On the other side, Deepseek announced a permanent price reduction at the end of May, making the 75% discount for V4-Pro permanent and reducing API prices to a quarter of the original price. Deepseek's cache hit input is as low as 0.02-0.025 yuan per million tokens, with output about 2-6 yuan per million tokens, creating a new global low. Regarding Doubao's charging, there has been a lot of online criticism.

“Doubao is foolish to charge” and “I will uninstall Doubao if it starts charging” have recently continued to trend online. In contrast, Liang Wenfeng has received much praise, and Deepseek is described as a shining star of domestic products. Why did two leading AI companies in China announce major moves at the same time? After such opposing actions, which company will ultimately succeed?

Why the divergence?

One is raising prices while the other is lowering them, essentially two AI companies taking different paths. Doubao focuses on product experience, primarily targeting end-users. Deepseek aims to capture the B-end market, focusing on model invocation. Looking at Doubao, it currently has the highest daily and monthly active users among domestic AI products, with QuestMobile data showing Doubao's monthly active users reaching 345 million, while Deepseek has about 127-130 million.

Doubao's charging basically announces the imminent end of the free model for AI models, and subscription fees are currently the main way for AI companies targeting end-users to generate profits. ByteDance has massive end-user traffic, leveraging platforms like Douyin and Toutiao to quickly capture users through free services and subsidies in the early days.

However, currently, Doubao's daily average of over 120 trillion token calls brings significant computing costs, especially in complex productivity scenarios such as PPT, video generation, and data analysis, which consume Tokens heavily. Currently, ByteDance claims that its investment in AI will increase to 200 billion by 2026, with daily investments exceeding 500 million, a significant portion of which is invested in computing power and other basic resources, making the free model unsustainable. The huge investments have also substantially lowered ByteDance's profitability in the first quarter.

Currently, Doubao's paid features target heavy users and high-end functionalities, while basic chatting remains free, aiming to achieve a balance between free growth and value-added monetization. However, users are concerned whether the free version of Doubao will be reduced in intelligence and whether there will be usage restrictions in the future. Deepseek's price reduction is more than just a price war; it also demonstrates a solid confidence that has formed a moat.

Through architectural innovation, Deepseek has made it so that the V4 series consumes only 27% of the computing power of its predecessor when processing millions of tokens in long contexts, achieving a technical reduction in per inference cost; it has also realized a degree of computing power autonomy, deeply adapting models to home-grown computing power like Ascend, thus reducing dependence on overseas high-end computing power and significantly lowering hardware procurement costs; furthermore, engineering optimization plays an important role by performing extreme optimization on the inference side to enhance computing power utilization and creating a beneficial cycle of “usage recouping costs” by diluting fixed costs through economies of scale.

This technology-driven cost reduction provides sustainability for the price cuts. In the enterprise-level model invocation arena, Deepseek has established a certain degree of moat, continuing to expand its enterprise market deeply through price reductions, and may become the model foundation with the highest usage domestically. According to Openrouter data, in the past month, the model invocation volume of Deepseek V4 ranks first globally. Perhaps one day Deepseek could become the “Android system” of AI usage.

Is there a better profit model for AI?

However, whether it is Doubao or Deepseek, both are currently in a phase of burning money and losses. Even if Doubao heads towards charging, it is merely compensating for huge computing costs and still finding it hard to achieve profitability. Compared to ChatGPT, most rely on a subscription model for revenue generation, and currently, OpenAI is still facing substantial losses.

In contrast, Anthropic has already achieved profitability, with market estimates predicting that its ARR (annual revenue) will reach $47 billion by 2026, bringing new thoughts to the entire AI industry. So how did Anthropic do it? Mainly because over 80% of its revenue comes from enterprises and developers, its customers have high customer value per account and predictable queries, and the company focuses on high ROI scenarios such as code/agents.

This provides the market with a new perspective. If AI is just chatting with users, then it can only charge a few dozen or a few hundred for membership fees. But if AI can save manpower for enterprises, it can charge software fees. When AI truly integrates into workflows and solves work problems, this is where AI starts to make substantial profits. In other words, merely developing a generic AI large model will lead to increasing internal competition and, ultimately, due to cost pressures, become unsustainable. Only by forming a complete commercial closed loop, embedding the model into real ecosystems and application scenarios can we completely escape the trap of losses.

DeepSeek's latest news indicates a first round financing of about $7 billion, with its valuation climbing to $59 billion, during which Liang Wenfeng invested 20 billion from his own pocket, and Tencent entered the scene with 10 billion, among others. After financing, it can continue to maintain its leading position in the B-end market, and if Tencent and other richer industrial scenarios come in, it will greatly strengthen its advantage. Doubao excels in its scale closed loop. The ultimate winner in the industry is the player that truly converts AI into ROI—whether it is in the C-end productivity ecosystem or the B-end agent platform.

The commercialization of AI is still in its early stages; let's wait and see.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。