Alibaba Cloud triggers a fierce battle

On May 21st, Aliyun announced that the price of its main model Qwen-Long, a GPT-4 level model, has been reduced by 97% from 0.02 yuan/thousand tokens to 0.0005 yuan. Just a week before, ByteDance officially released the Dou Bao large model, with the Dou Bao universal model pro (0.0008 yuan/thousand tokens) claiming to break the industry's bottom price, being 99.3% lower than Ali's Qwen 2.5-max (30k) at 0.12 yuan/thousand tokens.

It is evident that Aliyun is challenging ByteDance head-on. With the leading cloud service company entering the battle, the price war for large models has intensified.

After Aliyun announced the price reduction, Baidu announced that its two main models, ENIRE Speed and ENIRE Lite of the Wenxin large model, are now free; iFlytek also announced that the iFlytek Xinghuo Lite API is permanently open for free, and the iFlytek Xinghuo top version (Spark3.5 Max) API is priced as low as 0.21/ten thousand tokens; Tencent's Hybrid model also followed suit, with the lite model becoming completely free, and the standard and pro versions seeing price reductions of 50-70%.

Commercial large models have now entered the era of "sub-yuan" or even "zero yuan". In practical terms, with 1 yuan in the real world, you can only take a bus ride, but in ByteDance's Dou Bao, you can read three copies of "Romance of the Three Kingdoms" (1.25 million tokens, equivalent to about 2 million Chinese characters); in Aliyun's Tongyi Qianwen, you can directly master five copies of the "Xinhua Dictionary" (2 million tokens, equivalent to about 3.2 million Chinese characters).

This price war in the domestic large model industry resembles past internet strategies such as the "ride-hailing war" and "billion-dollar subsidies", indicating that the industry has moved from a technological competition to an early stage of customer acquisition and market expansion. They are trying to attract more users to develop applications in their own ecosystems and accelerate the implementation of AI applications.

In fact, before Ali and ByteDance slashed prices, the trend of price reductions had already begun in the large model industry.

In early May, the AI company DeepSeek under the private equity giant Magic Cube fired the first shot by launching a new version of a large model with a price reduction to one thousand tokens/0.001 yuan. Subsequently, AI star companies such as Zhipu and Mianbi announced price reductions, and soon the four giants ByteDance, Aliyun, Baidu, and Tencent joined the melee, with a strong scent of blood in the air.

In the eyes of industry insiders, AI companies that have been riding the wave of technology for over a year have yet to see a trend of large-scale explosion in the application of large models, with commercial monetization still a distant prospect. However, at a time when the internet industry has reached its peak in terms of users and traffic, and technology is at a key point of platform ecosystem reconstruction, there is immense room for imagination. Perhaps the price war for large models at this moment is an efficient catalyst President Tan Dai of Volcano Engine pointed out that the landing of AI application scenarios requires both technological improvement and the aggregation of scattered individual wisdom in the application ecosystem. "It must be easy to get started in order to do this."

"In the past, enterprises were worried about the difficulty of achieving a positive ROI in innovation, with 90% of innovative projects failing. Now, after the model prices have dropped, the cost of trial and error has directly decreased from 800,000 yuan to 8,000 yuan. The cost of failure is controllable, which has allowed many enterprises to join the AI wave unburdened." Tan Dai said, "Imagine in the future when large models are used in people's daily lives every day, low prices are the real prerequisite for widespread application, so price reduction is the most urgent matter at present."

In fact, after a year of brutal growth, large models have also tended to homogenize the user experience, and the growth curve of iterations has begun to slow down. Even the super unicorn OpenAI has started to emphasize "cost-effectiveness." In mid-May, OpenAI announced the latest GPT-4o, where iterative changes at the technical level are no longer phenomenal revolutions, but the eye-catching point is the "50% off everything" promotion.

Since the beginning of last year, OpenAI has continuously released new models and reduced prices four times, with an overall reduction range of 50-75% per year. Even with industry benchmarks doing so, AI companies in China, influenced by the internet era, naturally started to play the familiar "low-price competition."

For downstream enterprises, the cost of reasoning is further reduced, and the AI application industry is expected to reach a cost inflection point. Huolong Securities believes that more C-end users are expected to use basic AI applications for free, and the huge volume of visits will help enterprises further improve their model service capabilities, completing a virtuous cycle.

"Alibaba Cloud's significant reduction in large model inference prices is aimed at accelerating the explosion of AI applications. It is expected that the daily API call volume for large models this year will exceed 10 billion." Liu Weiguang, Vice President of Alibaba Cloud and President of the Public Cloud Business Unit, said frankly.

Using the large model ecosystem as bait and leading user companies into their own cloud service "fishpond" is Alibaba Cloud's "sunny strategy" to solve the growth dilemma. Currently, more enterprises call large models through "public cloud + API," reducing the price of large models or even offering them for free, which is beneficial for Alibaba Cloud to attract more developers and enterprise users, build a brand-new ecosystem in the field of AI, and even further expand its market share.

This also makes Alibaba Cloud's actions in this round of large model price wars so rapid.

The fourth-quarter financial report for the 2024 fiscal year shows that although Alibaba Cloud continues to be profitable, its revenue only increased by 3% year-on-year to 25.595 billion yuan. If the revenue from Alibaba's consolidated businesses is excluded, Alibaba Cloud's revenue has been declining year-on-year for several quarters.

During the financial report meeting, Alibaba stated that its current cash flow is sufficient for strategic reinvestment, and it is expected to produce three major results in the future, one of which is to have Alibaba Cloud return to double-digit growth in the second half of the 2025 fiscal year. Eddie Wu, CEO of Alibaba Cloud, stated that public cloud and artificial intelligence products will be the main driving force.

According to Eddie Wu's expectations, Alibaba Cloud aims to become Alibaba's "second growth curve" towards the future Including Bytedance's Volcano Engine, Baidu Cloud and other cloud service providers are all trying to follow the benefits of AI, launching attacks on cloud giants such as Alibaba Cloud and Tencent Cloud. The cloud service market is also reshuffling and restructuring in the era of AI.

A large-scale elimination match is unfolding. However, as this competition allows more developers to develop AI applications at a lower cost, various internet giants and AI startups are gradually building an ecosystem for AI. AI is already in the hands of ordinary people, not far away