Zhang Yong launched a surprise attack.

After Zhang Yong's frequent actions since taking over, Alibaba Cloud has once again thrown a bomb at the AI industry, announcing the open source of two AI models and launching a surprise attack on Huawei, Baidu, Tencent, JD.com, and other major companies.

On August 3rd, the 7 billion parameter general model and dialogue model (Qwen-7B, Qwen-7B-Chat) of Tongyi Qianwen were launched on Alibaba Cloud's AI Magic Community. Both models are open source and can be used for free or commercially.

Alibaba Cloud has also become the first domestic internet giant to join the queue of large-scale model open source.

With the wave of AI, more and more players are entering the game, and the large-scale models that have sprung up like mushrooms after the rain are no longer mysterious, and the industry is beginning to return to rationality. At this time, Alibaba Cloud, which has opened the door to open source, may be brewing a "conspiracy".

Zhang Yong intends to leverage the power of developer users to break through the opponents' moats with a small force and consolidate his own position. And open source is the leverage to launch this attack.

"Slaying the Dragon"

According to Wall Street News, the open source Qwen-7B this time is a base model that supports multiple languages such as Chinese and English, trained on a dataset of over 20 trillion tokens; Qwen-7B-Chat is a Chinese-English dialogue model based on the base model.

Alibaba Cloud revealed that users can deploy and run the above two open source AI models on "consumer-grade" graphics cards, and they can also access and call them through Alibaba Cloud's Lingji platform.

Insiders said that the open source of these miniaturized model versions is expected to eliminate the threshold for model usage, allowing a large number of small and medium-sized enterprises and AI developers to use Tongyi Qianwen earlier and faster.

This means that many research institutions and commercial companies can use the open source Tongyi Qianwen for free, without having to develop a large-scale model at a huge cost, reducing enterprise R&D costs and also benefiting Alibaba Cloud in attracting more developer partners.

Binding more users through open source is just the surface, and the greater significance behind it is to generate "viral" dissemination and seize market mindshare.

At the WAIC Forum this year, Wang Hua, CEO of Innovation Works, mentioned that in the future, large-scale models will explode in all aspects, and the open source model is a truly valuable scenario.

"Now everyone thinks that large-scale models themselves are very valuable, but in two or three years, they will become commonplace. What will be more important in the future is the scenario." Wang Hua pointed out that developers need to be leveraged, "they can create value by supporting applications in various industries through services such as SaaS".

Zhang Meng, founder of TabbyML, an AI solution provider, also believes that open source itself is a very ideal option from a commercial perspective in the developer scenario.

In his opinion, the applications in the industry are diverse, which is a state that closed-source models find it difficult to catch up with. And open source that is free and can be used commercially also allows users who are unwilling to pay to have the opportunity to become contributors to the community and generate value.

In fact, the industry has already staged the script of the "dragon-slaying" with open source as the killer.

In July, Microsoft announced a partnership with Meta to release the commercially available version of the open source AI model Llama2, providing a "like-for-like" alternative to OpenAI and Google models. Afterwards, various "modified" llamas flooded open-source communities such as HuggingFace, including Stanford's Alpaca, UCB's Vicuna, and many others. Numerous modified models based on Llama emerged and climbed up the leaderboard for large-scale model performance.

Many developers believe that Llama's "open-source" nature weakened its competitors and caused deep anxiety for Google and OpenAIII.

"We didn't win this competition, and neither did OpenAIII. While we were still arguing, a third party quietly took our rice bowl - open-source." On May 5th, leaked internal Google documents recorded the above content in a Discord group.

Similarly, for competitors like Baidu's "Wenxin Yiyuan," Huawei's "Pangu," and even Tencent's "Hunyuan" and JD.com's "Yanxi," which are still in the testing phase, the first shot fired by Tongyi Qianwen has made them fearful.

Positioning

Open-source has changed the competitive landscape for large-scale models, but for Alibaba Cloud, this is not all. Zhang Yong has a bigger plan in mind - the explosion of the model ecosystem community, which will drive the Matthew effect to manifest itself in Alibaba Cloud.

At the AGI Playground conference at the end of July, Yin Yifeng, an engineer from the AI model open-source community Hugging Face, pointed out that open-source has a catalytic effect. "From 0 to 1 work, it may require exceptionally smart experts working behind closed doors. But from 1 to 100 work, it can be done quickly in an open-source community."

This means that "open-source + community" will synergistically catalyze greater impact.

The race for the ecosystem is already in full swing among major foreign companies. For example, AWS recently reached revenue-sharing agreements with well-known AI large-scale model companies such as Hugging Face and Stability AI, strengthening the MaaS ecosystem.

Zhou Jingren, CTO of Alibaba Cloud, revealed at WAIC in early July that their Moda community has become the largest AI model community in China, attracting over 1.8 million developers.

And they all have the opportunity to become soldiers in Alibaba Cloud's conquest. With the continuous attraction of new participants due to the open-source nature of models, the community will continue to grow.

The combination of open-source and community will ultimately bring greater imagination to players in the AI model community. "When the picture is exhausted, the dagger is revealed." This is what Zhang Yong is looking forward to.

Since the GPT frenzy last year, AI model open-source communities like Hugging Face have received capital favor. In May last year, Hugging Face announced the completion of a $100 million Series C financing, with a valuation of $2 billion, becoming a unicorn.

At this time, as the process of independent separation gradually unfolds, Zhang Yong, who is busy working to boost Alibaba Cloud's value in preparation for an IPO, is intending to take advantage of the situation and elevate Alibaba Cloud's worth. After all, this is a tough battle. He hopes that Alibaba Cloud, as an independent company, can grow as big as Alibaba Group today.

From price reductions and enhanced AI model community services to open source initiatives, Alibaba Cloud, the leader in domestic cloud services, always strives to stay ahead. The "big brother" who disrupts the market is actually driven by concerns over stagnant market share and revenue growth.

Since the first quarter of 2016, after achieving a peak revenue growth rate of 175%, Alibaba Cloud has been experiencing consecutive annual declines. By the 2023 fiscal year, the year-on-year growth rate dropped to single digits at 3.5%. Behind them, there are state-owned cloud providers such as China Telecom Cloud, China Mobile Cloud, and China Unicom Cloud, which are rapidly catching up at a double-digit growth rate.

In addition, Alibaba Cloud also faces new competitors in specific market segments.

According to recent data released by IDC, in the ranking of China's AI public cloud service market share in 2022, Baidu Cloud Intelligence ranked first with a 28.9% market share, while Alibaba Cloud ranked second with 27.4%. Although the gap is not significant, Alibaba Cloud still feels the threat.

By leveraging open source initiatives this time, Alibaba Cloud will greatly stimulate user demand and find new possibilities for the growth of cloud computing market share, the expansion of the AI community ecosystem, and the cost reduction of large-scale model iterations, while also providing an extra layer of security.

However, after the wave of open source initiatives in China, major players such as Baidu, Tencent, Huawei, JD.com, and Meituan are not without the possibility of launching counterattacks.

Can Alibaba Cloud and Zhang Yong, who have launched another surprise attack, reshape the landscape and secure their position? What kind of prospects will this bring to Alibaba Cloud's spin-off listing? The market is eagerly anticipating the outcome.