Text/Chen Feng
Edit/midnight
In the past six months, the big model boom triggered by OpenAI has created the hottest technology boom in China's business world.
People call this change a great AI revolution, and some even believe that this is another landmark paradigm revolution ushered in by mankind after the steam age, and the change is no less powerful than the mobile Internet.
Under the boom, Internet giants represented by Ali and Baidu actively bet on large models, and at the same time transform existing businesses through large models.
At the Alibaba Cloud Summit on April 11, Alibaba Cloud launched the "Tongyi Qianwen" large model, and announced that all Alibaba products will be transformed with large models; Later, in a conference call after Baidu's first-quarter earnings report, Baidu chairman Robin Li also said that Baidu plans to steadily incorporate Wenxin into all of Baidu's businesses.
At the same time, cloud computing, which is the cornerstone of the underlying model of a large model, has also undergone new changes. At the current stage, cloud vendors have basically reached a consensus that the future business model of cloud computing will evolve in the direction of MaaS (model as a service), and applications will be built on large models on a large scale.
For cloud vendors, especially Internet cloud vendors, this is a new challenge and a new opportunity.
In 2022, due to multiple factors such as changes in the macro environment, the acceleration of market grabbing by operators' cloud vendors, and the accelerated transfer of cloud computing customers to non-Internet customers, Internet cloud vendors will generally enter a cycle of declining growth rates.
How to reverse this situation is a major challenge for cloud vendors.
Since 2022, cloud vendors have strategically reduced costs and increased efficiency, gradually eased their posture, from determined to do general contracting to acceptable "integration", constantly exploring automotive cloud and other outlet industries in layout, and constantly starting price wars in competition, etc., are all adjustment measures they have made in response to the slowdown in growth.
In the first half of this year, embracing the large model has become a new direction for cloud vendors.
Under the fierce competition, each family closely follows the actions of their peers, and a new round of racing is being staged.
1. The growth rate has slowed down, competition has intensified, and the pressure on Internet cloud vendors has doubled
Since 2022, cloud computing vendors have not had a good time.
According to Canalys, China's cloud computing market grew by 10% to $30.3 billion in 2022, far less than the strong performance of the past few years. Over the past three years, the cloud computing market has grown at an annual rate of more than 30%.
The slowdown in growth is inseparable from uncertainties such as the epidemic. When faced with too much uncertainty, enterprises are less enthusiastic about adopting the cloud, and they are more cautious about IT budgets and more focused on the operational cost efficiencies that the cloud can bring.
Canalys pointed out that while the current negative impact on enterprises is gradually being eliminated and cloud demand is expected to return, cautious IT budgets will make it difficult for cloud growth to return to peak. It expects spending on cloud infrastructure services in China to grow by 12% for the full year in 2023.
In this context, cloud vendors are also under pressure to grow.
Alibaba Cloud is still the largest vendor in the domestic public cloud market, but in 2022, its revenue growth rate dropped from 29.7% in 2021 to 7.0%, and its growth rate remained at a high level of more than 50% for a long time before 2020. Baidu Intelligent Cloud's revenue growth rate also dropped from 64% in 2021 to 23% in 2022.
Source/IDC
However, the same cloud vendors, in 2022, Internet cloud vendors and operator clouds have shown different trends - operator clouds accelerate to grab customers and become the biggest competitors of Internet cloud vendors.
IDC pointed out in the tracking report "China's Public Cloud Service Market (First Half of 2022)" that China Telecom's Tianyi cloud market share increased by 0.7 percentage points compared with the second half of 2021, reaching 11.0%, and surpassed Tencent, ranking third after Ali and Huawei.
Let's look at the rate of revenue growth. In the first half of last year, Tianyi Cloud's revenue increased by 100.8% year-on-year, and its half-year revenue has exceeded that of the whole year of 2021; Mobile cloud revenue increased by 103.6%, and the market size has increased tenfold in the past three years, with a compound annual growth rate of more than 200%. In the second half of the year, the operator cloud also maintained this growth trend.
For comparison, for the whole year of 2022, Alibaba Cloud, HUAWEI CLOUD, Tencent Cloud, and Baidu Intelligent Cloud combined revenue growth rate of 9%.
This is closely related to the increasingly aggressive market strategy of operator clouds. According to Caixin, at present, the three operators have given a lot of discounts to grab cloud orders, especially China Unicom, and some products can even be discounted by four or five.
Externally, it is a price war, and internally, Operator Cloud continues to strengthen the sales incentive policy of selling cloud, encouraging salespeople to sell more cloud. A middle executive of Unicom Province revealed to the media that the cloud sales incentives given by Unicom companies in various places are different, generally between 5% and 15% (5% to 15% higher than selling traditional businesses).
The logic behind the acceleration of operators' cloud market grabbing is not difficult to understand - the high growth rate of Internet cloud vendors in the past was mainly due to the dividends of large-scale cloud migration of Internet enterprises, but as the cloud migration rate of Internet enterprises continues to rise, the dividends disappear, and non-Internet customers have become a new increment in the current cloud market.
Referring to the North American market, Accenture pointed out in a report that in the North American cloud computing market, the proportion of pan-technology enterprises is only about 20%.
At the same time, in the face of more traditional enterprises' cloud migration requirements, the advantages of operator cloud will be further reflected, such as more in line with the compliance requirements, cost requirements, and security needs of government and enterprise customers.
It is worth noting that Internet cloud vendors have also increased their cloud layout for non-Internet enterprises in recent years, and the proportion of revenue from this part of customers is also rising. However, their overall revenue growth began to slow down, showing the strength of the operator cloud.
Source: Huatai Securities
In other words, with the structural transformation of cloud customers in China's cloud market, the operator cloud, which was questioned in previous years that "has lost opportunities in the public cloud market", has ushered in the opportunity to break through again, while bringing new pressure to Internet cloud vendors.
2. Price war, lowering posture, Internet cloud vendors actively defend
Dismantling the pressure faced by Internet cloud vendors, the problems they want to solve can actually be boiled down to two points: how to expand more non-Internet customers, and how to deal with the fierce offensive of operator clouds.
Looking back at the various actions of cloud vendors in the past year, it is not difficult to find that their initiatives can be roughly divided into the following types.
On the one hand, it is to fight a price war and exchange low prices for the market. According to Wired Insight's statistics, since the first half of this year, except for Baidu Intelligent Cloud and Huawei Cloud, almost all other cloud vendors have been involved in the price war.
First, on April 26, Alibaba Chairman and CEO Daniel Zhang announced at the Alibaba Cloud 2023 Partner Conference that Alibaba Cloud will carry out the largest price reduction in history, with the price of some core products reduced by 15% to 50% across the board, and the maximum price reduction of storage products by 50%. Immediately after May 16, Tencent Cloud also announced price reductions for a number of core products, with some product lines falling by up to 40%.
Source: Alibaba Cloud WeChat public account
Less than 50 minutes after Tencent Cloud announced the price reduction, Mobile Cloud quickly followed up and reduced the price of a number of cloud products, with some product lines reducing by up to 60%; On the same day, Tianyi Cloud also released a head-image advertisement on its official website that "90% of the core product profit is as low as 1 discount", saying that the new S7 cloud host is 2.5 folded and the database is 1.5 folded.
This is not the first time the price war in the cloud computing market has begun.
As early as 2017, when the Xiamen Information Center publicly tendered for extranet cloud services, Tencent Cloud won the bid with a penny, which caused heated discussions in the industry.
Earlier, in the second half of 2016, Alibaba Cloud announced price reductions in October and December, with core cloud products reduced by up to 50%; Tencent Cloud also launched a discount campaign of 3% off for four core products, and later launched a minimum discount of 3-5% off for all products.
Even from a global perspective, cloud computing price wars are not unusual, Amazon CFO Tom Sukutak has said that price cuts are a common thing - in the past few years, Amazon's price cuts have been extremely frequent, setting a record of more than 50 price cuts in ten years.
However, for Internet cloud vendors, price wars are not a long-term solution, especially in recent years, cost reduction and efficiency improvement have become the keywords of large manufacturers. In other words, the price war is only a response to the escalation of competition in the short term.
On the other hand, Internet cloud vendors began to change their thinking and began to pursue more quality growth.
The obvious change is that from Alibaba Cloud to Tencent Cloud, they have begun to let go of the role obsession of only wanting to be an integrator in the past, and have instead lowered their posture and are willing to be integrated.
On December 15 last year, at Tencent's internal staff meeting with the theme of "reducing costs and increasing efficiency", Ma Huateng officially set the tone and expressed full support for CSIG's strategy of abandoning its role as an integrator and switching to the integration of self-developed products.
"Don't be ridiculed by people, saying that your cloud has not been surpassed by Huawei, you are only the third (you can't help it)." Ma Huateng said that in the past three years, CSIG has done a lot of impulse things, pursuing front-end integration and pursuing large digital revenue. He said that Tang Daosheng did not want to do this before, but Tencent is often caught up in market share and public opinion, and is forced to do general contracting to gain market share.
Alibaba Cloud proposed the strategy of "being integrated" earlier, in 2019. On April 26 this year, Daniel Zhang said at the partner conference that the core of making integration do what is said is to firmly move towards "product integration".
Source: Alibaba Cloud WeChat public account
The latest strategic development is that at the partner conference, Alibaba Cloud announced that Alibaba Cloud will be integrated into Alibaba Cloud after a new round of organizational changes, and DingTalk as a coordination and development platform will also enter the integration sequence. In addition, cloud-native products including cloud computing and serverless and large-model products, including Tongyi Qianwen, have also joined the integration sequence.
It is worth noting that the shift to integration not only means that they are more pursuing growth quality and actual profits, but also means that they are no longer purely competitive with operators.
At a previous forum, Kang Ning, president of the second division of China Telecom's Internet Industry Division, said in his speech that he is willing to gather various cloud head vendors when becoming an integrator, and is also willing to be integrated in the enterprise-level services of the head cloud vendors to jointly serve customers.
3. Volume to large models and increments to MaaS
Time to 2023, embracing the big model has become a new exploration for cloud vendors.
The popularity of OpenAI and its impact on Microsoft's Azure intelligent cloud provides a realistic path for cloud vendors to embrace the big model.
"No matter how ChatGPT makes money in the future, it is certain that Azure Intelligent Cloud will make a lot of money with AI big models." A cloud vendor NLP business expert told the media.
Objectively, the substantial return of Microsoft's extensive access to GPT business has not yet been highlighted, but more and more enterprises have begun to use OpenAI services to run large models - Microsoft's latest quarterly financial report shows that Azure OpenAI has 2,500 customers, a 10-fold increase from the previous quarter.
This is why, since the beginning of this year, Baidu, Ali and other large manufacturers have launched large models, while accelerating the pace of transforming existing businesses through large models, and are also exploring how large models can transform cloud computing.
Source: Northeast Securities
From the perspective of Ali and Baidu's layout on large models, their layout of "Tongyi Qianqian" and "Wen Xin Yiyan" is not simply a "tool" or "product", but an artificial intelligence base-type empowerment platform.
Because the transformation of the industry by the big model is not only reflected in the fact that a company chooses to access the big model such as "Tongyi Qianqian" or "Wen Xin Yiyan", and improves its products and services through the natural language understanding, expression ability, reasoning ability, etc. expressed by the large model.
More importantly, for different industries, enterprises can also have their own industry-specific large models through AI big models.
"Every enterprise can have an exclusive large model space on Alibaba Cloud, where it can not only call all the capabilities of Tongyi Qianwen, but also combine the enterprise's own industry knowledge and application scenarios to train its own enterprise big model." Zhou Jingren, CTO of Alibaba Cloud Intelligent Group, said at the recent Cloud Summit that ecological partners from all walks of life can combine their own industry knowledge and scenario needs to retrain and fine-tune the Qianwen model to create an exclusive large model and integrate it into their own overall industry solutions.
In other words, based on the accumulation of computing power, algorithms, and data reserves by cloud computing vendors over the past many years, the capabilities they can provide externally are divided into two layers:
One layer is the capabilities that come with a large model as a tool. For example, Tmall Genie will become smarter after accessing Tongyi Qianqian. Users can ask it to "synthesize a 1-hour playlist, 50% interspersed with rock-style songs" while running, and chat with Tmall Genie about history and life.
For example, the emergence of large model application products will also bring new changes to the industry.
On June 1, Alibaba Cloud announced the latest progress of the Tongyi Big Model, and the new AI product "Tongyi Listening Understanding" focusing on audio and video content was officially launched, becoming the first large model application product in China to open public testing, which accesses the understanding and summary capabilities of the Tongyi Qianwen Model, which can help users complete the transcription, retrieval, abstraction, and organization of audio and video content, such as automatically taking notes, organizing interviews, and extracting PPT with the large model.
According to Zhou Jingren, Tingwu provides both a personal version and an enterprise application, and has been widely used within the Alibaba Group.
The other layer is the platform capability as the underlying infrastructure in the AI era. For example, in theory, each enterprise can have its own intelligent customer service, intelligent shopping guide, intelligent voice assistant, copywriting assistant and other related models, in short, it is the ability to enable enterprises in the ecosystem to train their respective industry models based on large models at a lower cost.
These two capabilities ultimately point to the transformation of the cloud computing business model - MaaS (Model as a Service).
Among the intelligent computing MaaS services provided by Alibaba's "Feitian Intelligent Computing Infrastructure", there are Tongyi large model series and enterprise customized large models based on it, as well as the first "model-as-a-service" open platform Magic Tower Community in China. According to official introductions, this community has now gathered more than 800 high-quality open source models, with a total of more than 1 million users and a cumulative number of model downloads of more than 16 million.
Previously, when Baidu released Wen Xin's words, Robin Li made it clear that after the advent of generative large models and becoming mainstream, the business model of cloud computing will evolve in the direction of MaaS, that is, applications are built on large models, rather than based on cloud computing power or storage in the past.
In his words, in the future, customers' needs for manufacturers will focus more on intelligent services, depending on whether the framework is good, the model is good, and the synergy between the four layers of model, framework, chip, and application. He described it as a "decisive battle in the field of to B." ”
From this point of view, although the current layout of cloud vendors for large models is still in the initial stage, in the long run, large models are expected to bring them new increments. As cloud vendors deploy large models as a trend, cloud wars will gradually evolve into big model battles.