2024/05/17 REPORTED BY CISSY ZHOU
China's cloud service providers are racing against the clock to turn the AI boom into real revenue as they struggle to grow in an already crowded market.
Over the past few years, businesses such as Alibaba Cloud have struggled with stagnant growth, while cloud service providers in the U.S. have seen a boom: Microsoft's revenue from Azure and other cloud services grew by 31% in the first quarter, Google's cloud division grew by 28%, and Amazon's cloud division grew by 17%, all thanks to strong AI demand.
In China, the boom in AI models is expected to spur a significant increase in demand for AI computing, big data, storage, databases, and related services. For example, according to official data, China currently has more than 100 self-developed large language models (LLMs) or the underlying algorithms of generative AI, each with more than 1 billion parameters. More parameters usually mean better performance, but it also requires more computing power.
But the most recent quarter's earnings report did not paint the expected good picture.
In the first three months of this year, Alibaba Cloud's revenue grew by just 3% year-on-year. Tencent did not disclose its cloud revenue, saying only that it had "increased" in the quarter.
Baidu lags behind Alibaba in both revenue and market share in the cloud computing market, but it is aggressively establishing itself as a leader in Chinese artificial intelligence and hopes to close the gap with its ErnieLLM. Unlike most of its competitors that are only geared towards businesses, Baidu is pushing its ErnieLLM to both businesses and individual users. While the company's overall revenue grew by only 1%, its AI cloud service revenue grew by 12% in the first three months of the year, largely due to its model training fees on the public cloud.
Baidu founder Robin Li said on Thursday that while most of the revenue from its AI cloud service currently comes from model training, revenue from model inference, which is the calculation of output by artificial intelligence using training results, is also growing rapidly. He added that another growth driver for cloud revenue is the cross-selling of CPU cloud services to GPU cloud customers.
However, experts believe that overall, the cloud computing industry is still in the early stages of reaping the benefits of AI.
Kai Wang, a senior equity analyst at Morningstar, an investment research firm, said: "We don't think the contribution of AI to cloud service providers is yet very large. "It's too early to monetize, and it's going to come mainly from the corporate side. Use cases at the consumer level are currently quite limited.
Cloud services are very sensitive to macroeconomic conditions, and due to their high cost, many companies still don't see the need to integrate AI into their operations. However, if there is a large-scale digital transformation of business, companies such as Baidu may capture more market share, Wang added.
Still, China's cloud service providers believe that the AI-driven demand wave is in its early stages and will see significant growth.
Alibaba Group CEO Eddie Wu said on Tuesday that the company's public cloud product revenue had achieved "double-digit growth," while AI-related revenue grew "triple-digit" year-over-year. However, this growth was offset by the continued impact, although it is waning, from the company's previous focus on low-margin, project-based businesses. The company said it expects faster revenue growth, primarily driven by public cloud services and artificial intelligence products. Public cloud services are available over the internet and can be purchased and used by anyone.
Ng added that the demand for cloud services comes from foundational model companies and internet companies, as well as customers in the financial services and automotive industries.
Alibaba, which has been undergoing management reforms, remains ahead of China's public cloud market in terms of revenue and market share, although the latter has been steadily declining. In an effort to recoup its losses, Alibaba has placed greater emphasis on public cloud services and slashed prices for more than 100 products by up to 55 percent in February, followed by rival JD.com. In April last year, the e-commerce giant launched a similar price war.
Alibaba has also invested in China's top five AI unicorns: Zhipu AI, 01.AI, Baichuan AI, Mini Max and Moonshot AI. These companies train large-scale models on Alibaba Cloud and provide services to the public.
Unlike Baidu and Tencent, Alibaba has open-sourced several of its synonymous Qianwen large language models, and developers who want to use the free model will choose Alibaba Cloud when deploying their applications in the future.
Alibaba said the growing demand for AI will also have some "halo effects."
"Starting in 2024, we are seeing a rapid increase in customer demand for AI. This has also spurred the growth and demand for traditional cloud computing, including general-purpose computing, storage, and big data. "The more customers invest in and use AI, the more they will demand our various other cloud products, and the two things go hand in hand." ”
At the same time, Baidu is also actively investing in generative AI, hoping to take the opportunity to close the gap with its cloud computing rivals. The company is exploring the use of generative AI applications with a group of software vendors and has also established an AI app store.
At the other end of the spectrum, Tencent slashed its cloud computing business last year and began to shift its strategy from grabbing market share at all costs to reducing losses.
A late participant in China's LLM carnival, ByteDance on Wednesday unveiled a batch of large language models for businesses called "Doubao," which shares the same name as its previously launched chatbot. It follows the industry model and deploys its self-developed AI model on its own cloud service "Volcano Engine".
The announcement was unusually high-profile for ByteDance and added fuel to the fire of China's cloud computing price war. According to ByteDance, the price per 1,000 "tokens" of the Doubao Pro model is as low as 0.0008 yuan (0.011 cents), while Baidu's Ernie and Alibaba's 100 billion Tongwen are priced at 1.12 yuan per 1,000 "tokens." Tokens are words or snippets of words used in natural language processing. One token is about four characters, or 0.75 English words.
While convincing customers to switch cloud service providers is always a challenge, ByteDance believes the growing demand for AI is creating new opportunities to do so, an employee familiar with the situation said.
According to Canalys, China's cloud services market will grow by 16% in 2023, up from 10% in 2022. The consultancy expects revenue from China's public and private cloud infrastructure and platform services to grow 18% this year, as cloud consumption is expected to be driven by generative AI.
Zhang Yi, an analyst at Canalys, said: "Under the competitive pressure of price cuts, cloud computing vendors are currently facing the dual challenges of maintaining revenue growth and maintaining profitability."
"At the same time, generative AI continues to evolve at a rapid pace, placing increasing demands on cloud vendors. In addition to concerns about profitability, they must also invest more in emerging technologies. In this case, effective cooperation with the partner helps to reduce the risk and gain access to the partner's expertise and resources." "This will enable them to overcome competitive pressures and promote sustainable growth in an AI-driven future," she said.