Author of this article: Zhang Yifan

Source: Hard AI

Strong three-digit year-on-year increase in AI revenue?!

At the May 14 earnings conference, Ali revealed that in 24Q1, Ali's AI-related products achieved strong three-digit revenue growth over the same period last year.

Group CEO Wu Yongming said that the growth of the AI business mainly comes from two aspects:

• Model training and inference service requirements of basic model companies and internet companies;

• Driven by growing demand for vertical applications in finance, automotive and other industries;

Regarding future expectations, management believes that under the company's strong AI layout, Alibaba Cloud's commercial revenue will resume double-digit growth in the second half of the fiscal year.

At the same time, judging from changes in financial report cash flow, Alibaba Cloud was still the company's key capital investment direction in the first quarter.

Why can Alibaba Cloud's AI business achieve such strong growth, and how will Alibaba Cloud's AI strategy be carried out? This article attempts to find out one or two through the company's business changes over the past six months.

I. Ali's AI layout

In September of last year, after becoming the new CEO of Alibaba Group, Wu Yongming proposed two major strategic priorities: “user-first and AI-driven.” At the November conference call, Wu Yongming also officially identified three directions: technology-driven internet platform business, AI-driven technology business, and a global business network.

For Alibaba Cloud, after 2 months of separation, Wu Yongming determined that Alibaba Cloud will implement an AI-driven, public cloud-first strategy for the next five years.

To this end, the company later had a more comprehensive layout and a clearer strategy in the field of AI. Everything is in line with the growth of public cloud services. Whether it's the original open source model base, Tongyi series of AI applications, model communities, or further expanding foreign investment and expanding Ali's “AI ecosystem of friends,” they are all leading the way to Alibaba Cloud's core public cloud services.

Hard AI Statistics Currently, Alibaba Cloud's main AI product ecosystem:

II. AI Cloud Service

There is no doubt that Ali's public cloud has become the core of the cloud sector, and all of its businesses are being laid out around this topic. Focusing on establishing competitive advantages in technology and scale, and reducing the prices of global public cloud products, it has become an advantage of Alibaba Cloud.

Further split, the advantages of AI cloud services are reflected in elements such as computing power, communication, and model libraries.

1) Computing power/network

As a leader in cloud services, Alibaba Cloud has an innate advantage in computing power and network resources. Relying on its own cloud platform, Ali has built sufficient AI infrastructure (including computing power and communication facilities) and launched the cloud development platform ModelScope, which enables developers to easily develop and deploy various AI models in the cloud.

At the beginning of April this year, Ali Chairman Cai Chongxin pointed out in an interview that the current computing power reserves are sufficient to support 18 months of big language model training requirements. Also, in the “inference” stage, the company also has corresponding deployments to support the operation of the model.

This was confirmed by the company's cash flow consumption this quarter.

2) Communication networks

The HPN7.0 network architecture introduced by Ali is currently the highest level of Ethernet AI clusters. It uses Ethernet RDMA low latency technology to support cluster scalability up to 100,000 cards. Fast response and low latency.

On May 14, Alibaba Cloud's AI high-performance network architecture HPN 7.0 results paper was included in SIGCOMM2024, making it the first paper in SIGCOMM's history on AI intelligent computing cluster network architecture. SIGCOMM is the world's most authoritative computer communication network summit. Previously, Google Jupiter Network developed into a classic architecture after being selected by Sigcomm. Some experts pointed out that Alibaba Cloud HPN7.0 is expected to become a new paradigm for next-generation AI high-performance network architectures.

Recently, the Tongyi Qianwen 2.5 large model released by Alibaba Cloud was implemented based on a high-performance network cluster built on HPN 7.0.

3) AI cloud development platform

Alibaba launched ModelScope, a cloud development platform, which provides developers with an extensive library of models, so that developers can easily develop and deploy various AI models in the cloud.

3. Model base

While providing cloud computing services to large model customers, Alibaba Cloud is also developing its own big open source model, but this is not to seize the customer market, but to help the industry grow faster through Alibaba Cloud's open source technology.

Up to now, Ali's generalized model has been iterated to version 2.5. Recently, the latest evaluation results released by an authority show that the performance of Tongyi Qianwen 2.5 launched by Alibaba Cloud in Chinese is on par with GPT-4 Turbo.

The release of the Tongyi model is, on the one hand, to establish the company's AI ecosystem (using an open source form) to help the industry develop faster, expand the market, and increase demand for AI cloud computing power; on the other hand, to promote the development of AI applications. Based on this big model, developers can quickly build AI applications suitable for different scenarios, such as intelligent customer service, image recognition, natural language processing, etc., to increase the growth of users within the group and the C-side, thereby increasing the demand for cloud computing power services.

The rapid development of both can quickly drive the growth of Alibaba Cloud's business.

IV. Model application

Considering the company's AI-driven strategy, a model suitable for each different scenario/business is needed to support it.

Based on the big model, based on multiple modes, Alibaba Cloud has launched a series of vertical industry applications.

According to the official website, the Tongyi series has now launched the conversational AI application Tongyi Thousand Questions, the speech-to-text AI application Common Sense and Understanding, and the AI application for image generation, Tongyi Wanxiang.

In addition, there are vertical applications for the special needs of different business scenarios, including Tongyi Farui (law), Tongyi Spirit Code (programming), and Tongyi Xiaomi (customer service).

More businesses are driven by AI, which also means creating more incremental demand for cloud computing.

At last year's results conference, Ali revealed that there is huge room for improvement in the company's search and advertising business through AI functions. However, the company also mentioned that the feature is currently still in the early stages of research testing. However, they believe that with in-depth research and investment, this feature will be a very positive help to Taotian's future search and advertising.

Additionally, for customers who want to personalize AI applications, Ali has launched the Bailianda Model Platform. Developers can “drag and drop” to develop a large model application in 5 minutes, and “refine” an exclusive model in a few hours.

5. Foreign investment to establish Alibaba's AI ecosystem

After Wu Yongming proposed a strategy to prioritize public clouds, Ali's investment in AI models accelerated markedly. Currently, the big domestic model Five Dragons have received investment from Ali without exception.

From investing in Smart Spectrum AI in September 2023 to March of this year, Ali completed a new round of investment in MiniMax.

At this point, Alibaba has completed investments in China's five leading model unicorns (Dark Side of the Moon, MiniMax, Smart Spectrum, Baichuan, and 10,000 objects). These leading model manufacturers have gradually increased their influence in the B-side and C-end markets, and have also provided huge increases in Alibaba's AI cloud services, becoming an important source of three-digit growth for the AI business.

From a strategic perspective of the company's AI-driven, public cloud-first strategy, this investment is extremely important today.

Large model makers need to rely on stable and cost-effective AI infrastructure to train and reason models. As a leading domestic AI cloud service provider, Ali can provide this kind of service with high quality.

This was also confirmed in this quarter's earnings report. As a result of establishing a competitive advantage in technology and scale, Ali lowered the prices of global public cloud products this quarter. And this has further boosted the development of the AI industry.

At the same time, Alibaba seizes opportunities through investment and services, enabling big model companies to deploy their business on Alibaba Cloud. As large-scale model users grow rapidly in the later stages, demand for AI cloud computing also grows, thus driving the growth of Alibaba Cloud's core business.

Ali's investment in AI big model companies becomes capital expenses of AI big model companies, and then flows back to Alibaba Cloud, forming a closed loop of business.

There are similar examples of this investment method among overseas giants, and Amazon's investment in Anthropic is also based on a similar logic. Furthermore, using cloud computing power as capital to invest in large model vendors has also become a way to further strengthen the contacts between cloud computing vendors and big model vendors in their business, while also reducing pressure on the cash flow of both parties.

6. What about the future of Alibaba Cloud?

At present, Alibaba Cloud's “AI-driven, public cloud-first strategy” has begun to bear fruit, and Alibaba Cloud has chosen a path of focus. However, the most important future growth depends on the major development of the generative AI industry, and the certainty of this is constantly growing.

As expected by Ali in its earnings report, “Alibaba Cloud will return to double-digit growth in the second half of the year.”