Generative AI: A Critical Battle for Cloud Computing

"There's a railroad company in the U.S. and we've been talking for the last two years. Finally, they called us this time and wanted to move quickly to the cloud. Recently, Uwem Ukpong, vice president of global customer technical support and services at Amazon Web Services, talked to reporters about the latest needs of enterprises to go to the cloud.

One background is that generative AI is sweeping thousands of industries, and to get the ticket to generative AI, the cloud is usually the first choice. "First, enterprises need to fully embrace cloud technology, and moving to the cloud is the first step, second, enterprises need to ensure that their data is in a unified data lakehouse so that they can efficiently use computing power to harness data, and finally, they need to identify potential use cases for generative AI technology and develop a proof of concept (POC)," said Uwem Ukpong. ”

He also told the 21st Century Business Herald reporter that some enterprises are very slow to adopt cloud technology, but now they are facing pressure from different stakeholders and shareholders to make a difference in generative AI, and enterprises to start the generative AI journey means that they need to go to the cloud.

As a result, generative AI is also becoming a new driver in the cloud computing market. Generative AI requires cloud computing, and cloud computing is also reshaping the service format because of generative AI, and as large models enter the cloud, model-as-a-service (MaaS) has attracted much attention.

In this regard, Dai Kun, vice president and chief analyst of Forrester, said in an interview with the 21st Century Business Herald: "The overall competition of generative AI will further heat up, and accelerate market integration in the process of commercial landing from hype to value." Model-as-a-Service itself will not bring about large-scale revenue growth and changes in service methods, but it will effectively promote enterprises' investment in underlying infrastructure, such as computing, storage, networking, databases, data lakehouses, etc., and inject new ideas into customer interaction, internal operations, and ecosystem construction for vendors and enterprises in various industries. ”

"Three-tier architecture" and full-scale competition

According to Forrester analysis, the top four global cloud computing vendors in terms of market share in the public cloud infrastructure sector in 2022 are Amazon Web Services (AWS), Microsoft Azure, Google Cloud, and Alibaba Cloud.

Among them, Amazon Web Services (AWS) ranks first, leading the way in cloud infrastructure. Amazon Web Services also continued to contribute high profits, with an operating profit of $24.631 billion and Amazon's operating profit of $36.852 billion in 2023.

And it's worth noting that with the rise of generative AI, the seemingly unknown Amazon has been attacking frequently since 2023. After investing $4 billion in OpenAI's rival Anthropic last year, it has recently continued to invest an additional $2.75 billion in Anthropic, which shows its determination.

As for the selection of large models, Uwem Ukpong said that Amazon Web Services has cooperated with a number of large models, including AI21 Labs, Anthropic, Cohere, Meta, Stability AI, Mistral AI, etc., as well as its own Amazon Titan base model.

As a global cloud vendor, Amazon Web Services proposes a three-tier service architecture based on generative AI. Uwem Ukpong told reporters that the first layer is generative AI cloud infrastructure, and in the past five years, Amazon Web Services has continued to increase its investment in self-developed chips, such as Amazon Trainium and Amazon Inferentia chips, which can provide the lowest cost of training models and running inference on the cloud. In addition, in terms of GPUs, a number of NVIDIA's advanced computing power have taken the lead in landing on Amazon Web Services.

The second layer is the model, which is based on the Amazon Bedrock cloud managed service, which allows enterprises to access a series of powerful basic models from text to images through APIs, and the third layer is the generative AI-driven application, which has been launched Amazon Q, an AI assistant that can be customized according to the customer's business and dedicated to the needs of office scenarios.

It can be seen that Amazon Cloud is expanding from the underlying infrastructure to the upper-layer applications, and Uwem Ukpong also pointed out that the layout of Amazon Web Services in 2024 will focus on Amazon Q and Amazon Bedrock. Both are closely related to generative AI and are an important fulcrum for Amazon Web Services to develop AI.

At the same time, Amazon Web Services is also facing fierce competition, and Microsoft Cloud, which ranks second, is struggling to catch up with generative AI. Microsoft bet on OpenAI early, following its $13 billion investment in OpenAI, followed by an investment in Mistral AI. It can be seen that leading manufacturers are also constantly investing in AI large model companies to increase their chips. It is reported that Azure expanded its support for OpenAI's latest models in the fourth quarter of 2023, including GPT-4 Turbo, GPT-4 with Vision, and Dall-E 3, and provided fine-tuning capabilities, while AI also drove Azure's performance growth.

Not to be outdone, Google Cloud, one of the top three, has rapidly iterated after launching the AI model Gemini, and with the help of generative AI, Google Cloud has begun to show new growth momentum in the fourth quarter of 2023. For the cloud market, a full-scale race for generative AI has begun.

A critical battle for generative AI

The current cloud computing manufacturers are facing fierce competition, and the white-hot degree can also be seen from this year's fancy price reduction information. For a long time, the differentiated competition of cloud vendors has intensified, and in the past, everyone mainly fought for infrastructure or made single-point breakthroughs.

But at present, the era of "one trick is eaten all over the world" has passed. For large manufacturers, if they only do cloud, or only make large models, and only make chips, there will be shortcomings. In fact, the global giants have already laid out in these fields, and the competition for clouds and large models is needless to say, and the race for chips is more fierce. For example, Google's TPU has been developed for a long time, and has launched the fifth-generation chip TPUv5e for large model training and inference, and Microsoft has also released its first self-developed AI chip, Azure Maia 100 AI chip and Cobalt 100 CPU.

The consensus is that at least get involved in order to have more possibilities, or that enterprises need to understand this field in order to better carry out systematic AI business. Whether it is to expand large models or chips, it is not necessary to make it a main business, but in the long run, these product lines cannot be without, especially for large manufacturers, these elements need to be equipped as standard.

In the process of expanding from cloud infrastructure to chips and large models, a new battlefield has opened, and cloud vendors have their own paths on the journey of generative AI. Dai Kun analyzed to reporters: "AWS, Azure and GCP have different generative AI development strategies in many areas such as AI hardware infrastructure investment, AI software infrastructure service planning, AI basic model development, ModelOps service orchestration, application development empowerment, solution focus, model ecological construction, associated development and application service collaboration, and business investment methods. ”

Uwem Ukpong told reporters: "One of the biggest differentiators of Amazon Web Services is that it has a wide selection of models. We're putting more LLMs on Amazon Bedrock to give customers plenty of freedom of choice, including the latest Claude 3 model currently released by Anthropic. ”

He also added that on the one hand, Amazon Web Services has the technology of self-developed chips, and Anthropic chose Amazon Web Services as its main cloud service provider, and used Amazon Trainium and Amazon Inferentia chips to train and deploy future base models. So you can get a lot of first-hand feedback directly from Anthropic for the improvement of chip technology. On the other hand, Amazon Web Services has both external partners and Amazon Titan basic models in the large model, which has a wide range of applications.

While overseas manufacturers are accelerating the promotion of generative AI, domestic cloud vendors such as Alibaba Cloud, Huawei Cloud, Tencent Cloud, and Baidu Intelligent Cloud are also rapidly seizing the AI dividend. Previously, Alibaba Cloud released an industry model based on Tongyi Qianwen, HUAWEI CLOUD developed its own Pangu model series, Tencent launched a hybrid model, and Baidu Intelligent Cloud launched 7 large model application products based on the Qianfan model platform, including Baidu Intelligent Cloud Xiling Digital Human Platform, Baidu Intelligent Cloud Keyue Intelligent Customer Service Platform, Baidu Comate Code Assistant, etc., all of which are continuously iterating and applying.

Talking about Chinese manufacturers, Dai Kun said: "Although China's cloud vendors are constrained by many factors such as the availability of high-performance GPU chips, the availability of high-quality data, the lack of awareness of community ecology and innovation collaboration, the cognitive barriers of innovation long-termism, the shortage of high-tech talents in production and research, and the adverse impact of the overall macroeconomy, the core performance of generative AI models, AI governance, AI-oriented data management, In general, there are gaps in many aspects such as service collaborative orchestration, but some leading enterprises have kept pace with the international market in terms of the application of industry models for specific business scenarios, including customer service, digital human, and modern knowledge management. ”

For more information, please download the 21 Finance APP