Technology cloud report original.
In the past year or so, the AI large model has gradually moved from the initial framework to the landing stage.
However, as AI large models have penetrated into thousands of industries, the market has begun to realize that although general large models are powerful, they do not seem to be able to fully meet the individual needs of different companies.
Comprehensive indicators such as security, interpretability, and ease of use of large-scale model technology are becoming the key to this round of AI competition. Various business models such as privatized deployment of large models.
AI and cloud computing usher in the era of great integration
Since its inception, cloud computing has been considered the future development direction and the "second growth curve" of Internet companies.
However, due to the peak of the Internet growth dividend and the impact of stricter policy regulation. Beginning in 2021, cloud service providers who were originally in a stage of rapid growth have fallen into a stage of slowing down. From a growth rate of more than 50% at the peak, it will drop all the way to 20% or even lower in 2022.
The decline in global industrial demand has led to an imbalance in the growth rate of the cloud computing industry. According to an early warning from research firm TrendForce in February this year, the server purchases of Meta, Microsoft, Google, and Amazon may slow down from an annual growth rate of 6.9% to 4.4%.
However, the turning point came soon. After the emergence of ChatGPT, even if the industry demand slows down, domestic and foreign cloud service providers are actively shifting their business focus to the AI field.
It can be seen that cloud service providers have gradually changed from "seeking speed" in the first half to landing in the depths of the industry to seek higher value increments.
With the completion of the first wave of cloud migration, enterprises are entering the stage of deep cloud use. The combination of AI and the cloud can not only further reduce the threshold for enterprises to go to the cloud, but also realize in-depth intelligence for enterprises.
According to Gartner's "2022 Artificial Intelligence Technology Maturity Curve" report, the early adoption of AI technologies such as composite artificial intelligence (AI) and decision intelligence will bring obvious competitive advantages to enterprises and organizations, and alleviate the problems caused by the vulnerability of AI models. Helps capture business context information and drive value realization.
In mid-April this year, Cai Yinghua, Chief Commercial Officer of Alibaba Cloud Intelligence, said that the rapid development of computing power has made digitalization a certainty and intelligentization possible.
In the future, Alibaba Cloud will take cloud computing as the cornerstone and AI as the engine to participate in the epoch-making transformation from digitization to intelligence.
AI is a big step towards Pratt & Whitney
Nvidia’s performance in the second fiscal quarter was astonishing. Behind it was the industry’s urgent need for high-performance computing power in the generative AI revolution. However, under the high cost of computing power, AI large models still seem to be a “rich man’s game” .
In the face of skyrocketing GPU prices and the possible "computing power shortage" in the future, large companies can still enter the AI arena leisurely by spending money and hoarding cards, while small and medium-sized enterprises do not have the "money ability". , more easily constrained by computing power.
What can be used to save the current situation of extremely unbalanced AI computing resources? How to let more entrepreneurs participate in the market competition of large models?
As the world's leading GPU supplier and the biggest beneficiary of this round of AI boom, Nvidia has given a way to break the situation-computing power leasing.
In March of this year, Nvidia officially launched the computing power leasing service solution "DGX Cloud". DGX AI supercomputer does not need to purchase and own server equipment.
In fact, Nvidia DGX Cloud is not the first case in the AIGC industry, but due to the outstanding performance of the DGX AI supercomputer, DGX Cloud has pushed the AIGC cloud computing power industry to a higher starting point. The launch of this service marks that AI cloud computing power has entered a new stage . Cloud computing power adopts the method of "dividing the whole into parts" to empower all parties in the industry chain, which is sustainable.
Computing power leasing, that is, renting out computing power, is a mode of renting computing resources through cloud computing service providers. Computing power manufacturers have cooperated with cloud platforms for a long time. Users can rent NVIDIA graphics cards and AI processors through Alibaba Cloud, Tencent Cloud and other platforms.
For Nvidia and cloud service providers, computing power leasing is a win-win strategy.
Oracle, the world's seventh largest cloud service provider, was the first to respond to Nvidia's DGX cloud plan. The company migrated Nvidia's accelerated computing stack tools (including GPUs, systems, and software) to its flagship IaaS business in October 2022— —On the cloud service platform OCI (Oracle Cloud infrastructure).
Judging from the latest quarterly report data, this has a very significant boost to Oracle's performance.
In the fourth fiscal quarter of 2023 (March 1-May 31 of the natural year), its cloud business (IaaS+SaaS) revenue reached US$4.4 billion, a year-on-year increase of 54%; among them, IaaS business revenue was US$1.4 billion, A year-on-year increase of 76%, the cloud business revenue growth rate ranked first among cloud vendors in a single quarter.
For Nvidia, this is also a business with a short payback period and a considerable gross profit margin.
Taking the A100 (80G) rental service as an example, the unit price of the A100 (80G) graphics card is 100,000 yuan. Now assuming that each card is fully rented, the A100 (80G) will be rented by the domestic cloud computing platform on August 19, 2023. The average price of the server is 15.1 yuan/hour. Considering that major platforms compete for customers and often launch preferential activities, assuming that the average actual rent is 7.6 yuan/hour, the actual payback cycle of investing 1 billion yuan is 1.5-2 years, according to Based on the lowest pricing calculation on the platform, the gross profit margin is at least 46.3%.
At present, Nvidia is actively expanding its "friend circle" and co-hosting DGX cloud infrastructure with leading cloud service providers. In addition to Oracle, Microsoft Azure has also started hosting DGX cloud. This service will also be extended to Google Cloud in the near future.
Based on computing power leasing, users only need to pay on demand, and do not need to bear the cost of hardware equipment purchase, maintenance, upgrade, etc., and do not have to worry about waste caused by idle or outdated equipment; users can access the required computing power resources through the cloud anytime, anywhere, and start quickly Training and application; users can choose different computing power platforms and models according to their needs, and can also conduct more experiments and explorations without being limited by regions or time, models, tools and other resources.
When multi-cloud meets generative AI
In order to provide highly stable and cost-effective AI infrastructure to large-scale customers, the model of generative AI + multi-cloud has become a new point of contention for technology manufacturers. The cloud is regarded as the bearer of AI, and AI is also the core of the cloud. The algorithms, computing power, and data capabilities required to develop large models, as well as solutions covering IaaS, PaaS, and MaaS.
Recently, VMware launched Intelligent Assist and Private AI architecture solutions. Prior to this, Alibaba Cloud proposed the concept of "Model as a service", and Amazon Cloud Technology launched new generative AI tools including Amazon Bedrock and Amazon Titan models.
Raghu, CEO of VMware, said: "Generative AI is a perfect match for multi-cloud. Customers' data is everywhere, in their data center, edge, cloud and more.
Together with NVIDIA, we will help enterprises confidently run generative AI workloads near the data and address their concerns about enterprise data privacy, security and control. "
Jen-Hsun Huang, founder and CEO of NVIDIA, said: "We can train AI models, fine-tune AI models, and in order to deploy AI models and large language models across multiple GPUs, especially large language models, one computer cannot run and must be distributed. To multi-machine multi-card, and reasoning on it, generating tokens, and realizing interaction, its speed is comparable to that of human daily interaction."
At the same time, he said that by expanding our cooperation with VMware, we will be able to provide thousands of customers in financial services, medical care, manufacturing and other fields with the full-stack software and computing they need, enabling them to use custom-made solutions based on their own data. applications to fully exploit the potential of generative AI.
Private AI consists of a set of integrated AI tools that enable enterprises to customize models and run various generative AI applications, such as intelligent chatbots, assistants, search and summarization, etc.
The platform will be built as a fully integrated solution using generative AI software and accelerated computing from NVIDIA, built on VMware Cloud Foundation and optimized for AI.
“Initially, AI was built and designed by some data scientists for the convenience of other data scientists,” said Chris Wolf, vice president of VMware AI Labs.
With the launch of the new VMware Private AI offering, VMware is bringing compute and AI model selection closer to the data so that the future of AI works for everyone in the enterprise. "
Can AI drive the growth of cloud services?
According to the 2022 global cloud computing tracking data released by the industry research organization IDC, the global cloud computing IaaS market has grown to 115.496 billion US dollars, a year-on-year increase of 26.2%.
The top three cloud vendors in the world are Amazon, Microsoft, and Alibaba Cloud, accounting for 48.9%, 14.4%, and 6.2% respectively, followed by Google and IBM, accounting for 5.6% and 2.9% respectively. Huawei Cloud, China Telecom, Tencent Cloud, China Mobile and Baidu Cloud rank six to ten.
Compared with 2021, Alibaba Cloud's market share is decreasing, while the market share of Amazon and Google is increasing.
After the development of the mobile Internet peaked, the revenue growth of global cloud service providers has slowed down. The growth rate of AWS has dropped from 40% to 12%, and the growth rate of Microsoft Azure has dropped from 31% to 15%. However, it is clear that Alibaba Cloud is facing greater challenges For some, the growth rate has dropped to less than 10%, the first negative growth in the last quarter.
In contrast, Google Cloud, which is closely behind, still maintains a growth rate of nearly 30%, and is very likely to replace Alibaba Cloud to occupy the third position among global cloud vendors.
In the second quarter of 2023, Alibaba Cloud's revenue increased by 4% year-on-year. The financial report shows that Alibaba Cloud’s revenue growth in this quarter was mainly driven by storage, network and AI computing-related products, and part of the increase was offset by the normalization of CDN demand.
From the perspective of customer distribution, revenue growth was mainly driven by the financial services, education, electricity and automotive industries, partially offset by initiatives to actively reduce project-based revenue.
The demand for computing power and model services brought about by the AI boom is driving cloud computing giants back to growth, so can this growth continue in the long run?
According to IDC's statistics on the revenue split of the world's major cloud computing vendors, the revenue structures of Alibaba Cloud and AWS are basically similar. Both are mainly based on IaaS business, supplemented by some PaaS business, while Microsoft Azure's PaaS and SaaS revenue accounted for more than 60%. %.
In fact, in the cloud computing industry chain, from the bottom layer IaaS, to the middle layer PaaS, and then to the upper layer SaaS, the higher the product differentiation is, the higher the gross profit margin will be.
Microsoft Azure just connects its own Windows, Office, SQL Server and other software with cloud services to obtain higher gross profit margins.
IDC's "Global Public Cloud Service Semi-annual Tracking Report" released on July 6 shows that the total revenue of the global public cloud service market in 2022 will be 545.8 billion US dollars, of which SaaS (software as a service) is the most important revenue of public cloud services. sources, accounting for more than 45% of total revenue in 2022.
In the face of fierce IaaS competition in the global market and the domestic market, Alibaba Cloud is focusing on PaaS and MaaS, and Microsoft Aure can be used as a reference for comparison.
As an investor in OpenAI, Microsoft obviously benefits more from the development of generative AI. Earlier this year, Microsoft announced the release of the Azure OpenAI service on the global Azure platform. The service aims to provide developers with easy access to large language models that can be seamlessly integrated across other Azure products to assist enterprises in developing and deploying conversational AI. services and solutions.
In addition, Microsoft announced Microsoft 365 Copilot pricing at its Global Partner Conference, launched the Bing Chat Enterprise AI chatbot, and jointly announced with Meta that the Llama 2 open source model will be introduced into the Azure cloud and Windows.
In Microsoft's 2023 fourth-quarter conference call, Microsoft Chief Financial Officer Amy Hood said that although the current demand for Azure AI services is strong, the current contribution of AI services to Azure's revenue is only about 1 percentage point. Investing in cloud infrastructure, the impact of AI on Microsoft's revenue will be concentrated in the second half of fiscal 2024.
Zhang Yong also said at the financial report conference: "The artificial intelligence AI revolution is an incremental opportunity, all walks of life, all companies will hope to use artificial intelligence to improve their services.
But this is inseparable from the use of a large amount of high-performance computing power, not only for the training of the current model, but also for supporting them to provide various services in the future. So we think this is a very important, long-term growth engine. "
Enterprise digitalization and industrial intelligence are a long way to go, but cloud computing giants are helping more companies to step into the wave of industrial intelligence upgrade through continuous technological innovation and open empowerment, greatly shortening the The time required for this upgrade path.
Under the reality that artificial intelligence technology is integrated into all things, technology, demand and industrial evolution will never stop and rush forward, let us wait and see how far the future technical framework will progress.
[About Science and Technology Cloud Report]
Focus on original enterprise-level content experts - technology cloud reports. Founded in 2015, it is the top 10 media in the cutting-edge enterprise IT field. Recognized by the Ministry of Industry and Information Technology, Trusted Cloud, one of the official media designated by the Global Cloud Computing Conference. In-depth original reports on cloud computing, big data, artificial intelligence, blockchain and other fields.