AIGC is the game changer of cloud computing, which will fundamentally change the rules of the game in cloud computing and even the entire technology industry. As the Game Rulemaker of the cloud computing industry, Amazon Cloud Technology has also shown new achievements. On April 13, Amazon Cloud Technology announced the launch of new generative AI tools, including Amazon Bedrock and Amazon Titan models, to cover the cost-effective Amazon EC2 Trn1n and Amazon EC2 Inf2 instances, and the AI programming assistant Amazon CodeWhisperer was released at the same time.
AI is the fruit on the high branch of the cloud industry
There is a term in economics called "low-hanging fruit". On a fruit tree, there are always some fruits in the lower position, and people under the tree can reach them. It is very suitable for the field of cloud computing. The past ten years have seen the rapid development of the cloud computing industry. After Amazon Cloud Technology clarifies the technology and business model of cloud computing, cloud vendors only need to tiptoe to pick the low-hanging and full fruits. However, this Such growth cannot continue indefinitely.
From the United States, where cloud computing is most widely used, to other parts of the world where digitalization is relatively lagging behind, from the Internet industry that first adopted cloud computing, to large-scale government and enterprise departments that migrated and transformed to the cloud, cloud vendors have always been able to pick and choose from the perspective of overall income. to the low-hanging fruit until the low-hanging fruit is gradually plucked.
In the latest quarterly financial reports of cloud vendors, everyone inevitably encountered a decline in growth rate. Although there are factors such as the impact of the macro economy and the recovery of corporate IT spending in the post-epidemic era that is not as expected, combined with the growth rate changes over a period of time, the cloud Vendors' growth continues to slow down, and hit a new low in growth rate, which is already a common problem faced by the cloud computing industry.
In a cloud environment, AI software based on machine learning (ML) algorithms can provide customers and users with an intuitive and connected experience, but Amazon's investment in cloud technology goes back to its previous footprints. Examples include the computer vision technology in Prime Air (Amazon drones) and Amazon Go, the billions of answers Alexa has with customers every week.
Clouds and mockups, wide and deep moats
It is recognized in the industry that large-scale models are the game of giants, and there are definitely not too many manufacturers who have enough funds and resources to build large-scale models by themselves. From another dimension, the big model is not a one-man show of the giants. The development of cloud and artificial intelligence is highly dependent on the construction of industrial ecology.
The algorithms, computing power, and data capabilities required to develop large models, as well as solutions covering IaaS, PaaS, and MaaS, all take into account the technical reserves of cloud vendors themselves. In terms of the new generative AI tools launched by Amazon Cloud Technology this time, the most impressive one is Amazon Bedrock. Users can access the basic models from AI21 Labs, Anthropic, Stability AI and Amazon Cloud Technology through the API, from text to Images are available, including Amazon Titan's two base models (Amazon Titan Text and Amazon Titan Embeddings).
This solves the binding problem that some industry customers worry about. Amazon cloud technology has assembled more large models, and customers can freely choose Amazon's basic model or other companies' large models, which can not only achieve targeted tuning, but also provide more Choose, so that you don't have to worry about being bound by a large model or cloud vendor.
One of the most prominent features of Bedrock is that the threshold is low enough. Customers only need to show Bedrock a few labeled data examples in Amazon S3, and Bedrock can fine-tune the model for specific tasks. With at least 20 examples, you can Train a customer's own scenario-based product.
The Trainium and Inferentia chips self-developed by Amazon Cloud Technology constitute a crushing advantage over other manufacturers. According to the data released by Amazon Cloud Technology, the Trn1 computing instance supported by Trainium can save up to 50% of the training cost, supports the interconnection of 30,000 Trainium chips in the same availability zone, which is equivalent to more than 6 exaflops of computing power, and has a PB-level network.
The Inf2 instance powered by Amazon Inferentia2 is also officially available to the public. Compared with the previous generation, the Inf2 instance throughput has increased by 4 times, and the latency has been reduced by 10 times. It can also achieve ultra-high-speed connections between accelerators to support large-scale distributed reasoning. Compared with comparable Amazon EC2 instances, inference price performance has been improved by 40%.
AI Cloud, Crossing the Singularity Moment
Before artificial intelligence changed all walks of life, it changed cloud computing first. Cloud computing and artificial intelligence have become a strong binding relationship. The future unlimited artificial intelligence scenarios mean the same scale of cloud computing consumption. There is a view that cloud computing has almost completed the domination of technology, and cloud solutions have become common. Compared with other technical solutions, the cost effect of cloud is prominent, and there are more optional capabilities, and no company will refuse to lower the cost And choose more solutions, artificial intelligence will naturally grow on the cloud.
The demand of cloud customers for large models is enthusiastic and realistic. Amazon Cloud Technology integrates the needs of customers for large models on the cloud. First, they need to be able to directly find and access high-performance basic models. These models need to be able to provide the best matching business models. Excellent feedback results for the scene. Second, customers want to seamlessly integrate with applications without having to manage large clusters of infrastructure and without adding prohibitive costs.
Ultimately, customers want to be able to start easily and build differentiated applications with their own data (more or less) based on the base model. Since the data customized by customers is very valuable IP, it is necessary to ensure data security and privacy protection during processing. At the same time, customers want control over data sharing and use.
Based on these needs, Amazon Cloud Technology has a series of generative AI tools. It can be clearly seen that Amazon Cloud Technology has two current directions. One is to lower the price, and the current training and future reasoning are at low cost. Only when customers can accept it will there be room for large-scale application of large-scale models; second, the value is sufficiently prominent, and the value of artificial intelligence will be recognized only when customers improve quality and efficiency in their own business scenarios and use large-scale models with a simple and low threshold. Relevant data show that when estimating the technical impact of AI, just by integrating AI into online work, it can generate a value of 3.5 trillion to 5.8 trillion US dollars per year in 19 countries.
Through the artificial intelligence and machine learning services of Amazon Cloud Technology, enterprises and partners can easily build and deploy industry models, allowing businesses to make a leap towards artificial intelligence.