- 2021-01-05
Amazon develops AI chips to reinforce the moat
November 26-30, 2018 is the day when Amazon Cloud Services held the re:Invent conference (AWS re:Invent 2018) in Las Vegas. In order to maintain the leading position in the Internet computing market as always, AWS has intensively released multiple series of new products.
On the 27th, AWS released a server processor (CPU) based on the ARM processing core and named it Graviton, and said that the Graviton chip will function through the new version of the EC2 cloud computing service. The main attraction of this product is that it can provide lower-cost computing services. With this customized CPU, Amazon has become the first company to launch a customized CPU among mainstream cloud computing service providers.
AWS Debuts Ground Station
Not only that, on the 28th, AWS also launched Ground Station (satellite receiving ground station), becoming the first company in the world to launch this type of service. It is reported that this service allows users to use satellite ground stations to connect with satellites and send the information to Amazon's data center for analysis, while effectively saving costs.
The Glacier Deep Archive, which will be launched in 2019, is a data storage service. Amazon said its fee will be a quarter of Amazon's current storage price.
Amazon FSx for Windows file servers is a service compatible with Windows. AWS is designed to facilitate enterprises to run fully compatible Windows applications in the cloud environment. Industry insiders said that this product may help Amazon grab some business from Microsoft's Azure cloud computing.
Although AWS CEO Andy Jassy previously stated that AWS is not interested in blockchain, at the re:Invent conference, AWS announced the official launch of two blockchain services, namely QLDB (Quantum Ledger database) and AWS Managed Blockchain (management Blockchain) can be used to manage peer-to-peer payments, process loans, and help companies conduct transactions with distributors and suppliers.
In the field of hybrid cloud, AWS has cooperated with VMware, a global cloud infrastructure and mobile commerce solution provider, to transform into an internal hardware supplier. Starting next year, it will begin to sell hardware to customers, and use the newly launched AWS Outposts service in its own data Operating in the center, the two companies share revenue.
In addition, Amazon has also intensively launched dozens of new products such as DeepRacer's global autonomous racing program, SageMaker Ground Truth training set labeling service, Amazon Elastic Inference, Amazon Forecast time series forecasting and other dozens of new products.
Dive into the field of machine learning
Among the many new products, the most eye-catching is the first AI chip Inferentia launched by Amazon to further deepen the field of machine learning. The chip will go on sale next year.
Amazon said that Inferentia will be a highly cost-effective processor with high throughput, low latency and sustained performance that will help researchers in the inference process and support popular frameworks such as INT8, FP16 and mixed precision. More importantly, it can support multiple machine learning frameworks, including TensorFlow, Caffe2 and ONNX.
At the beginning of 2018, it was revealed that Amazon was independently developing AI chips, and now AWS has "dust settled". This move is also considered to be a sign of Amazon's challenge to the two giants in this field, Intel (Intel) and Nvidia (Nvidia).
In the field of AI chips, Intel and Nvidia have always been leaders. Data shows that Intel’s Xeon processors are installed in 98% of the world’s data centers, and data center revenues account for 30% of Intel’s total revenue, and its business scope covers many artificial intelligence segments.
For another giant Nvidia, chips are also its fastest-growing business. As we all know, both companies generate revenue by designing and manufacturing chips for technology companies such as Apple, Amazon, and Microsoft.
As one of the largest customers of Intel and Nvidia, in terms of machine learning, Amazon previously installed Nvidia's high-end Tesla GPU and Intel's Xeon to provide itself with powerful computing power. However, Amazon is now starting to independently develop AI chips, gradually reducing its dependence on Intel and Nvidia, in order to strengthen its control of its own industrial chain. The impact on Intel and Nvidia is still unknown.
Some analysts also believe that Amazon chips do not directly threaten the businesses of Intel and Nvidia because they will not sell chips externally. However, Amazon's external cloud computing services will begin to use its own chips in 2019, which will undoubtedly cause Nvidia and Intel to lose a large customer.
Before Amazon entered the market, international Internet giants such as Microsoft, Google, Facebook, etc. had already made arrangements in the semiconductor chip industry. The top ten global semiconductor chip companies are currently all European and American companies without exception... and the top ten account for 58.8% of the global market share. Industry mergers and acquisitions have reduced the breadth of special products provided by semiconductor suppliers, and Internet companies have developed special functional requirements for their new equipment after a certain stage of development. Luo Wei, general manager of Foxconn Industrial Internet (Fii) Fuhua Branch, said that companies have deepened their own moat by extending to the upstream chip industry. Wei Shaojun, dean of the Department of Micro and Nano Electronics at Tsinghua University and director of the Institute of Microelectronics, once publicly stated that AI has not yet seen killer applications. The development of AI chips is likely to encounter a setback period in the next 2 to 3 years. Today, meeting specific applications is the main focus The purpose of the AI chip needs to think about where to go.