On November 28th local time, Amazon Cloud Technology (AWS) unveiled two chips at the ReInvent 2023 conference held in Las Vegas and announced the expansion of cooperation with Nvidia.
Against the backdrop of a shortage of GPUs, major technology giants are increasing their investment in self-developed chips. The two latest chips released by AWS are the AI chip Trainium2 for artificial intelligence large model training and a customized general-purpose chip Graviton4. Earlier this month, Microsoft also launched two custom chips at the Ignite Developer Conference held in Seattle to address the increasing challenge of large model training costs.
AWS CEO Adam Selipsky stated at the opening ceremony that AWS will begin offering a new training chip, Traineum2, next year.
But AWS and Microsoft's newly released AI chips are still unable to replace Nvidia's AI chips. At the opening ceremony of the reInvent conference, AWS stated that it will provide customers with access to Nvidia's latest AI chip H200. NVIDIA founder and CEO Huang Renxun attended the opening ceremony of the conference.
The latest second-generation AI chip released by AWS, Trainium2, has four times the speed and twice the energy efficiency of the previous generation. Selipsky stated that startups Databricks and Amazon supported large model leaders plan to use the new Traineum2 chip to build large models.
David Brown, Vice President of Computing and Networking at Amazon Cloud Technology, said, "Chips are the foundation of all user workloads, which is why Amazon Cloud Technology has always regarded this field as its crucial innovation area. With generative AI gaining widespread attention, Tranium2 can help customers train machine learning models faster at lower costs and better energy efficiency."
It is worth mentioning that Aerospace is a direct competitor to OpenAI supported by Microsoft. Anthropic has developed the artificial intelligence big model Claude 2, which directly benchmarks against GPT-4. Dario Amodei, co-founder and CEO of Anthropic, also appeared at the opening ceremony of the reinvent conference. Amodei is a former executive at OpenAI.
In September of this year, Amazon announced an investment of up to $4 billion in cash in Aerospace to expand its layout in the artificial intelligence cloud field, which is also Amazon's largest investment in the field of generative artificial intelligence to date. As part of the transaction, AWS will become the main cloud provider for Antitopic's critical task workloads, providing access to the AWS Training and Inferentia chip computing infrastructure for the Antitopic team. Antitopic will utilize a large number of proprietary chips purchased from Amazon to train its future artificial intelligence models.
The latest Graviton4 chip processor released by AWS is a general-purpose CPU chip based on the Arm architecture, which has a 30% performance improvement compared to the previous generation and consumes less energy than Intel or AMD chips. This is also the fourth generation CPU launched by AWS within five years. AWS stated that there are currently over 50000 customers using the company's Graviton chips.
AWS's Graviton chips also face direct competition from Microsoft. Microsoft announced the launch of custom CPU chip Cobalt at the Ignite conference, with Graviton and Cobalt both using the Arm architecture.
Behind the increase in investment in self-developed chips by technology companies is the surge in demand for large model training driven by generative AI. Earlier this year, AWS launched a generative artificial intelligence service called Bedlock, which allows customers to develop their own chatbots and image generation services using language models from Amazon and other startups.
At the opening ceremony of the reInvent conference, AWS announced the launch of the latest chatbot service Amazon Q for enterprise users, which can connect to various business software tools and starts at $20 per month.
The launch of Amazon Q means that Amazon will pose a challenge to Microsoft and Google in the field of productivity software. The cost of Copilot for Microsoft 365 and Duet AI for Google Workspace is both $30 per month.
Steven Dickens, Vice President of Future Group, a technology industry research company, wrote in a report, "AWS's Amazon Q will change the rules of the game for customers, and I expect Amazon Q to be widely adopted by developers and cloud administrators in the coming months."
The recent personnel tremors at OpenAI have also raised ethical concerns throughout the artificial intelligence industry. Selipsky intercepted a news article at the opening ceremony of ReInvent about Microsoft's brief restriction on employees using OpenAI's ChatGPT due to security concerns, and provided Amazon's response to the security of generative AI.
AWS has recently announced a new protection measure called Guardrail for offensive content in generative AI applications. Selipsky stated that the service allows users to filter out "harmful content".