AWS re:Invent 2023: Key announcements for enterprises

Amazon Web Services (AWS) commenced its annual AWS re:Invent event in Las Vegas on November 28, featuring a rapid succession of announcements, primarily centered around advancements in its signature platform. The nearly 2.5-hour keynote by AWS CEO Adam Selipsky focused predominantly on artificial intelligence (AI) developments. 

Here are some of the key announcements from the AWS event! 

Amazon Q for Enterprises

Microsoft promotes its Copilot AI, and Google advances with duet AI, both aiming to assist cloud operations. Now, AWS follows suit with Amazon Q, deeply integrating it across various cloud services like Amazon CodeWhisper and Amazon Connect. According to AWS CEO Adam Selipsky, Q can answer questions, generate text and visualisations, and execute actions, streamlining workload management. Priced at $20 per user per year and currently in public preview, Q is trained on 17 years of AWS knowledge. Q goes beyond answering questions to understanding app workloads and suggesting AWS solutions for short-duration apps. The AI-powered chatbot is expected to save customers time in architecting, troubleshooting, and optimising workloads. 

New Features for Amazon Bedrock

Selipsky, during his keynote, revealed updates to Amazon Bedrock, a fully managed service offering leading AI models via API and developer tools for generative AI applications. The new customisation options were added, empowering users to adapt AI models to their specific requirements and data. 

The enhanced customisation includes a feature for fine-tuning models with labeled examples from a customer's data, enabling the teaching of business-specific information. Additionally, Retrieval-augmented generation (RAG) allows models to incorporate extra details from a customer's internal data sources. Another feature, Continued pre-training, utilises a customer's unlabeled data to enhance a model's knowledge and reasoning in their industry or field. 

AI Chip and Nvidia Partnership

During the event the company revealed the latest iteration of Trainium 2, a specialised chip crafted for training large language models (LLMs) and foundational models like OpenAI's GPT-4. This chip is optimised for handling neural networks with trillions of parameters, emphasising the industry-wide trend of scaling AI models. Trainium 2's focus lies in efficiently managing the vast parameters of neural networks, enhancing their scale and computational power. 

Amazon also introduced Graviton 4, an upgraded version of its custom microprocessor. Additionally, the company announced an extended partnership with Nvidia, incorporating Nvidia's cutting-edge chips into its cloud computing service. 

S3 Express One Zone

AWS introduces a significant upgrade to its S3 object storage service with AWS S3 Express One Zone, a high-performance and low-latency tier promising a 10x boost over standard S3. This enhancement caters to data-intensive applications like AI/ML training, financial modeling, and high-performance computing. It excels in handling small files with rapid access, outperforming S3 Standard by up to 10x in reading small objects. 

Notably, AWS ensures a 50% reduction in request costs for Express One Zone. The backend enhancements include a new bucket type, authentication model, and a refined naming scheme. This improved S3 tier is now widely available. 

Fire TV Cube

Amazon has introduced affordable $195 devices for enterprise users to access virtual desktop environments, such as Amazon WorkSpaces, online. These devices are integrated into Fire TV Cube hardware, a strategic move by Amazon to utilise its expertise in streaming media players. The decision to develop this hardware stems from customer feedback, expressing a desire to reduce IT expenses by replacing traditional desktops and laptops with more cost-effective alternatives.