AWS re:Invent 2023 Keynote by Adam Selipsky

Here I have captured key highlights from 1+ hour of AWS re:Invent 2023 Keynote by Adam Selipsky, AWS CEO.

New AWS services announced are preceded by New & bold font

Keynote by Adam Selipsky, AWS CEO

Summary

ReInventing Cloud Computing

Highlights

12th AWS re:Invent in Las Vegas: 50k+ in-person & 300,000+ attendees registered virtually, over 2200 sessions
ReInventing Infrastructure:
- AWS Infrastructure spans 32 Geographical Regions around the world and each region consists of atleast 3 AZs
- Compared to other cloud providers, AWS has 3x Data Centres, 60% more Services, 40% more Features
ReInventing Storage:
- Amazon S3 (2006 launched)
  - S3 Deep Archive
  - S3 Intelligent-Tier: Saved Customers 2+ Billion dollars
  - New Amazon S3 Express One Zone:
    - for high speed storage needs of Customers
    - Highest Performance & Lowest Latency Object storage
    - 10x faster than S3 Standard storage
ReInventing General Purpose Computing:
- In 2018, AWS became first major cloud provider to launch GP compute silicon chip - Graviton
- 2020 - Graviton 2 - provides 7x performance than G1
- 2022 - Graviton 3
  - 25% performance than G2
  - 60% less energy consumption
  - Best price performance in EC2
- 2023 - New Graviton 4
  - Most powerful & energy efficient chip ever made
  - 30% faster than G3
  - 40% faster for DB applications
  - 45% faster for large Java apps
  - Preview of R8g EC2 instances based on G4
ReInventing with Generative AI
- Generative AI stack
- AWS has been collaborating with Nvidia for 13 years to bring GPUs to the cloud:
  - All GPU instances released in last 6 years is based on AWS Nitro
  - Jensen Huang - Founder & CEO of Nvidia: Dialog with Adam
  - New Nvidia DGX cloud to AWS
    - DGX cloud is Nvidia’s AI factory
- New EC2 capacitive blocks for ML:
  - Reserve EC2 ultra clusters with hundreds of GPUs
  - Ideal for training & fine tuning FMs, short duration workloads & handing capacitive surges
- AWS Trainium - 2nd gen chips launched - 4x faster than 1st gen
- AWS Inferentia - 2nd gen chips launched
- AWS Neuron:
  - SDK to optimize ML on Trainium & Inferentia
  - Supported Frameworks: TensorFlow, PyTorch & upcoming support for JAX
  - Supports 93 of top 100 Models
- Amazon SageMaker: managed service to build, train, deploy ML models at scale
- Amazon Bedrock: easiest way to build & scale Gen AI apps with LLMs & FMs
- Amazon Bedrock Customization:
  - (New) Fine Tuning
  - (New) Retrieval Augmented Generation (RAG) with Knowledge Bases
  - (New) Continued Pre-Training for Titan Text Lite, Titan Text Express
- (New) Agents for Bedrock
- (New) Guard Rails for Bedrock
- Dario Amodei - Anthropic CEO & co-founder: Dialog with Adam
- Amazon Titan Models:
  - High performing FMs from Amazon
  - Titan Text Lite, Titan Text Express, Titan Text Embedding model
- Lidia Fonseca - Chief Digital & Technology Officer, Pfizer: Speech
- Amazon Code Whisperer
- (New) Amazon Q:
  - Gen AI powered assistant for work, that is tailored for your business
  - Your expert assistant for building on AWS: Currently it has been fed 16 years of AWS docs/ knowledge base and is available in AWS Console & AWS Doc and also in popular IDEs like VS Code
  - Amazon Q Code Transformation: to upgrade code Framework / version, like Java upgrade
    - Coming soon: upgrade DotNet applications to help migrate from Windows to Linux
  - Amazon Q in Quicksight
  - Amazon Q for Amazon Connect
- Most comprehensive set of Data Services, since your Data is unique & most helpful tool:
- Zero ETL Integrations:
- Amazon Data Zone
  - Catalog, Discover, Share & Govern Data across your organization
  - (New) Amazon Data AI Recommendations
ReInvent: Bold bets by Amazon:
- Amazon Third Party sellers
- Amazon Prime
- AWS
- Amazon: project Kuiper: Global high speed satellite network for broadband / internet connectivity