AWS Re:Invent 2020 keynote by Andy Jessy
Andy Jessy, CEO of AWS, during his AWS re:Invent 2020 keynote on December, 1st

A summary of Andy Jassy’s keynote during AWS re:Invent 2020


Share on linkedin
Share on twitter
Andy Jessy, CEO of AWS was holding his almost 3-hour-long keynote on December, 1st kickstarting AWS' 3-week virtual conference re:Invent 2020. Here is what to remember.

Just like Tim Cook’s announcement of the iPhone 12, the AWS re:Invent 2020 had to be reinvented considering the Covid-19 constraints. The whole event has been shaped in a virtual conference and Andy Jessy was standing alone with probably some recorded applauses. BTW, Andy started his speech by stating how extraordinary was 2020 and that Covid has been so difficult for so many people. More surprising, the CEO of AWS took a strong and courageous position to condemn the way we treat black people, reacting to the killing of George Floyd. I was even hearing the emotion in his voice and I don’t think this was faint. This is probably announcing the post Trump era.

AWS is growing by 29% Y/Y but with at lower pace

Without transition, back to business and numbers. AWS has made $46B revenue annualized from Q3 2020. This is 10B more than one year before. Those are impressive numbers. Still, in his attempt of demonstrating the acceleration of AWS growth, the performance is less than before as CNBC is showing in the above diagram with an annual growth of “only” 29%. This was in contradiction with Andy Jessy’s speech explaining that “the pandemic accelerated cloud adoption by several years“.


AWS is now in the top 5 of IT actors

What was more convincing, it is today’s place in the big league of IT champions: compared to 2010 where AWS was absent from the top 10, AWS is now ranking at #5, according to IDC. It is interesting to see as well that the other newcomer in 2020 top10 is Salesforce who has recently acquired Slack to threaten more Microsoft power. 

How to reinvent a business?

It is no coincidence that AWS event is named re:Invent. It is true that businesses should try to reinvent themselves instead of being surpassed by a competitor or a new entrant. And it is better to reinvent your business when it is healthy. Andy Jessy listed 8 principles to reinvent a business:
  1. You need the leadership will to invent and reinvent
  2. You cannot fight gravity: if something has to change, it will change. Netflix was right to disrupt their DVD renting business with streaming. was right to open their e-commerce web site to 3rd party sellers.
  3. You need talents that are hungry to invent and probably this will come with new blood as they can more easily transform than the ones that are within the company for a long time. And I am adding: why not hiring an interim manager that however are is not a superhero
  4. The focus should be solving real customer problem rather than doing this because its cool
  5. Speed is key: a culture of emergency must exist in the company
  6. Do not complexify. This was clearly an attempt from Andy Jessy to push aside inclinations to opt for a multi-cloud strategy. 
  7. Use the platform wiht the broaest and deepest set of tools. Of course, the CEO of AWS is saying that AWS is having this platform and this is not totally wrong but Microsoft is also having a comprehensive cloud toolset.
  8. Set aggressive top-down goals
 As you can see, most of Andy Jessy’s advice are not technology driven and it is key to build a reinvention culture within the company. 
AWS offering

Compute Intances (EC2) with macOS and new chipsets

First, AWS is now supporting macOS instances

New Graviton2 chip from AWS
AWS Inferentia Machine Learning Processor
Habana Gaudi AI Processors
Habana Gaudi AI Processors

In terms of processors, AWS is supporting chipset from Intel, AMD and Arm. But AWS is also continuing to build their own chips like they did with Nitro chips: AWS annonced a new Graviton2 chip based on top of an Arm processor adding more horsepower than the previous Graviton chip and Inferentia, a chip focussed on Machine Learning inference. AWS is also now supporting the new Habana Gaudi processors from Intel, built specifically for Machine Learning training (Intel acquired Habana, an Israeli company with 150 employees for $2B in December, 2019). A Habana Gaudi-based EC2 instance supports the 2 major open source ML Platform TensorFlow and PyTorch. AWS also announced in 2021 AWS Trainium, a ML training chip custom designed by AWS and supporting also TensorFlow and PyTorch. As you can see in the pictures, Graviton2 and Inferentia are built by Annapurna Labs which is an Israeli company acquired in 2015 by AWS  for US $370M.

Containers anywhere


AWS is claiming 2 thirds of all containers. We remind that AWS is proposing 3 container offerings:

  1. Amazon Amazon Elastic Kubernetes Service (EKS)
  2. Amazon Elastic Container Service (Amazon ECS)
  3. AWS Fargate to run serverless containers
AWS annonced with Andy Jessy’s keynote for 2021 Amazon EKS Anywhere and Amazon ECS Anywhere to run EKS and ECS in your own datacenter. This is clearly an attempt to enter the private cloud market.

Event-driven serverless computing

AWS Lambda is AWS solution for serverless computing. As a first change, AWS is announcing that they are droping they 100 ms increment of billing to 1ms. AWS also announced Lambda Container Support a solution to use Docker or Open Container to manage serverless images. Moreover, AWS announced AWS Proton, a deployment service to deploy containers and serverless applications. 

More performant block storage and SAN in the cloud

In the pursuit of always better performances, AWS annonced Elastic Block Store (EBS) gp3 which more throughput and Io2 with more IOPS. But, more interresting, AWS is now proposing Io2 Block Express a SAN solution built for the cloud. SAN classical manufacturers may have good reasons to be worried about this announcement. 

AWS will release you from Oracle and MS SQL claws

The CEO of AWS has started his speach on databases with a wild attack against Oracle and Microsoft accused (rightly) to abuse from vendor lock-in situations and prohibitive prices. “This is something that the customers are fed up with“, he said. As a solution, he is proposing Amazon Aurora with MySQL and PostgreSQL databases at 1/10th of the cost. I must say that it is ironical that Andy Jessy is talking about vendor-locking, knowing that it is quite difficult to exit AWS once you have been addicted to it. To ease the transition, you can now count on Babelfish for Aurora PostgreSQL, a service translating SQL Server T-SQL proprietary dialect for PostgreSQL. Babelfish for PostgreSQL will be available on GitHub in 2021 as an open source project, using the Apache 2.0 license. 

Purpose-built databases

AWS purposed database
AWS has created 7 purpose-built database

But relational databases are not always good. Sometimes a purpose-built database is doing a particular workload or use case extremely well. AWS is proposing 7 of those purpose-built databases as shown in the figure above:

  1. Amazon DynamoDB to store Key / Value
  2. Amazon ElastiCache for In-Memory Store
  3. Amazon Netpune for graphs, launched in 2017
  4. Amazon DocumentDB to store documents, with MongoDB compatibility
  5. Amazon Timestream to store time-series, for IoT and operational applications that makes it easy to store and analyze trillions of events per day
  6. Amazon Quantum Ledger Database (QLDB) that provides a transparent, immutable, and cryptographically verifiable transaction log ‎owned by a central trusted authority. Ledgers are typically used to record a history of economic and financial activity in an organization. 
  7. Amazon Keyspaces for Apache CassandraApache Cassandra is used when you need scalability and high availability without compromising performance.

Purpose-built analytic stores and datalakes

As AWS is proposing purpose-built databases, AWS is also proposing purpose-built analytic stores, optimized for particular use cases:

  1. Amazon Ethena for interactive query on your data lake or your S3
  2. Amazon Elastic MapReduce (EMR) to process vast amounts of unstructured data using Apache SparkApache Hadoop or Presto
  3. Amazon Elasticsearch Service for operational Analytics on logs
  4. Amazon Kinesis for Real-Time Analytics
  5. Amazon Redshift for data warehouse
Lots of companies creates and maintain a datalake to take data from disparate silos so that there is one place where to do analytics and machine learning from. Lots of datalakes are built on Amazon S3 but also some customers want to use purpose-built databases listed earlier. So the question is how to move data back and forth between those different data stores. AWS is announcing AWS Glue Elastic Views that finds and replicates data across multiple data stores. Glue is AWS ETL (Extract Transform Load) service but Elastic Views is automatically keeping the data in the target data store updated. 

Machine Learning

On ML, Andy Jessy shared what do the ML practitioners actually asking to AWS:

ASK #1: We want to have the right tools for ML practitioners, which involves chips and frameworks, knowing that the top ML framework that practitioners are using are: Pytorch, TensorFlow, and MXNET.

ASK #2: We want ML tools for everyday developers and data scientists.
AWS answered last year by creating AWS SageMaker and SageMaker Studio, an Integrated Development Environment (IDE) for ML, which includes SageMaker Notebooks, SageMaker Debugger to debug models, SageMaker Experiments to organize ML training, SageMaker Model Monitor to detect concept drift, and SageMaker Autopilot to perform Automated Machine Learning (AutoML). This year, the effort has been made on SageMaker on data preparation for ML with Amazon SageMarker Data Wrangler.

Additionally, ML features are the properties ML models used during training and inference to make predictions. In an ML application that recommends a music playlist, features could include song ratings, which songs were listened to, and how long songs were listened to. The accuracy of an ML model is based on a precise composition of features. Often, these features are used by multiple teams training multiple models. To store ML features, AWS announced for immediate availability Amazon SageMaker Feature Store

And last but not least, AWS was also announcing Amazon SageMaker Pipelines, a Continuous Integration / Continuous Development CI/CD service, and workflows for ML.

ASK #3: ML for those who don’t want to build models

AWS is offering an impressive set of ready-to-use ML capabilities:

  • Amazon Rekognition to recognize something in images and videos with ML
  • Amazon Polly to go from text to speech (check Léa who speaks French)
  • Amazon Transcribe to go from speech to text
  • Amazon Translate
  • Amazon Comprehend, which is a Natural Language Processing (NLP) service that discovers insights and relationships in a text
  • Amazon Textract, which easily extract printed text, handwriting, and data from a document in a more powerful manner than an Optical Character Recognition (OCR) system
  • Amazon Kendra, an intelligent search service powered by ML
  • Amazon Lex, AWS conversational AI for chatbots
  • Amazon Personalize to create real-time personalized user experiences faster at scale
  • Amazon Forecast uses machine learning to deliver highly accurate forecasts.
  • Amazon Fraud Detector, a fully managed service that uses machine learning (ML) and more than 20 years of fraud detection expertise from Amazon, to identify potentially fraudulent activity
  • Amazon CodeGuru automates code reviews and optimizes application performance with ML-powered recommendations. Java was supported last year, AWS is now adding Python. They are also adding a CodeGuru Security Detector that provides real-time alerts when code may not be secure and Amazon DevOps Guru that uses ML to identify operational issues long before they impact customers: missing alarms, early warning of approaching resource limits, under-provisioned compute capacity, memory leaks, etc.
ASK   #4: I would like to benefit from ML without knowing that it is ML
As an example, Amazon QuickSight, launched in 2016, is a ML-powered BI service for the cloud. This year, AWS is proposing Amazon QuickSight Q to allow to type a question in natural language and get BI answers in seconds. 

Contact centers


Amazon Connect is AWS contact center in the cloud, built in 2017 with AI/ML in mind.

This year, AWS enriched their contact center solution with Amazon Connect Wisdom, a capability that uses ML to deliver agents the product and service information they need. Connect Wisdom is using ML to listen to the call transcription and then put the right information to the agents in real-time. 

AWS is also adding Amazon Connect Customer Profiles which gives agents data on a customer’s related activity. 

Amazon Connect also comes with Contact Lens a Contact Center analytics solution.  It automatically transcribes and analyzes customer calls. It also analyses positive or negative sentiment and long periods of silences. Now, AWS is able to do this in real-time, so that a supervisor can coach the agent immediately or pick the call. 

On top, Amazon Connect Voice ID is a real-time caller authentication system that uses ML voice analysis. After the user opts-in, he says a couple of sentences and the system build a voice print of this customer. Later, when the user will call in again, we will let him talking, explaining what is his problem about. And within a few seconds, Voice ID will show to the agent who is this customer with a confidence level of that prediction. 

Industrial processes and IoT

In an industrial context, one of the challenges is to predict when industrial equipment needs maintenance before it breaks. Therefore, Amazon is launching several products:

Hybrid cloud

Hybrid cloud was a term invented by infrastructure datacenter providers. It is usually understood as a combination of Cloud and on-premise datacenters. Which leads to a infinite debate: is it a binary solution: will we only use on-premise or only use the cloud. Andy Jessy confess that AWS contributes to the confusion by explaining that lots of companies won’t have datacenters anymore. But hybrid should also includes other edge notes, not only data centers. A factory or an hospital could be an example of an edge node. 

AWS Outposts allows already to run AWS infrastructure on-premise. Outposts is a a full rack of 80″ tall. AWS is now proposing smaller size of Outposts with a 1U size and a 2U size. 


Amazon cloud revenue jumps 29%, in line with expectations, CNBC, October 29, 2020

Photo credits

Screenshot of AWS re:Invent web conference

Photo by Johann Walter Bantz on Unsplash 

Image by Gordon Johnson from Pixabay

Photo by Museums Victoria on Unsplash

Leave a Reply