Amazon’s 2020 cloud strategy and my 5 takeaways from AWS re:Invent 2019
This week I’ve attended AWS’s annual re:Invent conference in Las Vegas. This visit was my second time at re:Invent following my visit to re:Invent 2016. This event has become huge over the years, with about sixty-five thousand attendees from all over the world and about three thousand sessions to choose from, many activities, expo, and more. The conference has become also more challenging to manage schedule-wise. It requires appropriate planning as sessions are scattered across locations, and those sessions, as well as the keynotes, require you to secure a seat in advance.
In contrast to my last visit, when the significant announcements were technical yet essential innovations, including serverless and container focused offerings and an initial set of AI-driven standalone services, this year’s re:Invent was with far greater strategic underlying themes. To me, this year’s re:Invent showed that Andy Jassy (Amazon Web Services CEO) and his team had done a great job addressing major fronts that would help AWS secure the top-right place in the “magic quadrant”.
So, what are those fronts, and what is AWS doing there?
1 — AWS wants to become the Machine-Learning-First cloud with end-to-end ML offerings
AWS wants to provide the broadest set of tools for ML software to build, train, deploy, and run — and has released SageMaker Studio, which is a web-based IDE for machine learning workflows to wrap this under one roof to achieve this. This is important since it helps drive developers to base their actual ongoing daily creative work on AWS and not just run, test, and deploy once they have something ready (see more in point 5 later).
But wait, there’s more. The state of ML development is not as mature and rich as the world of generic software development, and AWS is now offering ML developers with a place to work and collaborate on their notebooks (SageMaker Notebooks) new ways to debug, profile, and improve their models (SageMaker Debugger), monitoring deployed models (Model Monitor) and detect concept drifts. They also made significant advancements in the new world of AutoML with their new Autopilot offering, which allows a developer to load the data and let Autopilot come up with the best-trained model yet provide control and visibility. These and other SageMaker features offer probably the broadest and most complete set of ML capabilities among the cloud providers.
On a side-note, this is part of AWS’s continued effort to lock customers into AWS as a single cloud all-in approach, while the reality is that many customers require a multi-cloud strategy. The AWS approach and this fact is a growing conflict.
2 — AWS custom-built compute hardware as a competitive edge
Amazon’s acquisition of the Israeli Annapurna Labs in 2015 keeps paying off as a very successful one for AWS. AWS is making a strong move forward in two fronts empowered by the Annapurna Labs team, and this helps AWS compete with compute offerings coming from Google and Microsoft.
The main news is around AWS Graviton2, which is a 7nm ARM-based processor that supposedly offers 40% better performance from comparable x86-based instances (mostly Intel Xeon based) for 20% less. Like others mentioned, this has some parallel’s with Apple’s hardware strategy on mobile, by leading innovation at the processor in house and not waiting around for 3rd party suppliers. This new CPU is assumed 7x faster than the Graviton1 — impressive.
Secondly, AWS announced a new type of compute instances now generally available for ML inference powered by their Annapurna Labs built inference chips — inferentia. This is important as most cloud spend when it comes to ML usage is about inference (and not for training, which is relatively small in comparison). Both these HW capabilities would allow AWS to provide cost-effective, performance strong, offerings compared to their rivals.
3 — From former “Cloud Only” (or mostly) to “Anywhere-You-Need”
For many years since its launch, AWS was a “cloud-only” sort of thing, which seemed great to new startups, existing companies building new independent services, and to great success such as Netflix, who managed to build their entire service on top of AWS. But this never provided the full solution needed and prevented those who have regulatory constraints, or specific latency requirements, or needs relating to connected mobile devices and thus prevented a big part of the market from adopting AWS fully. Also, from a cloud competitiveness point of view, this situation always left Microsoft as the better positioned to provide hybrid cloud solutions as they were already in the data center, leaving AWS out the door for many large potential enterprise accounts.
Well, this is now quickly changing, and AWS is making 3 critical efforts to do so. The first is the previously announced Outposts that now are finally available. Outposts are AWS in a box that a customer could have delivered and connected on its premises and use the same APIs, integration with other AWS services, and central AWS console management while the “Outpost box” sits on-premises. Secondly, AWS Local Zones, which is AWS’s newly deployed “in your neighborhood” location (now starting in Los Angeles), meaning you could have deployments on specific pinpointed locations (suppose you need very low latency to particular resources). The third, AWS WaveLength — which is AWS push into integration with the 5G mobile network core (which is finally becoming easier given the virtualization of the telco stack and mobile network) meaning mobile application could have the network itself as an edge with ultra-low latency — this is achieved with a Verizon partnership leveraging the new 5G network and standardized to spread across telecommunication providers internationally. Looking ahead (and taking into account the already existing IoT related offerings such as Greengrass), AWS will soon be everywhere — very differently from not so long ago.
4 — AWS is after ALL the enterprise data with next-gen Enterprise Search
As part of the competitive effort against Microsoft and Google — AWS is now offering a new natural language-driven Enterprise Search offering called Kendra. Kendra is interesting as it brings AWS closer and more integrated with all the enterprise data, and not only that which found its way to AWS assets such as S3 or Redshift, and so forth. While Kendra is still in its early preview days, looking ahead, AWS seeks to build an open-source community around the connectors needed to bring in data from all siloed locations across the enterprise and become the core index and source of answers. This provides one more piece of the context switch cost that would tie in customers for the long run (while hopefully happy) once using Kendra.
5 — Locking in developers and data scientists with Productivity Tools
AWS is making sure the developer journey to ML-driven software begins with AWS from the first line of code. SageMaker Studio has the potential to become for the modern ML-driven developers what Microsoft’s Visual Studio was to software developers, and it is not only about SageMaker and ML its also about any software developer — with the newly announced CodeGuru, which provides automated code reviews and application performance recommendations, powered by ML. Imagine, now, a developer can detect a line of code that, if implemented more efficiently, could save big on compute costs. All this brings AWS to the fingertips of the developers already while first writing new code.
Final thoughts
So, what does all this mean? Well, as someone who was a software developer earlier in my career, this is, of course, exciting as it gets, right? Software developers today have so many tools and capabilities at their arsenal to be productive and bring in innovation. The next decade which is upon us, would be so exciting that I can’t wait.
From an investor’s point of view, Amazon announcements are always scary because almost every time AWS announces a new capability somewhere a startup dies (or at least faces some uncertainty). And in most times, the conclusion is that horizontal software startups may it be in DevOps, ML-Ops, Big Data, and what not are very difficult to win in and are less attractive for investors (the investors of UiPath may disagree with me here) while the more exciting opportunity for VCs are in vertical AI-powered end to end software and services — but, this year, Amazon is launching certain vertical offerings as well and partnering up with key domain entities.
What is certain is that the future of technology innovation is exciting, and we are just at the beginning, so bring it on!
