There is a battle going on in the X-as-a-Service (XaaS) world that barely anybody is even aware of. That's the battle to build the most comprehensive, most performant artificial intelligence and machine learning as a Service platform (AI/MLaaS). In this post, I will compare AWS, Azure, and GCP's AI and ML platforms side by side.
Let's backtrack for a moment. Speaking purely about multi-tenanted architectures, we have now had Software as a service (SaaS) for 20-odd years (Salesforce and Concur back in 1998-1999) and Infrastructure as a service (IaaS) for 14+ years (AWS launched in 2002 and relaunched in 2006). Of course, since then we have had the rise of Database-as-a-service, Desktop as a service, and others. all of which can be generally lumped into the "cloud computing" category.
In addition to their extensive IaaS platforms that have been built over the last 15 years, these players (as well as IBM, Tencent, Baidu, and Alibaba) have also been building up large machine learning and artificial intelligence as a service platforms. These platforms are designed like most development platforms, to reduce the time to value by providing hardware infrastructure and software environments preconfigured for AI and ML use cases.
I wanted to get an idea of what each of the big three is doing and how they map to each other so I pulled together a bunch of information from all three vendors as well as this excellent page over at comparecloud.in. (Big shout-out to Ilyas F for his work.)
If you are looking at platforms, I hope this will at least give you a starting point.
(Side-by-side simplified heat map of AI & ML Services at AWS, Azure, and GCP)
Robotics, Artificial Intelligence, and Machine Learning Platform
The thing that struck me when I started reviewing Amazon's catalog was just how big and developed it already is. Microsoft seems to be following close behind, and Google, despite having a huge amount of AI talent, seems to be trailing in terms of productization. But regardless, I think that most people would be surprised to learn that this many developer-friendly AI services are already available by the big three providers.
Check out the column called Functions. The amount of services already available is stunning.
(Detailed view of Robotics, AI, and ML services from AWS, Azure, and GCP)
Databases & Analytics
One of the interesting observations in Dr. Kai-Fu Lee's excellent book AI Superpowers, is that the US leads in two areas: basic research and having a lot of structured and semi-structured data (from corporate ERP systems).
On a related note, the big three have been building and providing a number of more cost-competitive and performant DBaaS options over the last several years and so they are now repositories of vast amounts of structured and semi-structured client data.
Add to this providers like Snowflake, that are building Datawarehouse-as-a-Service (DWaaS) sitting on TOP of AWS, Azure, and GCP. Now you have even more data to pull from for training AI and ML systems.
(Detailed view of Database & Analytics offerings that sit underneath the AI/ML platforms.)
Not much to say here except that all three are of course very developer-friendly and provide a lot of tools to help with the entire development process.
(Detailed view of the development tools from each vendor.)
Infrastructure as a service
AI and ML ultimately sit on top of big and fast and cheap compute and storage, which is IaaS in a nutshell.
(Detailed view of the basic IaaS functions at AWS, Azure, and GCP.)
This is obviously a snapshot in time. It is also evolving and developing rapidly. I can't wait to see what these service stacks look like in another year or two, in particular how the AI services develop.