RC

The future I want for AWS

By Richard Crowley

AWS is a juggernaut. Individual business units are larger than some entire industries. The scale of their infrastructure defies comprehension. And yet there’s chatter about AWS being on the decline. “Washed,” they say. I’d expect this from the VC set, eagle eyes trained on the would-be disruptors. But dozens of casual conversations with practitioners over the past year, people with no agenda beyond never returning to racking and stacking, got me thinking about the future. It would be such a tragedy if AWS really did decline. To prevent that, in my opinion, they should recommit themselves to infrastructure and dramatically shrink their product surface area.

The scale of AWS’ operations enables and forces a level of investment in scalability and reliability few other operators can match. As a practitioner, I benefit greatly from their investment. I’m almost happy to pay through the nose for network transit to build on this infrastructure.

It’s fashionable to complain about the cost of operating in the cloud, claim you can do it cheaper, and show your work by sharing your Dell shopping cart and a quote for a 95th-percentile transit contract. There’s just so much more to it than that. AWS is playing a different sport at every level from power infrastructure through chip and datacenter design, network architecture, and compute and storage flexibility. If you replace an AWS installation running in three availability zones with some gear you racked at even a very good colo, you may have saved money but you’ve also replaced your availability strategy with luck.

So you stick with the cloud. You might try, like I did for many years, to limit yourself to the basic services. “Just EC2 and S3.” You tell yourself this will make it easy to move to a different cloud provider or a datacenter down the road. Something (SDN minutiae, probably) will snap you out of that, you’ll realize even EC2 is lock-in, you’ll become zen about it, and you’ll start to explore the hundreds of other AWS services.

Some of them are great and will save you time. Some are not and will cause hair loss. I am not here to stack-rank the higher-level AWS services. I’m here to say that effectively all of them are a distraction from AWS’ core competency of providing scalable, reliable infrastructure. So why do they exist at all? AWS can create a larger profit margin by selling software running in EC2 than by selling EC2 alone. And AWS’ margins here would be other people’s opportunity except for one thing: The killer feature of most AWS services is that they show up on the same bill as EC2 and S3, thereby sidestepping the procurement department. AWS Marketplace existing is proof that this is true.

R&D and baseline operating costs mean that a theoretically high-margin AWS service can operate at a loss if it fails to achieve critical mass. If enough services don’t cover their fixed costs, their dead weight could drag EC2 and S3 down with them. That’s my real worry. I worry that we lose the best cloud infrastructure under the weight of a long tail of bundled but otherwise undifferentiated SaaS.

Which brings me to my dream for AWS. I wish AWS would shutter or spin out the vast majority of its products and recommit themselves to infrastructure. I think AWS would still be worth the price if they pared it down to just these eight services:

If we lose any of these technologies because AWS made one too many Sagemaker services, I’ll never forgive them.