Is "cloud first" the best business strategy?

I'll start by saying I have my CCSP (Certified Cloud Security Professional) certification from ISC(2)- so I am *NOT* anti-cloud. The cloud ABSOLUTELY has it's place and it's benefits. That said, today's article is a critical examination of what I am going to suggest is an overuse of cloud. I'll say the quiet part out loud- the cloud is not ALWAYS the best place for everything.

The NIST definition of cloud computing gives 5 key attributes of cloud computing.

1. On-demand self-service. A consumer can unilaterally and quickly provision computing capabilities without prior approval from the cloud provider.
2. Broad network access. Cloud service are available over the network to multiple heterogeneous platforms (e.g. mobile devices, laptops/desktops, legacy servers).
3. Resource pooling. The cloud provider’s computing resources are pooled/distributed to serve multiple consumers using a multi-tenant model and resources are dynamically provisioned according to consumer demand.
4. Rapid elasticity. Cloud resources can be elastically provisioned and released, in some cases automatically, to dynamically provide overflow capabilities and pressure relief. To the consumer, the capabilities available for provisioning often appear to be unlimited and can be appropriated in any quantity at any time.
5. Measured service. Cloud systems automatically meter reservations and use of resources (e.g., storage, processing, bandwidth, and active user accounts) for the purpose of controlling, reporting and billing.

To take full advantage of the cloud, you should be aware of and leveraging each of these 5 points to your company's advantage. Failing to do this could cost you significantly more than running a traditional data center- sometimes by orders of magnitude. I'll explain.

I want to focus on elements 3 & 4- resource pooling and rapid elasticity. Computing assignments inevitably require more or less resources depending on the time of day and year- but when you start to reserve these for long term use so you always have them, the benefits of pooling and elasticity start to fade quickly. Here's an example:

Amazon S3 storage costs at the time of this article are $0.023?per GB. For comparison, I found a 16TB drive for $300 - and when you do the math, is $0.018 per GB. So I can buy a physical hard drive to store the equivalent volume of data and in ONE MONTH, I'm already ahead compared to purchasing S3 storage. Said differently, at list price 16TB of S3 storage on Amazon will cost an amazing $4,521 compared to $300 for a 16TB physical disk locally.

Okay- someone is already thinking that I have to have a server to put the drives in, network switches to access, administrators to support the network and the server, the data center to put it in and all the other overhead- and they aren't wrong... so what I just did there isn't an exactly fair comparison but the principle is valid and the point I am trying to make should still be considered. At some point, you need to look at the resources you are trying to reserve and determine your break even point. When is it more cost effective to do it yourself?

Here is a real world example in micro scale- internally, my company was looking to use a 24x7 high compute GPU resource to build and maintain some AI models. An entry level cloud reservation was going to cost us $1.20/hour for the number of GPUs we needed- at a cost of $10,512/year. By comparison, we were able to build a physical machine with 4x the core CPUs, 4x the storage and 2x faster GPU for a one time cost of $3,500. For an additional $5k/year we can rent an entire dedicated rack with all network connectivity in a managed data center. So for less than the cost of a single GPU machine in the cloud, we can get a more powerful system that is still operated in a managed data center.

For some real world examples at scale, one company expects to save $7M over the next 5 years and another company is ditching cloud microservices to save millions.

Okay, so let's talk strategy and application... here are some ideas to consider to start the conversation:

• If you are looking to use the cloud as your 24x7 operation with reserved resources, I encourage you to instead consider leasing physical rack space in a managed data center and deploying your own hardware instead.
• If you need the rapid on-demand availability of resources (e.g. - I need a server by tomorrow and I can't wait 4 weeks for purchasing to do the PO, get it shipped, built and racked in my data center), that's okay... use the cloud to get your system immediately.... THEN order your physical machine and set it up, retire your cloud machine and reclaim the recurring expense.
• Take advantage of the inherent elastic capability of cloud (bursting). Maybe you don't need a 32 core machine. Maybe you can get away with a 4 core machine and dynamically scale to 32 cores under load.

My bottom line- it is difficult to financially justify 24x7 cloud with reserved capacity. When you do the math, it may not be the most prudent business choice. So in conclusion, i content "cloud first" should be "business first"! When cloud makes sense, use it! But don't do "cloud first" as a business strategy.

So that's my view and opinion to start a conversation. Hopefully I've given some information for you to consider and think about. I'd be interested to hear your position... and that's what the comment section below is for. Let me know what you think.