Cloud 2023

“EC2 is total s*** we are moving to kubernetes.”

A few more minutes into that chat and an all too familiar theme reared its head. Making tech & platform decisions purely based on economics and cost.

This growing trend is nothing new in the world of business & perhaps not new in tech(changing vendors based on pricing is a common practice in enterprise IT). However this “change” variable is happening faster.

HashiCorp’s 2022 state of the cloud found that 2 of the top 3 reasons for enterprises to adopt a multi cloud strategy were based on pure economics. First to avoid vendor lock in. 2nd to avoid price jumps and lower immediate costs.

This isn't news, but what is news is the growing number of startups I've chatted with that are hedging their technical bets on the emerging category of “serverless”, “containerized” or at least “managed” compute platforms such as Kubernetes(or any containerization for that fact), Lambda plus other serverless platforms. The play is simple. When people use your service, your environment scales, and you get billed for real usage, not provisioned usage.

However to make this drastic of a change in what is often an incredibly complex application stack based purely on (maybe) saving some cash, seems a bit much. You can just as well find tons of articles trashing serverless as an expensive headache, alleviating yes (some) infrastructure management burdens but creating new application development and software development burdens just to make the thing you already built work with its new medium.

But shouldn't companies start at why their cloud bill is expensive in the first place?

Cloud buying programs launch a massive barrier of entry that is the very antithesis of why people rushed to the cloud in the first place: time.

In theory the cloud should be about agility and speed(and it is), however cloud buying is very much the opposite.

Cloud buying asks users to make predictive commitments, they can't change/alter or move out of(for the most part). This causes 99% of cloud buyers to leave the best discounts on the table each month in exchange for agility.

99% of all AWS  buyers are buying their compute at less then 40% discount off list. This stat is wild, given the best programs offer users 66-72% off list. No wonder so many people are moving to multi-cloud in a world where most tech buyers think 60% off list is the negotiating starting point.

The good news: Cloud buying is in the midst of a massive, and once in a lifetime transformation. In 10 years, more than 50% of all cloud buying will be done by AI. (give me a call in 10 years if this isn't the case and you win a set of steak knives).

This is a look at how (and why) the next 10 years of cloud buying will drastically change, and why the hyperscalers that embrace this change will gain competitive advantage over those who don’t.

Humans hate managing cloud procurement.

The reality of buying cloud licensing is painful. The hyperscalers force long commitments on their customers that are not flexible to the changing architectures of their environments, and in many ways are against the spirit of why the cloud became cool in the first place. Build what you need when you need it, stay light and go fast. But they forgot to add in one detail to that pitch. Pay list pricing.

Each provider has 100s of products that can contain 1,000s of SKUs. It’s complex. Even for folks that live in the AWS or Google Cloud CLI, provisioning machine architectures at scale is a mind boggling exercise that leads to purchasing too much or too little. Not to mention that most folks when choosing what Ec2 instance to run, or how to size their databases, are doing so with their hands tied behind their backs. Meaning they have little to no performance metrics behind these decisions. All of this will be replaced by AI.

If something can be done better, cheaper, and humans hate doing it in the first place, then bet the house that AI will disrupt it very soon. Cloud procurement is a perfect example of this paradigm.

The age of multi-cloud will phase out. Multi-platform is coming.

The time of hyperscalers fighting over customers by offering credits/pricing breaks and one time discounts is phasing out. The real 2.0 of “cloud hedging” is going to be focused around right sizing allocation of resources within applications architectures native to the choice cloud provider.

We are already seeing this play out in the battle of server vs serverless. Customers will start to make intelligent and automated decisions based on performance, manageability,  availability and pricing for applications to live in various architectures within a single hyperscale. Software will help them simulate results, and quickly make these pivots. Many customers I'm talking to now are already playing tug of war between serverless and server based architecture and most of the debate is around pricing.  This battle used to be between one hyperscaler to the next, however it's become increasingly more internalized between products within a single provider.

Storage will go through a similar evolution, as more data workflows won't be in traditional SQL databases. Mature data pipelines using a variety of protocols and storage products within hyperscalers are now much more commonplace and feeding a variety of services/microservices within internal/external applications.

All of this will need automated and AI driven cloud procurement to dynamically adjust to real usage changes in tech stacks.

Cloud FinOps adoption is exploding.

98% of enterprise CTOs want to have a FinOps practice, or grow their existing one to be more effective. The CxOs of tomorrow (and the ones today that survive) will have a deep understanding of cloud tech. Vanishing are the days of CFOs ordering up cloud migrations because they read a cool article on digital transformation. The companies that will be leading categories in 10 years will be founded by & led by technical folks that understand that if you’re still talking about digital transformation in 10 years you’ve already lost.

Finops’ current limiting factor today is that if your company is less than 500 people, you have 998 other companies for every 1 certified Cloud FinOps practitioner in this space.

Not everyone will have a Finops team but everyone will have a Finops practice. Most Finops practices in ten years will be software tools only.

Reserved Instances are outdated and wonky to manage. Automated and natively agile buying programs are the future.

RIs were released in 2009. Also released in 2009? Disney’s “Up” (still a classic), Kid Kudi’s “Man on the Moon” (still a classic) and the iPhone 3Gs (the one before Siri).

The problem with RIs? They are too focused on single machines(yes i know about normalization factors but that still takes someone sitting around with a calculator). No one running 500-1000 servers wants to manage 500-1000 coupons or figure out how 25 coupons can cover the 500-1000 machines. Not to mention the fees, risk, and management overhead if you want to sell your RIs. Although I love seeing 3rd parties that offer flexible RIs, the fees associated with this model still eat into the customer side savings, and thus will be open to disruption over time.

AWS has led the cloud buying innovation cycle for a long time. They released Savings Plans a couple years ago, and this is the future. Savings Plans allow a customer to consume any compute, anywhere, and manage nothing other than top line spend commitments. They remove 95% of the work in managing RI coupons, and drastically increase the value, agility & ease of management. Oh and you can toggle between server and serverless. Win.

At North we backend our savings as a service product with Savings Plans, and they’re awesome. GCP is also trying to catch up with Flex CUDs although much remains to be desired on that product in my humble opinion.

Anyone using RIs or relying on RIs is wasting time & energy on a medium that simply will be phased out by a better experience over time. Better buying programs like Savings Plans will open and foster the growing AI Cloud buying ecosystem, while allowing users what they really want, total compute flexibility with no geographical, or sub-product limitations.

AI can and will bridge the gap(s) between performance analytics, architecture and buying posture.

These mediums have lived in separate domains within the Finops and Devops landscapes for years, but will soon be conjoined using AI.

If you can use software and machine learning to train AI to understand what cloud licensing to buy, based on environmentals.. why wouldn't you?

People don’t want to look at dashboards all day long.

Dashboards are becoming the new spreadsheet. I can't tell you how many conversations I've had with DevOps leaders and CTOs that have complained about dashboard overload.

If your software only puts dashboards in front of customers and leaves the decisions to the user, is it really intelligent at all? AI is not refactoring data to make it more viewable; it's taking the source data out of the decision front end. People don’t want 30 dashboards and a decision, they just want the problem to go away.

North is a certified AWS partner and software provider focused on helping Cloud Native customers reduce cloud overspend. Our software automates up to a 72% reduction in AWS costs, as a service.