In case you somehow missed it, the generative AI revolution is upon us. Tools like ChatGPT are famously capable of producing everything from computer code to poetry, with impressive accuracy and sophistication.
But here’s one thing that generative AI is not capable of doing, at least not anytime soon: Helping to solve cloud spending woes. It would be nice if figuring out how to cost-optimize your cloud were as simple as asking an AI engine for tips, but that’s not happening—not because generative AI technologies aren’t impressive (they are) but because AI tools currently lack the ability to understand feedback loops and nuanced business context.
To prove the point, let’s look at what AI can do in the context of cloud cost optimization, then detail why it remains incapable of helping businesses conquer cloud spending challenges.
The long history of AI and cloud cost optimization
Although it has only been in the past half-year or so that generative AI began making flashy headlines, the cloud cost optimization ecosystem has been leveraging AI in other forms for quite a long time.
The fundamental process at the heart of most cloud cost optimization tools is predictive analytics, which is a form of AI and machine learning. For cost management purposes, predictive analytics involves deploying algorithms that assess cloud spending and workload performance data.
[ Also read Generative AI: 3 do’s and don’ts for AI leaders. ]
Those algorithms are trained to predict spending patterns. Cost optimization tools can also parse historical spending data to identify anomalies associated with overspending. They then surface the mistakes to help businesses save money in the cloud. This is how tools like AWS Compute Optimizer have worked for years.
To be sure, generative AI tools do more than just predictive analytics and anomaly detection, and they are trained on data sets much larger than those used by your standard cloud cost optimization tools. Still, at the end of the day, both categories of tools—generative AI technologies and traditional cloud cost optimizers—deliver insights in the same fundamental way, by parsing data sets using machine learning models. There’s nothing remarkably new in this respect about generative AI tools as compared to the cloud cost optimizers that have been delivering insights via predictive analytics for years.
The limitations of AI for reducing cloud spending
Given that the application of AI-powered technology to cloud cost optimization is hardly new, it’s hard to imagine the latest breed of AI tools faring any better than traditional solutions when it comes to helping businesses to reduce cloud spending.
The main reason why is that next-generation AI tools, as well as traditional cloud cost optimization software, are subject to some major limitations in their ability to understand cloud spending patterns and needs. The most important is that they can’t incorporate customer feedback in order to refine their recommendations over time. Instead, they provide generic, crude recommendations that might be in the interests of the typical business but that may or may not make sense for your particular company or for the workload it operates.
In other words, generative AI, like conventional cloud cost management tools, offers users no way of saying, “I’m not going to accept your recommendation to move this workload to a different EC2 instance because doing so would violate my business’s governance rules. Please make a new recommendation.”
You can’t explain critical context to AI tools and expect them to respond effectively.
Nor can you say, “I took the recommendation you gave me last month to modify the hosting configuration for a little-used application, but today our CMO announced that we’ll be promoting that application heavily and we expect traffic to pick up significantly. How should we update the configuration to reflect the changing business context of the app?”
You can’t explain critical context like this to AI tools and expect them to respond effectively.
Relatedly, AI tools often need human guidance to achieve their goals. For example, although an AI tool could parse workload configuration data and analyze performance and spending metrics, it likely can’t tell the difference between a testing workload and production workload unless you tell it how to determine the difference (by, for example, looking at workload tags). Nor can it understand how the needs of different business units, different groups or customers, varying budgetary priorities and so on impact cloud spending unless, again, you provide guidance.
The only way to work around these limitations of AI in the context of cloud cost management is to bring humans into the fold. You need humans who understand the nuances of cloud spending and who can identify and assess cloud savings recommendations based on the unique context of every workload and every business.
The future of AI and cloud cost management
It’s worth pointing out that there are ways that modern AI technology could evolve to address the shortcomings I’ve highlighted, at least in theory. Conceptually, it might be possible for the algorithms behind a tool like ChatGPT to accept human feedback about what works and what doesn’t with regard to cloud spending initiatives, then use the feedback to refine recommendations over time. It is also theoretically feasible for these tools to collect and analyze data, such as email streams, which give them insight into business context.
But currently, no generative AI tool is designed to do these things. And even if such tools did exist, there will always be some amount of business context and nuance that AI tools can’t learn by simply ingesting data. There will also always be some level of unknowability about a business’s cloud workload requirements,
So, while AI might evolve over time and become somewhat better at helping to manage cloud costs, it’s virtually impossible to imagine a world where AI alone gets you more than perhaps 50% of the way toward optimizing cloud spending.
Human oversight still required
There are many ways in which generative AI tools could potentially help to smooth over some of the rougher edges of cloud cost optimization. For example, finance teams could use natural language models to interpret jargon-heavy technical descriptions from engineers about what cloud workloads require. Or businesses could use AI tools to explain pricing structures for complicated cloud workloads in a way that humans can more easily understand. AI tools can also play a key role in bringing both technical and non-technical stakeholders up to speed quickly on a wide variety of topics, including helping to explain and support cost optimization recommendations.
That said, there is simply too much nuance and complexity in the discipline of cloud cost optimization ecosystem for AI tools to automate away cost management tasks entirely. They just can’t refine their recommendations well enough in a context-aware way to deliver actionable results without requiring humans to participate in the process or, at a minimum, provide critical guidance.
If they could, you’d expect that there would be AI tools out there that can optimize cloud costs just as well as they can write code or poetry. But there aren’t, and it’s hard to imagine there ever will be, because cloud cost management is just not as simple or repeatable as the types of tasks that AI can handle more effectively without human assistance.
[ Want best practices for AI workloads? Get the eBook: Top considerations for building a production-ready AI/ML environment. ]
