The ‘AI tax’ on AI-enabled applications in the cloud

Back in 2019, I wrote about the “container tax.” In simple terms, this is the additional cost to use containers properly within a cloud-based application. It includes development, operations, and other expenses that containers incur. The goal of leveraging containers is to offset the additional costs with the benefits they offer.

Many other technologies come with additional costs, which may or may not justify using that specific technology. The latest example is artificial intelligence in cloud-based applications. Companies should consider the additional costs of AI versus its potential value.

AI is nothing new but is going through a renaissance due to the popularity of generative AI platforms and the potential value of leveraging AI from within applications. We’ve built AI-enabled applications since the 1960s. Their value sometimes outweighs their costs and sometimes not.

The biggest issue with AI enablement is its overuse. For a time, AI was rarely used, primarily because it was expensive and didn’t deliver much value to offset the additional costs and risks.

Most AI engineers of the 1980s, including myself, are excited to see the capabilities of today’s generative AI engines such as ChatGPT. The cloud brought AI back onto platforms with many times the capabilities of past AI systems at drastically reduced prices. So far, so good, right?

At issue are the additional costs that need to be considered when using AI subsystems from within existing or net-new applications—in other words, the AI tax. In many cases, AI is being tossed into applications without considering its purpose or the value it can generate. Sometimes the value is easy to spot. Most times, it doesn’t cover the additional costs of AI-enabling a new or existing application. That’s where the trouble comes.

What are the additional costs of leveraging AI, and what needs to be understood before implementing it? Here are some basic AI “taxes” to consider:    

Infrastructure costs: Developing AI-based cloud solutions will require additional computing power and storage capabilities. The required investment in more powerful hardware and required services from cloud providers will increase costs outright and ongoing.

Data acquisition and preparation costs: You need high-quality data relevant to your use case to build effective AI models. Acquiring and preparing this data can be time-consuming and expensive, especially if you must collect data from multiple sources or clean and preprocess it to ensure accuracy.

Training costs: AI models require training with large amounts of data to learn how to make accurate predictions or decisions. Training AI models is a computationally intensive process that requires significant resources and thus, more money.

Maintenance costs: Once AI models are deployed, they must be monitored and maintained to ensure they continue to operate effectively. This requires ongoing updates, bug fixes, and performance tuning that all add to the overall costs of the solution.

Talent costs: Developing AI-based cloud solutions requires specialized skills and expertise, which may not be available in-house. Hiring or contracting AI experts is an expensive undertaking.

I don’t think that any of these “taxes” come as a surprise to most cloud architects or cloud engineers. We’ve known about them for years. The disconnect is often between how they exist within the context of specific applications and, most importantly, the potential value the application can return using AI.

In some cases, the returned value justifies the use of AI. In many more cases, the price tag to AI-enable a new or existing application does not make sense—at least, not yet. Much like our discussions around container taxes, we must have viable and justifiable business reasons to leverage AI.

Today, it seems everyone wants to add “AI experience” to their CV. That’s not a good reason to introduce AI into an application or organization. Trial and error at this level rarely benefits an organization or a career. Before you go too crazy with this stuff, take a breath and justify your AI vision with ROI facts and figures. You might be surprised by the results.