The Hidden Costs of Public Clouds: Data Transfer

You've done your homework. With the help of your hyperscalers' pricing calculators, you’ve pinpointed what seems like the ideal option for your existing workflow. You’ve even secured enough cloud credits to cover your first 14-18 months of your cloud setup. But then, six months into the implementation, you’re staring at a bill that's double – or triple – what you anticipated. What went wrong?

The tough reality of public clouds is that they’re rife with difficult-to-anticipate costs that can take even the most expert IT professional by surprise. The good news is that these costs can be mitigated – and in many cases eliminated – with the right multicloud strategies. 

But before addressing these challenges, it’s crucial to understand them. In that spirit, I’ll start by diving into one of the more common causes of unanticipated cloud costs: Data transfer.

Widespread Data Transfer Confusion

Operating data is meant to move – whether that’s sending your data out for API calls, or transferring data across regions within your own cloud stack. Particularly when it comes to moving data out – data egress – this transfer isn’t free. Per two widely-quoted estimates, Gartner has found that between 10 and 15% of cloud spend goes to egress charges; while IDC finds that egress charges account for 6% of organizations' cloud storage costs.  

But data egress is not only a major source of cost -  it’s also a major source of confusion. 

I love this chart above from the DuckBill group as it does a great job of visualizing all of the AWS services that can add to a data transfer tax  - often one you have to pay many times over. 

Confusion Source 1: Provisioning Estimates vs. Pricing Realities 

For many firms, data transfer cost confusion can start even before the contract is signed. When teams run cost estimates through the major hyperscalers’ pricing calculators, many don’t realize that these estimates are for data provisioning costs like virtual machine compute, storage, and database instances only. The cost of data transfer, by contrast, is often placed on separate pages and in “the fine print.” While public data on this issue isn’t widely available, in my own conversations across the industry, this is an issue that comes up a lot. Teams plan their cloud credits against storage costs alone, while they’re eating away at their credits through data egress. They only find out once they get hit with the bill.

Confusion Source 2: Egress Unpredictability

Even if price calculators put all potential costs up front, actual costs can be slippery to estimate in advance – because data usage itself can be hard to estimate. Per a global survey of 1,600 IT decisionmakers by research agency Vanson Bourne on behalf of Wasabi Networks, 42% of organizations have found that they migrated more applications and data than originally planned. 

It’s also not hard to picture how this kind of underestimate plays out in a cloud computing context. All it takes is an app to grow a user base faster, or to get more usage than initially anticipated – and you’re potentially looking at astronomically more API calls. Costs go up accordingly.

Of course, AI is only making planning data transfer costs trickier to predict.

How AI is Exacerbating the Problem

In non-AI applications, there’s a fairly direct correlation between app popularity and API usage: more users, or more use, means more API calls to downstream systems to process data. But AI can make the math even more complex – and unpredictable. That’s because the data usage varies based on the complexity of the questions users are asking, and the length of the response. A chatbot greeting "Hi, how are you today?" will cost the AI developer more tokens than a curt “Hi.” Answering a tough question will cost more than a simple one – and of course, it can be tricky to understand what kinds of a questions an AI will find hard to begin with. AI asks humans to be their most human when interacting with human-like machines – which means a lot of unpredictability, and a lot of unpredictable data transfer costs.

Bringing Predictability to Data Transfer Costs

How can cloud users get a handle on data transfer costs – both bringing them down and minimizing the surprise? I recommend three key steps.

• Read the fine print on pricing calculators. Be sure you’ve read all the calculations the price calculators provide – not just the main page.

• Tier your cloud setup, moving the most data-intensive pieces of your architecture to the hyperscalers, and working with smaller and niche providers for pieces of your workflow that can do just fine with less data because, for instance, high speed is less of the essence. For instance, at FluidCloud we use a hyperscaler for production, but a smaller cloud for dev and staging.

• Work with FluidCloud to use our Cloud CloningTM technology for a precise, transparent estimation of true cloud costs. Talk to the FluidCloud team to get started.

Watch this space as we have a lot more to say about visibility, transparency and traversing a multicloud landscape so that you can lift the veil on cloud economics.