Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Bills Now Charge for Data Retrieval Requests
Transcript
- Lucas: So there's a new line item showing up on a lot of cloud bills this year, and it's not for data at rest, not for compute, not even for egress. It's for the simple act of asking for your data back. Luna: Asking for your data back? You mean retrieval fees? Those have been around for archive tiers for ages, haven't they? Lucas: Right, but historically they were buried in cold storage products like Glacier and Azure Archive. Now the big three are introducing per-request retrieval charges on their standard and even frequently accessed tiers. I've been tracking this since the start of 2026, and the shift is real. Luna: So it's not just for deep archives anymore. This is hitting everyday buckets and object stores. What's driving it? Lucas: Providers argue it's about aligning costs with actual infrastructure usage. Every GET request hits their metadata layer, consumes bandwidth, and spins up disk seeks. The old flat-rate model didn't capture that. So they're metering differently now. Luna: And that sounds reasonable on the surface, but it changes the calculus for anyone building data-heavy applications. Lucas: Exactly. Let's get specific. I looked at a retail analytics platform, let's call it ShopMetrics, that runs on AWS. They had a standard storage tier, and their bill suddenly showed a line for 'retrieval requests' at fifty basis points per thousand requests. They were doing about 40 million requests a month, so that's an extra $20,000 a year. Luna: Twenty grand for just asking for data? That's not nothing. And I'm guessing they didn't sign up expecting that. Lucas: Their contract said 'per-request fees may apply,' but it was buried in a service agreement update. When the bill jumped 18 percent month over month, their CFO noticed. Luna: So what are the actual rates across the big three? Is it the same model? Lucas: Not exactly. AWS charges per thousand requests on standard tiers, similar to ShopMetrics. Azure has a two-part structure: a per-request charge and a data access fee per gigabyte read. GCP has been the most aggressive, introducing retrieval fees even on regional buckets. Luna: Interesting. So it's not just a per-call thing; some providers also charge for the volume of data you pull. Lucas: Right, and that's the double whammy. With Azure, if you're running a big batch job that reads terabytes, you're paying per API call and per gigabyte. It can add up faster than you'd think. Luna: So what's the practical advice for a company like ShopMetrics? They can't just stop retrieving data. Lucas: The first thing is to look at access patterns. Are you reading the same hot objects repeatedly? If so, a caching layer or a CDN can absorb a lot of those GET requests before they ever hit the object store. Luna: Makes sense. Caching is the obvious answer, but what about for analytical workloads where you're scanning large datasets? Lucas: For analytics, the trick is to minimize redundant reads. Instead of running the same query ten times a day, you can materialize the results into a separate table or even a different storage tier. You're trading compute for retrieval fees. Luna: So you're saying you might shift to more compute-heavy approaches to avoid retrieval costs? Lucas: Exactly. And that's actually the broader trend. As retrieval fees spread, the cost of compute relative to storage access is changing. You might find it cheaper to preprocess data and store only the outputs. Luna: It's a whole new optimization problem. But I'm curious, how are providers justifying this to customers? Is it flying under the radar? Lucas: It's interesting. The cloud providers are framing this as a way to make pricing more 'granular' and 'cost reflective.' They say it encourages efficient use of storage and reduces waste. And honestly, there is some merit to that. Luna: There's a grain of truth, but it also makes it harder to predict your bill. For a small startup, a sudden 18 percent increase can be brutal. Lucas: That's the key tension. And I've seen this pattern before with egress fees. It starts small, then becomes the norm. I wouldn't be surprised if we see retrieval fees become a standard line item on every cloud bill by 2027. Luna: And just when you thought you had your cloud costs under control. But you know, before we go deeper into how to mitigate this, I want to step back for a second. A lot of what we talk about on this show comes from you digging into these obscure billing changes and making them make sense for our listeners. Lucas: Yeah, it's a lot of research and reading through service agreements that are dense with legalese. And honestly, I'm able to keep doing this because a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo. That's literally what funds making this many of these episodes. Luna: And we're genuinely grateful for that. It keeps us ad-free and independent, so we can cover topics like this without any sponsor pressure. Lucas: If you find these conversations useful for what you're building or running, and you want to see more, that's the way to support us. No pressure, just something to keep in mind. Luna: Right. So back to retrieval fees — what should a company do if they're already seeing these charges and they're not prepared? Lucas: First, audit your access logs. You need to see which buckets are generating the most GET requests and how much data you're actually pulling. That gives you the baseline. Luna: Then what? Is it just about trimming requests, or are there specific architectural changes? Lucas: There are several levers. One is lifecycle policies: if you have objects that are rarely accessed but still on standard tier, move them to a colder tier where retrieval fees might be lower or the per-request charge is waived entirely. Luna: So you're saying the archives actually become more attractive again, even with the retriever fees they always had? Lucas: Exactly. The calculation shifts. If your access pattern is truly infrequent, the cold tier might now be cheaper than the standard tier, once you account for the new retrieval charges. Luna: And for teams that can't move data around easily, what's the alternative? Lucas: You can also consolidate your reads. Instead of having twenty microservices each pulling from the same bucket, you can have one service that fetches the data and caches it internally. That reduces the number of unique requests. Luna: So it's about being intentional about how you access data, not just how much you store. Lucas: Precisely. And I think that's the core message: cloud costs are no longer just about what you store or compute; they're about how you interact with your infrastructure. Every API call now has a price tag. Luna: That's a big mental shift for a lot of engineers. It's not just about optimizing code; it's about optimizing your billable interactions. Lucas: And that's why I think this is a bigger story than just a new fee. It's about the evolution of cloud pricing models overall. We've seen compute get metered, then egress, now storage access. The next frontier might be metering API calls themselves, which were once free. Luna: You mean like charging for every time you call a cloud service's API, even just to list your buckets? Lucas: Something like that. Some providers are already experimenting with 'control plane' request fees, though they haven't rolled out widely. But it's a trend to watch. Luna: So as a cloud consumer, what's the best way to stay ahead of these changes? Lucas: Stay informed. Read the service announcements, not just the billing pages. And build a FinOps practice that includes monitoring these new metrics. The days of 'set it and forget it' with cloud storage are over. Luna: And for the finance folks listening, is there a way to model these costs before they hit? Lucas: Absolutely. You can estimate your retrieval volume from logs and apply the published rates. We've actually put together a simple spreadsheet template that listeners can grab from our show notes. It's a start. Luna: That's helpful. So, summing up: retrieval fees are here to stay, they're spreading beyond archives, and the key is to understand your access patterns and adjust accordingly. Lucas: And remember, you're not helpless. You have options: caching, lifecycle policies, and redesigning your data flows. The cloud is still cheaper than running your own data centers, but it's not a blank check anymore. Luna: It's a new era of accountability. I'm just glad we're here to help people navigate it. Lucas: Absolutely. And if you want to support us, you know where to find us. Otherwise, keep an eye on your next bill — you might just see a new line item.