Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Costs Surge When Data Gets Hot
Transcript
- Lucas: So you've got a cloud bill that seems to creep up every month, and you've already turned off idle instances and reserved capacity. But the bill still climbs. One culprit that flies under the radar: the temperature of your data. Luna: Temperature — as in hot data versus cold data? Lucas: Exactly. Every cloud provider offers storage tiers based on how often you access data. Amazon's S3 has Standard — that's hot — then Infrequent Access, then Glacier for archive. Azure has Hot, Cool, Cold, and Archive tiers. Google has Standard, Nearline, Coldline, and Archive. Luna: And the price difference is dramatic, right? Hot storage costs more per gigabyte but no retrieval fees, while cold storage is cheap to store but you pay to get data back out. Lucas: Right. But here's the trap: enterprises often default to hot tiers for everything because it's the easiest path at migration. They don't think about access patterns. A 2025 IDC report found that 35 percent of enterprise cloud storage spend is wasted on data stored in hot tiers that is accessed less than once a quarter. Luna: One out of every three dollars on storage — just thrown away. Lucas: Let me give you a concrete example. I spoke with the cloud architect at a Fortune 500 retailer — they had 200 terabytes of web server logs in S3 Standard. Those logs were from 2019. Nobody had touched them in three years. They moved the entire set to S3 Glacier Deep Archive, which costs about one-ninth the price per gigabyte. Luna: What was the impact on their monthly bill? Lucas: They cut storage costs for that data by 68 percent. That's about $40,000 a month saved. And the retrieval cost? They've only needed to access those logs once in the past year for a compliance audit. That retrieval cost them maybe $300. Luna: Forty thousand versus three hundred — that math is hard to ignore. Lucas: So why don't more enterprises do this? Two reasons. First, they don't have a data classification strategy. Second, they're afraid that moving data between tiers will break applications that might need it. Luna: But cloud providers have lifecycle policies that automate this. You can set rules like 'if an object hasn't been accessed in 30 days, move it to Infrequent Access; after 90 days, move to Glacier.' Lucas: That's exactly the solution. AWS has S3 Lifecycle Policies, Azure has Blob Storage lifecycle management, Google has Object Lifecycle Management. The tricky part is setting the right thresholds. You don't want to move data prematurely and then pay retrieval fees when an app needs it. Luna: So you need to understand your actual access patterns first. Lucas: Yes. And this gets more complex with AI workloads. Training data for machine learning models — that's hot data. You need GPU clusters pulling that data constantly. But once the model is trained, the training dataset can often be moved to colder tiers. Inference data — logs, predictions — that can live in cold storage. Luna: But AI teams often push back. They want everything hot because they might need to retrain at any moment. Lucas: And that's a valid concern. But you can set up a pipeline that automatically rehydrates data from cold to hot if a retraining job is triggered. AWS S3 Batch Operations can do this. You just need to bake it into your MLOps workflow. Luna: So the key is to have a policy that's dynamic, not static. Not just 'move after X days' but 'move unless Y happens'. Lucas: Exactly. And the best practice I've seen is to run a quarterly storage audit. Use tools like AWS Storage Lens or Azure Storage Insights to analyze access patterns. Then update your lifecycle rules accordingly. Luna: Speaking of audits — a lot of enterprises I talk to don't even realize they have data in multiple tiers across different accounts. They have no centralized view. Lucas: That's a real problem. I'd recommend using a cloud management platform like CloudHealth or Vantage to get a single pane of glass. Or even just a simple script that tags every bucket with its tier and last access date. Luna: Tags are huge. I've seen companies save 20 percent just by tagging data with 'retention: 90 days' and then automating deletion. Lucas: Right, but deletion is a different conversation. Today we're talking about tiering — keeping data, just in the right place. And the savings are real. If you're running a $5 million monthly cloud bill, storage is typically 20 to 30 percent of that. So a 30 percent optimization on storage is $300,000 to $450,000 a year. Luna: And that's just from one policy change. No new architecture, no application rewrite. Lucas: Exactly. So my takeaway: don't let your data temperature be a surprise. Audit your storage tiers this quarter. Set up lifecycle policies. And if you're building AI pipelines, design for hot and cold from day one. Luna: If this kind of concrete, no-fluff analysis helps you in your work, you know what keeps it coming — listener support. It's a simple way to keep these conversations ad-free and focused on what actually matters for operators and builders. Lucas: Yeah, we really appreciate that. If you found today useful, you can buy us a coffee at buy me a coffee dot com slash fexingo. It makes a difference. Luna: And on that note — next week we're looking at a related trap: why egress fees are still the silent budget killer, even with all the recent price cuts. Lucas: That's a good one. Thanks for listening, and we'll catch you next time.