Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / How Cloud Providers Lock You Into Expensive Contracts
Transcript
- Lucas: Let’s talk about the moment you realize your cloud bill is not just an expense, but a strategic liability. In late September 2026, the average enterprise cloud spend has climbed past ten percent of total operational costs for many tech-forward firms, and the growth isn’t slowing. Most leaders think this is just inflation or scaling pains, but I’ve been looking at the contract structures from Amazon Web Services, Microsoft Azure, and Google Cloud Platform, and what I’m seeing is less about usage and more about leverage. Luna: You mean they’re betting on you being too expensive to leave? That feels like a fair assessment of the current market dynamic. Lucas: Exactly. It’s not just that leaving is hard; it’s that staying becomes artificially cheap only if you commit deeply enough to trap yourself. Let’s take a specific example from a fintech client we worked with last quarter. They were running a legacy migration to Amazon Web Services and had signed a three-year commitment for compute capacity. On paper, they saved thirty percent compared to on-demand pricing. But when they needed to scale up for a new product launch, they found their reserved capacity was locked into older instance types that didn’t match their new AI workloads. Luna: So they paid for the discount but couldn’t actually use the right tools. That sounds like a classic mismatch between financial planning and engineering reality. Lucas: It’s worse than that. They ended up paying for the reserved instances they didn’t need, plus full on-demand rates for the new capacity. Their effective cost per transaction actually went up because they couldn’t shift resources fluidly. This is what we call the flexibility tax. When you sign those multi-year contracts, you’re trading agility for a percentage point reduction that rarely materializes if your workload is volatile. And volatility is the norm in today’s market. Luna: Right, because nobody knows if their next feature will be a niche tool or a platform-defining moment. So locking in capacity feels risky. Lucas: It is risky, yet every procurement team pushes for it. The problem is that the cloud providers have built their entire revenue model around predictability for themselves, not for you. Look at the data from the FOMC meetings earlier this year. Interest rates have stabilized, which means capital is cheaper, but cloud vendors aren’t lowering prices to reflect that stability. Instead, they’re increasing the complexity of their billing structures. Egress fees, inter-availability zone traffic charges, API request costs—these are all designed to make your data stay put. Luna: Data gravity is becoming a prison sentence rather than a natural law. How do companies even measure that friction? Lucas: Most don’t. They look at the headline number for compute hours and ignore the network layer. Let’s say you move data from one region to another within the same provider. If you’re using Amazon Web Services, that traffic between availability zones can cost up to four cents per gigabyte. If you’re moving terabytes daily, that adds up to thousands of dollars a month purely for internal communication. And if you want to pull that data out to another cloud or for disaster recovery, the egress fee jumps to twelve cents per gigabyte. Twelve cents might sound small, but for a video streaming service or a large language model training dataset, it’s a margin killer. Luna: Twelve cents per gigabyte is huge if you’re moving petabytes. That’s basically a toll booth on your own infrastructure. Lucas: Precisely. And here’s the twist: most engineers don’t control these settings. They’re configured at the account level by ops teams who are under pressure to minimize compute costs. So you optimize the wrong thing. You cut your EC2 instances or Azure VMs by half, thinking you’re saving money, while your data transfer bill doubles because everything is now churning through different zones to find each other. It’s a zero-sum game where the provider wins either way. Luna: So the incentive structure is misaligned. You save on compute but lose on network, and the provider keeps both streams flowing. Lucas: That’s the core issue. Now, let’s look at how some forward-thinking teams are pushing back. There’s a growing movement called FinOps, which stands for Financial Operations. It’s not just about cutting costs; it’s about unit economics. One company we tracked, a healthcare logistics firm, stopped buying reserved instances entirely. Instead, they used a hybrid model. They kept a small baseline of reserved capacity for predictable, steady-state workloads—like their database servers—and shifted everything else to spot instances or serverless functions. Luna: Spot instances can be interrupted though. Isn’t that a nightmare for reliability if the job gets killed halfway through? Lucas: It was, until recently. Cloud providers have improved their spot instance resilience significantly. Amazon Web Services now offers spot fleets that can automatically replenish interrupted instances across different availability zones. For batch processing, image rendering, and CI/CD pipelines, the cost savings can be seventy percent or more. The key is designing your software to be fault-tolerant. If your application crashes when an instance disappears, you weren’t really building for the cloud anyway. You were just renting a server. Luna: Fault tolerance becomes a financial asset then. That’s a reframing I haven’t heard often enough. Lucas: It should be. But there’s another layer: the contract negotiation itself. Many enterprises sign annual agreements without ever opening the fine print. The new trend is modular commitments. Instead of promising to spend fifty million dollars over three years, companies are negotiating flexible pools. You commit to a certain amount of compute, but you can shift that commitment between different services—say, from general-purpose servers to GPU clusters for AI training—without penalty. Microsoft Azure has started offering these kinds of dynamic reservations, and Google Cloud is following suit. Luna: So you’re paying for the potential capacity rather than the specific hardware. That makes much more sense for R&D heavy industries. Lucas: Exactly. And this brings us to the bigger picture. In September 2026, the race for AI dominance is driving massive infrastructure demand. Everyone wants GPUs, and everyone is bidding against each other. If you lock yourself into a rigid contract today, you might find yourself unable to access the hardware you need six months later when the technology shifts again. Flexibility is no longer a luxury; it’s a survival mechanism. Companies that treat cloud spend as a fixed utility bill are falling behind those who treat it as a variable investment portfolio. Luna: Treating it like an investment portfolio means constantly rebalancing. That requires a lot of discipline and visibility. Lucas: It does. And that’s why tooling matters. You can’t manage what you can’t see. The best teams use automated tagging strategies. Every resource—every virtual machine, every database, every storage bucket—is tagged with a project code, a cost center, and an owner. When the bill comes in, you don’t see a lump sum; you see a detailed breakdown of who is spending what and why. Without that granularity, you’re flying blind. You might catch a spike in costs, but you won’t know if it was a developer testing a new algorithm or a runaway script eating up memory. Luna: Tagging seems simple, but enforcing it across hundreds of engineers must be a cultural battle as much as a technical one. Lucas: It absolutely is. Engineers hate overhead. They want to build features, not fill out metadata fields. The solution is to bake tagging into the deployment pipeline. If you try to spin up a server without a valid project tag, the system rejects it. It sounds harsh, but it forces accountability from day one. Once that habit is formed, the financial data becomes clean, and then you can start doing real analysis. You can identify idle resources, decommission unused databases, and right-size oversized instances. These are low-hanging fruit that typically shave fifteen to twenty percent off the initial bill. Luna: Fifteen percent is life-changing for a startup’s runway. It’s not just about efficiency; it’s about longevity. Lucas: It is. And let’s not forget about storage tiers. Most companies store all their data in high-performance SSDs, even if ninety percent of it is cold data that hasn’t been accessed in two years. Moving that data to archival storage classes can reduce costs by eighty percent. Amazon Web Services Glacier, Azure Archive Storage, Google Coldline—these are designed for long-term retention. The trade-off is retrieval time. If you need to read the file instantly, it might take hours to restore. But if you’re just keeping it for compliance or backup, the savings are enormous. Luna: So you’re matching the storage speed to the business need, not just picking the fastest option by default. That’s basic supply chain logic applied to IT. Lucas: Exactly. Supply chain logic. Treat your data like inventory. You don’t keep fast-moving goods in a warehouse across the country if you can store them locally. Similarly, you don’t keep cold data in premium storage. The mental model shift is critical. It stops cloud spending from being a black box and turns it into a managed resource stream. Luna: I love that analogy. Managed resource stream. It makes it feel less like a bill and more like a supply chain challenge. Lucas: It should. And as we wrap up this deep dive into cloud economics, I want to pause for a moment. If today’s conversation helped you rethink how you approach infrastructure spend or gave you a new angle for your next budget review, there’s a simple way to support the show. These stays ad-free thanks to listener support. If you find value in this kind of practical, deep-dive analysis, consider buying me a coffee at buy me a coffee dot com slash fexingo. It helps keep the research sharp and the conversations honest. Luna: Yeah, and it lets us keep digging into these structural issues instead of chasing clicks. Really appreciate it. Lucas: Thanks, Luna. Back to the topic. Let’s look at one final tactic that’s gaining traction: multi-cloud redundancy done right. Too many companies adopt multi-cloud for fear of vendor lock-in, but they do it poorly. They replicate everything in duplicate, doubling their costs. A smarter approach is active-passive redundancy. You run your primary workload in one cloud and keep a lightweight standby environment in another. You pay minimal costs for the standby state, and if something goes wrong, you fail over quickly. This gives you insurance without the full price of duplication. Luna: Insurance is a good word for it. You’re paying premiums for resilience, not necessarily for daily operations. Lucas: Precisely. And that brings us to the future. As AI models grow larger, the energy and compute requirements will explode. Cloud providers are already struggling with power constraints. We’ll likely see a shift toward edge computing, where data is processed closer to the source to reduce latency and bandwidth costs. Companies that start experimenting with edge architectures now will be ahead of the curve. The cloud isn’t disappearing; it’s fragmenting. Your job is to figure out where each piece of your stack belongs. Luna: Fragmentation sounds chaotic, but if it drives efficiency, it might be the necessary evolution of infrastructure. Lucas: It is. The question for listeners is: where is your data currently living, and is it paying rent for the space it doesn’t need? Audit your tags, review your storage tiers, and renegotiate those contracts with fresh eyes. The leverage is yours if you’re willing to walk away. Luna: Walking away is the ultimate leverage. Thanks for breaking down the mechanics of that today. Lucas: Anytime. Keep building smart, and we’ll see you next time.