Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Cloud Contracts Now Penalize Unused Reserved Instances
Transcript
- Lucas: If these conversations are useful for what you're building or running, we keep this show ad-free by design. More on that in a moment. Luna: We've talked a lot about new fees in cloud contracts — data egress, API calls, GPU bonding. But there's one charge that's been quietly creeping into enterprise renewals that I think catches people off guard. Lucas: You're talking about penalties for not using reserved instances you've already committed to. I saw a term sheet last month from a mid-size SaaS company — they had a clause that charged them 15 percent of the unused value of their three-year AWS Reserved Instances if utilization dropped below 70 percent. Luna: Fifteen percent on unused value — that's not a small rounding error. How much did that end up costing them? Lucas: They had about 2.3 million dollars in annual RI commitments, and their utilization dropped to roughly 55 percent during a re-org. The penalty came out to about 340,000 dollars for Q2 2026 alone. And it was buried in an appendix — not in the main pricing table. Luna: So this is a shift from reserved instances being a discount vehicle to being a liability if you don't manage them. When did providers start pushing these clauses? Lucas: It's been building over the last 18 months. AWS started adding utilization minimums to Enterprise Discount Program renewals in late 2024. Azure followed with their Microsoft Customer Agreement templates. GCP has always had committed use discounts, but now they're adding true-up penalties if you fall below your committed amount by more than 10 percent. Luna: So it's not just one cloud — it's the whole big three. And I imagine the logic from the provider side is, 'We reserved capacity for you. If you don't use it, we lose money.' Lucas: Exactly. Cloud providers operate on utilization rates themselves — if a customer buys a reserved instance and runs it at 40 percent, that's compute they could have sold to someone else at on-demand rates. So they're essentially forcing customers to do better capacity planning. Luna: And we deliberately don't run ads on these conversations. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Now, back to the penalty structures — are these clauses standard across all tiers of customers? Lucas: Not yet. The big hyperscalers are mostly targeting customers with over 500,000 dollars in annual commit. But as this becomes standard, I think it'll trickle down. We're already seeing it in some enterprise-grade contracts for mid-market companies. Luna: What about convertible RIs versus standard ones? Does the penalty differ? Lucas: Good question. AWS Convertible RIs have more flexibility — you can change instance family, maybe even region. But the contracts I've seen often have higher utilization minimums for convertibles — sometimes 80 percent versus 70 for standard. The thinking is, if you have more flexibility, you should be able to keep utilization up. Luna: So you're paying a premium for flexibility, but that flexibility comes with a tighter leash. Lucas: Right. And Azure has a similar dynamic with their Reserved Virtual Machine Instances versus their savings plans for compute. The savings plan is more flexible, but the minimum commitment period is one or three years, and if you drop below your hourly commitment, you still pay — no penalty per se, but you don't get the discount on the shortfall. Luna: So the penalty is effectively the lost discount. That's a softer version than AWS's explicit penalty clause. Lucas: Exactly. And GCP's committed use discounts are the most straightforward — you commit to a certain dollar amount per month, and if you spend less, you pay the difference at on-demand rates. No additional penalty. But I've heard they're testing a penalty tier in some enterprise agreements. Luna: Let's talk about the practical side. If you're a cloud buyer listening to this, how do you audit your reservation portfolio to avoid these penalties? Lucas: First thing: look at your utilization reports. AWS has a Trusted Advisor that shows RI utilization. Azure has the Advisor recommendations. GCP has the Committed Use Discounts report in the console. Identify any reservations that are consistently below 70 percent. Luna: And then what? You can't just return them. Lucas: You can sell them on the AWS Reserved Instance Marketplace. That's one option — you offload the commitment to another AWS customer. But you might take a hit on the upfront cost. Another option is to modify the reservation — change the instance size or availability zone to better match your actual usage. But that only works within certain limits. Luna: Or you could consolidate workloads. I've seen companies shift batch jobs to run on underutilized reserved instances to boost utilization. Lucas: That's a great tactic. If you have a reservation for a c5.2xlarge that's running at 50 percent, you can schedule non-critical workloads on it during off-peak hours. Some finops teams call that 'sweating the assets.' Luna: And what about contract renewal timing? Should you renegotiate utilization minimums? Lucas: Absolutely. If you're doing a three-year renewal, negotiate a grace period — say, six months where the penalty doesn't apply, to give you time to adjust. Or ask for a lower utilization floor, like 60 percent instead of 70. The providers are willing to negotiate if you have a credible plan to improve. Luna: I think the bigger picture here is that cloud procurement is becoming more like traditional IT vendor management. You can't just sign and forget. Lucas: Exactly. The days of 'buy reserved instances, get a discount, ignore it' are over. These penalty clauses are a signal that providers want customers to treat capacity as a managed asset. And for companies that do it well, the discounts are still substantial — 30 to 60 percent off on-demand rates. But the risk is real. Luna: So the key is active management. Have you seen any tools that help automate this? Lucas: There are third-party finops platforms like CloudHealth, Spot by NetApp, and Vantage that track utilization and even automate instance modifications. But most of them require integration with your cloud accounts and some setup. The bigger enterprises are building their own scripts using the cloud providers' APIs. Luna: One thing I'm curious about: do these penalties apply to all types of reserved instances, or are there exceptions? Lucas: Good point. Some contracts carve out exceptions for test and development environments, or for instances that are stopped but still count as reserved. But the trend is to narrow those exceptions. I've seen clauses that require you to run instances for at least 8 hours a day to count toward utilization. Luna: So even if you stop an instance overnight, it might not count. That's aggressive. Lucas: It is. And it's pushing companies toward using more serverless and spot instances for variable workloads, reserving only the baseline capacity. That's probably the smartest strategy going forward. Luna: So the golden rule is: reserve only what you know you'll use at least 70 percent of the time, and use on-demand or spot for the rest. Lucas: Exactly. And if you're about to sign a new cloud contract, read the appendix on reserved instance utilization. That appendix could cost you hundreds of thousands if you're not careful. Luna: I think that's a great note to end on. Next episode, we'll look at how cloud contracts are starting to charge for idle load balancers. Lucas: That's a fun one. Load balancers sitting there doing nothing while the meter runs. See you next time.