Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / How Cloud Contracts Now Guarantee Instance Availability
Transcript
- Lucas: So there's a quiet revolution happening in cloud contracts, and it's not about AI compliance or carbon credits. It's about something much more basic: whether you can actually get the instance you need when you need it. Luna: You mean the GPU shortage that's been going on for years? That's still a thing? Lucas: It is, but it's evolved. The shortages of 2024 and 2025 were acute — everyone remembers scrambling for A100s and H100s. What's changed now is that hyperscalers are starting to put contractual guarantees around instance availability into their enterprise agreements. Luna: Wait — so AWS or Azure will actually commit in writing that you'll get, say, 100 H100s next Tuesday at 2 PM? Lucas: Exactly that. And they'll pay a penalty if they can't deliver. It's a massive shift from the old model where you'd request capacity and hope for the best. Now, especially for high-value workloads like AI training or financial simulation, buyers are demanding firm commitments. Luna: So what triggered this? Was it just the GPU shortage, or something else? Lucas: The GPU crunch was definitely the catalyst. But it's broader than that. A lot of enterprise workloads need specific instance types — memory-optimized instances for in-memory databases, compute-optimized for batch processing. And when you're running a critical pipeline, you can't afford to be told 'sorry, we're out of capacity' and have to fall back to a slower instance. Luna: Right, so the contract becomes a risk management tool. Who's pushing hardest for these clauses? Lucas: Two groups, primarily. First, hedge funds and trading firms running time-sensitive quantitative models. If their backtest window closes because the cloud is maxed out, that's real money lost. Second, biotech companies doing drug discovery — they have these massive molecular dynamics simulations that need a specific GPU count for a specific duration, and a disruption can set their timelines back months. Luna: And the hyperscalers are actually agreeing to this? I'd think they'd push back because capacity planning is hard. Lucas: They are agreeing, but with conditions. The typical clause works like this: you commit to a minimum monthly spend, say $500,000, and in return the cloud provider guarantees availability of a defined set of instance types within a given region and availability zone. The guarantee usually applies during business hours or a defined 'critical window' — not 24/7. And the penalty for failing to deliver is typically a service credit of 10 to 30 percent of the affected workload's cost. Luna: So it's not a free ride. You're paying a premium for that certainty. Lucas: You are. And the premium varies. For standard instances, it might be a 5 to 10 percent uplift over on-demand pricing. For GPU instances, especially the latest generation, the premium can be 20 to 30 percent. But for the buyers I mentioned, that's a cost they're willing to bear because the alternative — a pipeline stalling — is far more expensive. Luna: How do the providers actually enforce this on their end? Like, do they have to keep a pool of cold servers sitting around? Lucas: That's the interesting operational challenge. They're effectively creating a new class of capacity — call it 'contractually reserved' — that sits between on-demand and reserved instances. They have to hold that capacity, which means they're betting that the customer will actually use it. There are true-up penalties if you under-utilize. So it's a two-way commitment: the provider guarantees availability, and the customer guarantees consumption. Luna: So it's almost like a forward contract for compute. I can see why hedge funds would like that. Lucas: Exactly. And it's not just hyperscalers. Some of the gpu specialized cloud providers, like CoreWeave or Lambda, have been doing this for a while. The difference is that now AWS, Azure, and GCP are baking it into their standard enterprise agreement frameworks. I've seen this in the last six to nine months. Luna: What about multi-cloud? If you have workloads spread across three providers, can you get a unified availability guarantee? Lucas: That's the next frontier. Right now, each provider's guarantee is siloed. But some large customers are starting to push for what's called a 'portfolio availability SLA' — where the provider commits that across their entire global footprint, the customer can always spin up a certain number of instances. That's much harder to fulfill, but we're seeing pilot agreements with big financial institutions. Luna: And what about the small and medium businesses? Are these clauses only for the whales? Lucas: For now, mostly yes. The minimum spend thresholds are still high — typically hundreds of thousands of dollars per month. But I think over time, as the practice matures, we'll see tiered offerings. Maybe a 'standard' availability guarantee for anyone with a one-year commit, and a 'premium' guarantee for larger commitments. Luna: So the cloud is becoming less of a utility model and more of a partnership, where both sides have skin in the game. Interesting. Lucas: Right. It's a recognition that compute is a finite resource, especially at the high end. And if you're building a business on top of that resource, you need certainty. The days of treating cloud capacity as infinite are over. Luna: That is a really useful lens for thinking about cloud procurement right now. And if this kind of practical analysis is helpful for what you're building or running, I want to mention that the way this show stays ad-free is listener support. You can buy us a coffee at buy me a coffee dot com slash fexingo. No pressure, just a way to keep these conversations going. Lucas: Yeah, and honestly, the feedback we get is that episodes like this one — drilling into a specific contract clause — are exactly what our audience wants. So if that resonates, the coffee link is there. Now, back to the availability guarantee — one thing I didn't mention is how this is affecting cloud architecture decisions. Luna: How so? Lucas: Companies that have these guarantees are now more willing to build tightly coupled, stateful workloads on cloud, because they know the underlying capacity won't vanish. Before, they'd design for failure — stateless apps, frequent checkpointing, fallback to cheaper instances. Now, they can optimize for performance and cost, because the availability risk is priced in and managed. Luna: So the contract clause is actually changing the architecture. That's a pretty big deal. Lucas: It is. It's a sign that cloud is maturing from an experimental platform to a core part of enterprise infrastructure, with all the contractual rigor that comes with that. And I think we'll see more of these 'guarantee' clauses — for network latency, for data durability, for support response times. The availability guarantee is just the first. Luna: Well, that's a great note to end on. Thanks for breaking that down, Lucas. Lucas: Always fun. See you next time.