Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Contracts Now Bundle Networking and Compute
Transcript
- Lucas: You've been hearing a lot about cloud providers competing on AI compute, GPU access, all that. But there's a quieter shift happening in how these contracts are structured — and it might matter more to your bottom line than which chip you're renting. Luna: Okay, you've got my attention. What kind of shift? Lucas: Bundling. Specifically, bundling networking and compute into a single commitment. So instead of signing separate agreements for virtual machines and data transfer, providers like AWS, Azure, and Google Cloud are increasingly offering one contract that locks in both GPU clusters and guaranteed network bandwidth. Luna: That sounds convenient. But also like a way to make it harder to mix and match providers. Lucas: Exactly. And that's the tension. On one hand, for workloads that are extremely latency-sensitive — think real-time AI inference or high-frequency trading — having a committed network path with guaranteed throughput can be a game-changer. On the other hand, you're tying yourself to one provider's ecosystem more tightly. Luna: Is there a real example of a company that went for this? Or is it still mostly theoretical? Lucas: There's a concrete case I saw recently. A Fortune 500 financial services firm — I won't name them because the deal is still under NDA — but they were running a large-scale fraud detection model that needed sub-millisecond responses. Their cloud bill was ballooning because of data egress charges between compute regions. Luna: So they were spreading the workload across multiple regions to get lower latency, and paying for it in egress. Lucas: Right. Their broker went to Azure and negotiated a three-year deal that bundled a fixed number of GPU instances with a dedicated network pipe between two regions. The result? They cut their annual cloud spend by 18 percent — and their inference latency dropped by 40 percent. Luna: Eighteen percent is real money. But I'm curious — what did they give up in return? Because that kind of discount usually comes with strings. Lucas: Great question. They committed to a minimum of 80 percent of that compute capacity being used at all times. If they dropped below, they'd still pay for the reservation. And they agreed not to run any of that workload on another cloud for the duration of the contract. Luna: So effectively, they're locked into Azure for that specific workload for three years. That's fine if the technology stays competitive, but what if Google or AWS releases a better GPU architecture next year? Lucas: That's the risk. And it's why you're seeing more enterprises bring in third-party brokers to structure these bundles. The brokers run the math — they model different scenarios, break-even points, and exit clauses. Some contracts now include a 'technology refresh' clause that lets you renegotiate if a significantly better compute option comes along. Luna: Interesting. So the bundle isn't necessarily a straightjacket if you negotiate well. But I imagine smaller companies don't have the leverage to demand those clauses. Lucas: That's the tricky part. The hyperscalers are rolling out these bundled contracts primarily for large enterprises with multi million dollar commitments. If you're a startup spending a few hundred thousand a year, you're probably still on standard pay as you go or a simple one-year reserved instance. Luna: But even for startups, the concept matters. Because the underlying driver — network cost being a bigger share of the total bill than compute — applies at any scale. Lucas: Exactly. And that's what's really changing. Historically, cloud pricing was compute-centric. You optimized for CPU or GPU hours. Now, with AI workloads moving huge amounts of data between training clusters and inference endpoints, networking can account for 30 to 40 percent of the total cost. Luna: So bundling is really a response to that shift in cost structure. The providers are saying, 'We'll give you a discount on the whole thing, but you have to commit to both.' Lucas: Precisely. And it's not just Azure. AWS has something they call 'Network Optimized Compute Reservations' — at least that's the internal name. And Google Cloud offers 'Committed Use Contracts with Network Tier' that bundle premium-tier networking. They're all converging on the same idea. Luna: Speaking of Google — I remember in a previous episode we talked about how Google runs its own network and that changes the math. Does that give them an advantage in these bundles? Lucas: It does, because Google's internal network is massive and they control the entire stack. They can offer very specific latency SLAs that AWS and Azure have to piece together from different backbone providers. But the catch is that Google's bundles tend to require longer commitments — sometimes five years — to get the best rates. Luna: Five years in cloud time is an eternity. That's a real bet on their architecture. Lucas: It is. But for some workloads, especially if you're training a foundation model that'll be in production for years, it might make sense. The key is understanding your own workload's trajectory — and having the data to back it up. Luna: Let me ask about the flip side. If bundling reduces flexibility, could it push more companies toward a multi-cloud strategy just to avoid being locked into one bundle? Lucas: It could. But ironically, the bundles themselves often include discounts for 'single-cloud concentration' — meaning you get a bigger discount if you run more workloads on that one cloud. So there's a tension. The economics of the bundle might pull you toward single-cloud, while the risk management of lock-in pushes you toward multi-cloud. Luna: That's the classic enterprise dilemma. And I think that's why we're seeing the rise of cloud brokers who specialize in modeling these trade-offs. It's become too complex for most internal finance teams. Lucas: Absolutely. And you know, conversations like this — unpacking the real mechanics behind the cloud pricing headlines — are exactly the kind of analysis that makes this show valuable to operators and builders. If you find these deep dives useful for what you're building or running, it's actually listener support that keeps this show ad-free and independent. Luna: Yeah, we don't have sponsors pushing a particular cloud provider's narrative. So if the show helps you make smarter infrastructure decisions, you can support it at buy me a coffee dot com slash fexingo. Lucas: And every bit helps us keep digging into contract clauses that most people never see. Now, back to the bundle angle — there's one more nuance I want to bring up. Luna: Please do. Lucas: Some of these bundles now include 'network capacity guarantees' that go beyond just speed. They guarantee that your traffic won't be preempted by other customers' traffic during peak times. That's a big deal for real-time applications. Luna: So it's not just about cost — it's about performance predictability. That could be worth more than the discount itself for mission-critical workloads. Lucas: Right. And the providers are starting to offer different tiers of these bundles. For example, Azure has a 'Performance Plus' tier that guarantees no packet loss and sub-100 microsecond latency between specific region pairs. It's expensive, but for certain financial services or autonomous driving workloads, it might be a must-have. Luna: Are these tiers standardized? Or is it still very much a bespoke negotiation? Lucas: Mostly bespoke. The listed prices are just starting points. The real deal happens when you bring in your workload profile and let the provider's solutions architect model it. And that's where having a good broker or an internal cloud center of excellence pays off. Luna: So the takeaway for our listeners: if you're about to sign a big cloud contract, ask specifically about network-compute bundles. Don't assume the standard pricing is the best you can get. Lucas: Exactly. And if you're already locked into a contract, check if there's a mid-term renegotiation clause. Some providers are open to restructuring mid-contract, especially if your usage patterns have changed. Luna: Good advice. I'll be looking at our own cloud spend with fresh eyes after this. Lucas: That's the goal. Thanks, Luna. Luna: Thanks, Lucas. See everyone next time.