Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / The Next Frontier in Cloud Contracts Is Bandwidth Reservation
Transcript
- Lucas: So we're at episode 100 of this show, which feels like a good moment to look at a clause that's been quietly appearing in enterprise cloud agreements over the past year — committed bandwidth reservations. Luna: Okay, that sounds dry, but I have a feeling it's one of those things that could really change how companies budget for infrastructure. Lucas: Exactly. Unlike standard data egress fees, which charge you per gigabyte as you send it out, these provisions lock in a minimum throughput level between specific regions for a fixed term — typically one to three years. And if you don't use that committed bandwidth, you still pay for it. Luna: So it's like a reserved instance, but for the network. Lucas: Right. And just like reserved instances, it shifts cost risk from the provider to the customer. The provider gets predictable revenue from the network link, and the customer gets a lower per-unit price — but only if their traffic patterns are consistent. Luna: Before we go too deep — if you're someone who builds or runs infrastructure, you probably think about bandwidth costs all the time. And honestly, conversations like this are exactly why listeners support this show. If these discussions are useful for what you're building, you can help keep the show ad-free and independent at buy me a coffee dot com slash fexingo. No pressure, just if it's worth something to you. Lucas: Yeah, it's a small gesture that makes a big difference for us. So back to bandwidth reservations — I want to walk through a concrete case. There's a mid-stage fintech company we'll call PayFlow, handling real-time payment processing across Europe. They signed a three-year committed 10 Gbps link between AWS us-east-1 and eu-west-1. Luna: Ten gigabit per second — that's roughly 3.2 petabytes per month if they used it fully. Who actually needs that kind of throughput? Lucas: In their case, they were doing real-time fraud detection that required constant model updates between training in Virginia and inference in Dublin. They also had synchronous database replication. Before the reservation, their monthly data egress bill was around $180,000. After negotiating a committed bandwidth deal, they dropped to about $110,000 — a 39 percent reduction. Luna: That's significant. But what happens if their traffic drops? Say a competitor takes market share or they migrate some workloads to another region. Lucas: That's the risk. The penalty for underutilisation is essentially the difference between the committed amount and what they actually use, billed at the on-demand rate. So if they commit to 10 Gbps but only use 6, they pay for the full 10. Some contracts have a 'make-up' provision — you can roll over unused capacity to the next month — but those are rare. Luna: It sounds a lot like how airlines overbook seats, or how hotels charge for no-shows. The provider is protecting themselves against idle capacity. Lucas: Exactly. AWS actually has a dedicated product for this called AWS Dedicated Bandwidth, where you can reserve capacity between specific pairs of Availability Zones or Regions. Google Cloud has something similar with their 'Partner Interconnect' committed bandwidth. Azure calls it 'ExpressRoute' with committed circuit bandwidth. Luna: But here's the thing — is this really a new clause, or just a rebranding of what MPLS or sd wan carriers have been doing for decades? Lucas: It's both. The concept isn't new in telecom, but what's new is applying it to cloud networking, where traffic patterns are more elastic. Traditional carriers assumed fairly predictable traffic. Cloud workloads can spike or plummet based on product launches, marketing campaigns, even time of day. Luna: So who benefits most from these clauses? It can't be every company. Lucas: It's most beneficial for companies with steady-state, high-volume data pipelines — think media streaming platforms, financial exchanges, large-scale AI training jobs that run for weeks. For them, the savings are real. But for a startup that's still figuring out its product-market fit, committing to a fixed bandwidth for three years could be a trap. Luna: What about multi-cloud setups? If you're running some workloads on AWS and some on Azure, you might need cross-cloud bandwidth. Do these clauses apply there? Lucas: They do, but they're more complex. Some providers offer 'direct connect' partnerships — for example, AWS and Azure have peering in some regions. But the clauses are usually provider-specific. So if you want a committed link between AWS and Azure, you're often dealing with a third-party carrier like Equinix or Megaport, and then you have to manage separate contracts. Luna: That feels like another form of lock-in. The more you commit to one provider's network, the harder it is to switch. Lucas: That's a fair critique. But there's also an infrastructure logic here. As AI training pipelines stretch across multiple regions — think training a large language model on data in Oregon and running inference in Tokyo — predictable bandwidth becomes more valuable than burst capacity. You don't want your training job to stall because of network congestion. Luna: So the trade-off is cost certainty versus flexibility. And I imagine finance teams love cost certainty. Lucas: They do. In fact, I've spoken with a few CFOs who say that unpredictable bandwidth costs are one of their top three frustrations with cloud. A committed bandwidth clause turns a variable cost into a fixed one, which makes budgeting easier. But it also means you need good forecasting. Luna: How are these clauses being negotiated? Is it a standard part of enterprise agreements now? Lucas: Not yet standard, but they're becoming a common addendum. For large deals — say over $10 million in annual cloud spend — the provider will often include a committed bandwidth option as a 'discount lever'. The customer gets a lower overall price in exchange for a network commitment. I've seen it in about 30 percent of enterprise agreements signed in the last six months. Luna: That's up from almost zero a couple years ago. What's driving the increase? Lucas: Two things. One is the growth of AI workloads, which are bandwidth-hungry. The other is that cloud providers themselves are trying to manage their own network capacity more efficiently. By locking in commitments, they can plan capital expenditures on fiber and routers with more confidence. Luna: So it's a win for the provider, a potential win for the customer if they forecast right, and a risk if they don't. Sounds like a classic financial instrument. Lucas: It really is. And that's why I think we'll see more of these clauses, maybe even bundled with reserved compute or storage. Imagine a three-year commitment that covers compute, storage, and bandwidth in one package, with a single penalty if you underuse any of them. Luna: That would be a true 'all-in' reserved instance. It would also make switching providers even harder. Lucas: Right. But for companies with predictable needs, the savings could be substantial. Let me give you a specific number from another case — a large video streaming platform that committed to 40 Gbps between AWS Oregon and Northern California. They saved about $2.8 million over two years compared to on-demand egress. Luna: Wow. That's real money. But again, they had to be sure their traffic wouldn't drop. I imagine there's a whole consulting niche now for 'bandwidth forecasting'. Lucas: There is. Some firms specialize in analyzing network telemetry to recommend optimal commitment levels. They look at peak to average ratios, growth trends, seasonal patterns. It's surprisingly nuanced. Luna: What about penalties for exceeding the committed bandwidth? Do you get throttled or just billed at a higher rate? Lucas: Typically, you get billed at an overage rate that's higher than the committed rate but still lower than standard on-demand egress. Some contracts have a 'burst' allowance — say up to 20 percent above the commitment without penalty. But it varies. Luna: So the key is to set the commitment slightly above your expected peak, not your average. Lucas: Exactly. And that's where the risk comes in — if you set it too high, you pay for unused capacity. Too low, and overage charges eat into your savings. Luna: One hundred episodes in, and we're talking about bandwidth reservation clauses. I love that we’ve found a niche that keeps getting deeper. Lucas: The cloud is an iceberg — most people see the tip, the compute and storage, but the networking underneath is where a lot of the hidden costs live. And these clauses are a sign that providers are getting more sophisticated about extracting value from that layer. Luna: Or that customers are getting smarter about demanding predictability. Either way, it's a shift worth watching. Lucas: Agreed. So here's a question for the listener: if your cloud bill has a significant bandwidth component, is it more predictable than it was a year ago? And if not, maybe it's time to ask your provider about a committed bandwidth clause. Luna: Good place to leave it. Thanks for listening — see you next time.