Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Google Cloud Runs on Its Own Network and How That Changes the Math
Transcript
- Lucas: There is a question that almost never gets asked when a company chooses a cloud provider, and it is: 'Whose network is my data actually traveling over?' Luna: Most people just assume AWS, Azure, and Google Cloud all pipe traffic the same way. And that assumption is wrong. Lucas: Completely wrong. And it costs enterprises real money. So here is the difference: when you move a terabyte of data between two AWS regions, say from us-east-1 to eu-west-1, that traffic almost always leaves Amazon's network and goes over the public internet. Same for Azure. It's not that the cloud providers have no backbone—they do—but the bulk of inter-region data transfer is routed through third-party transit providers. Luna: And you pay egress fees for that. AWS charges roughly five to nine cents per gigabyte for inter-region data transfer, depending on volume. Lucas: Right. Which adds up fast if you're moving datasets regularly. Now Google Cloud is structurally different. Google built its own private fiber-optic network starting in the early 2000s. It connects all their data centers, all their regions, and it carries almost all of Google's internal traffic. When you use Google Cloud, your inter-region traffic is also routed over that same private network—unless you explicitly configure it to go over the public internet. Luna: So the data stays entirely on Google's infrastructure. No third-party carriers. No internet hops. Lucas: Exactly. And the pricing reflects that. Google Cloud's inter-region egress is roughly four to six cents per gigabyte—cheaper than AWS, but more importantly, the performance is more consistent because there are no internet bottlenecks. You get lower latency, fewer packet drops, and better throughput. Luna: I want to anchor this in a real example. A mid-size analytics company I spoke with recently moved their data pipelines from AWS to Google Cloud. They had about 50 terabytes of data moving between regions daily. Lucas: Fifty terabytes a day is significant. That's not a trivial workload. What were they doing? Luna: They were running batch analytics on IoT sensor data—industrial equipment data streaming from factories in Asia, processed in us-central1, then stored in europe-west4. They told me their data-transfer costs dropped by 40 percent after the migration, just from the network difference. No change in compute usage, no optimization of storage—just the network. Lucas: Forty percent. That's a concrete, measurable advantage. And it's one that doesn't show up in the usual spreadsheet comparisons of compute pricing or storage tiers. It's a network effect that Google has spent decades building. Luna: And it's not just about cost. It's also about what the network enables. Google's private network makes distributed AI training more practical. Lucas: Yes. Training large language models or vision models often requires sharding the work across hundreds or thousands of GPUs, sometimes across regions. The communication between those GPUs—the gradients, the weight updates—is extremely bandwidth-sensitive. If that traffic is going over the public internet, you get jitter and latency, which can slow down training and reduce model quality. Luna: Google's network allows those gradients to flow over a dedicated private backbone with very low jitter. That's one reason DeepMind and Google Brain were able to scale training to massive clusters earlier than others. Lucas: And now enterprises can access that same infrastructure. Google has been marketing this as 'the network that serves YouTube and Search is the same network that serves your workloads.' Which is a powerful pitch—if you can trust that your workloads won't be deprioritized behind consumer traffic. Luna: Has that been an issue? Do enterprise workloads actually get the same priority as Google's own services? Lucas: Google says yes, and they have SLAs to back it up. But I think the more interesting question is what happens as more companies pile onto that network. Google's private fiber is enormous—they own about 200,000 kilometers of terrestrial and submarine fiber, according to public estimates. But it's not infinite. If every enterprise customer starts moving terabytes across regions, can the network maintain that same consistent performance? Luna: That's a valid concern. AWS and Azure are building their own private networks too. Amazon has a project called 'Amazon Global Network' which is expanding, but they started later. Microsoft has similar investments. Lucas: They're playing catch-up. AWS announced a private backbone for inter-region traffic in 2019, but it's still not the default for all inter-region traffic. You have to opt into it, and it costs extra. Google's private network is the default. That's a big difference. Luna: So for a company that is data-intensive—say a genomics firm moving sequencing data, or an ad-tech company with real-time bidding data—the choice of cloud provider might hinge on this network advantage more than on compute pricing. Lucas: I think that's exactly right. And I'm starting to see enterprises make cloud decisions based on network topology, not just compute. They map out where their data sources are, where their users are, and then ask: which cloud provider can move data between those points with the lowest cost and lowest latency? Luna: And the answer increasingly is Google Cloud—for certain use cases. But it's not a blanket win. For workloads that are compute-bound and don't move much data, the network advantage is negligible. Lucas: Right. If you're running a web application that only serves static content from one region, the network doesn't matter. But for the big, growing workloads—AI training, real-time analytics, IoT pipelines, media transcoding—the network is the bottleneck and the cost center. Luna: Which brings us back to the earlier point: the unasked question. Companies spend weeks evaluating compute instances and storage tiers, but they rarely benchmark inter-region network throughput or egress costs across providers. Lucas: They should. And there are tools now that help. Google has a 'Network Intelligence Center' that visualizes traffic flows. AWS has 'CloudWatch' but it's not as network-focused. Third-party tools like 'Netreo' or 'ThousandEyes' can benchmark performance across providers. Luna: I want to talk about one more angle: the submarine cable investments. Google has invested in at least 20 submarine cable systems, including Dunant, Equiano, and Grace Hopper. Those are private cables connecting continents. Lucas: Those cables are not exclusively for Google Cloud customers—they also carry internal Google traffic—but they give Google unmatched control over transoceanic data routes. If you're a company with offices in Singapore and Frankfurt, and you need low-latency data transfer between them, Google Cloud can route you over a private cable that avoids the congested public internet paths. Luna: AWS and Azure are also investing in cables, but they often co-invest with consortiums. Google tends to own or lease entire cables outright. That gives them more flexibility in pricing and routing. Lucas: And that's an asset that's hard to replicate quickly. You can't just spin up a submarine cable in a year. These are multi-year, multi billion dollar projects. So the network advantage Google has today is likely to persist for the next several years at least. Luna: Unless the other providers make private networking the default and lower egress fees aggressively. Which they could do. Lucas: They could. And I think they will, eventually. But right now, the pricing disparity is real. And for a company moving 50 terabytes a day, like your analytics friend, the savings are substantial. Luna: What about smaller companies? Do they see the same benefit? Lucas: The percentage benefit is similar, but the absolute dollar amount is smaller. However, for startups that are building data-intensive products, every dollar counts. And choosing Google Cloud could mean paying significantly less for data transfer, which improves unit economics. Luna: One thing I've noticed is that Google Cloud sales teams now lead with the network narrative. They say 'our network is faster and cheaper' before they mention compute pricing. Lucas: That's smart. Because compute pricing is increasingly commoditized. All three providers offer similar instances at similar prices. But the network is a differentiator. And it's one that's rooted in real infrastructure, not just pricing games. Luna: So if I'm an enterprise architect evaluating cloud providers, what should I actually test? Lucas: I'd run a simple test: provision an instance in us-east1 and another in europe-west4 on each provider. Transfer a large dataset—say 100 gigabytes—between them. Measure the throughput, latency, and cost. Do it at different times of day. You'll see the difference clearly. Luna: And you'll probably find that Google Cloud's private network delivers more consistent performance, especially during peak hours. Lucas: That's been my experience. And it's one of those 'you have to see it to believe it' things. The data is out there, but most people haven't run the test. Luna: We've been deep in the network weeds, but I think there's a broader point here about how cloud competition is evolving. It's not just about compute and storage anymore. It's about infrastructure depth. Lucas: Exactly. And network infrastructure is the hardest to replicate. You can match a CPU spec sheet in a quarter. You cannot match a global private fiber network in a quarter. So Google has a durable advantage, at least for now. Luna: And that advantage matters most for the workloads that are growing fastest. AI, real-time analytics, IoT. The workloads that define the next decade of cloud. Lucas: It's interesting to hear you say that, because this show is built for people who are thinking about exactly those workloads. And honestly, the reason we can keep doing these deep dives is that a small group of listeners chips in monthly through buy me a coffee dot com slash fexingo. It's listener-funded, no ads, and that support directly lets us keep the conversation going. Luna: Yeah. It's a small thing that makes a big difference. And it keeps us independent, which matters when we're talking about things like network pricing. Lucas: Alright. So network is the hidden variable. Next time you evaluate a cloud provider, don't just compare compute and storage. Compare the network. Because that's where the real cost and performance differences live. Luna: And where the next competitive battles will be fought.