When teams model the cost of moving a media archive to the cloud, the spreadsheet usually has a clean line for storage at rest and little else. But the link between your facility and the cloud region is a cost center of its own. For archives measured in petabytes, it shapes both the schedule and the budget. A recent Tech Insider comparison of the three hyperscaler options is a good prompt to look at what this decision means for media and entertainment workloads.
Three clouds, three on-ramps
AWS Direct Connect, Azure ExpressRoute, and Google Cloud Interconnect all provide private connectivity that bypasses the public internet, terminating at colocation facilities. Each one is the only private on-ramp to its own cloud, so for most teams this isn't a head-to-head shopping decision. Your choice of cloud largely makes it for you. The comparison matters most if you run multi-cloud, or if you're still deciding where an archive tier should live.
What the headline prices look like
Production setups need redundancy, so the useful comparison is a redundant pair. Using the figures published in that comparison, monthly cost looks like this:
| Redundant configuration | 10 Gbps / month | 100 Gbps / month |
|---|---|---|
| AWS Direct Connect (two ports) | $3,285 | $32,850 |
| Google Cloud Interconnect (two circuits)* | $3,398.88 | $33,988.80 |
| Azure ExpressRoute Direct (port pair, Zone 1) | $6,000 | $50,000 |
*Before VLAN attachment fees, which Google bills separately. Prices are as published in the Tech Insider comparison (October 2026); confirm against each provider's current pricing page before budgeting.
Azure's premium is real, roughly 76–83% at 10 Gbps, but part of it is that the port pair includes redundancy by design, where AWS and Google leave you to build it. The most common budgeting mistake is pricing a single connection when production will need two, which nearly doubles the bill.
The line item that can outgrow the port fee
Inbound data is free on all three. Outbound is not. The comparison lists egress over these connections at $0.02–$0.1923/GB for AWS (varying by region), $0.025–$0.14/GB for Azure, and $0.020–$0.11/GiB for Google.
For a media archive, those rates matter more than the port fee. Think about restoring a library for a remaster, pulling content back for re-transcoding, or repatriating a tier. One petabyte is about 1,000,000 GB. At $0.02/GB, that's roughly $20,000 in transfer charges. At $0.1923/GB, it's over $190,000. A redundant 10 Gbps pair on AWS runs about $39,000 a year in port fees, so a single large restore can match or dwarf a year of the connection itself. If your workflow involves regular retrieval, model egress as carefully as you model storage.
How long will it actually take?
Bandwidth also sets your timeline. At full line rate, a petabyte takes about 92 days over 1 Gbps, about 9 days over 10 Gbps, and about 22 hours over 100 Gbps. Real-world throughput will be lower. Your source systems matter too: a single LTO-9 drive streams around 400 MB/s native, or about 3.2 Gbps, so filling a 10 Gbps link from tape means running several drives in parallel. A bigger pipe doesn't help if the source can't feed it.
Details that bite later
Location is not region. Colocation facilities can sit hundreds of miles from the cloud region you're targeting, and the backbone hop adds latency. Check this before you commit to a facility.
Cross-connects are the long pole. None of the three providers publishes guaranteed provisioning timelines. The physical cross-connect at the colocation facility usually takes longest, so order it first.
SLAs are credits, not guarantees. The percentages describe when you get money back, not a promise that outages won't happen. Google offers no SLA at all on a single connection. Its 99.9% and 99.99% tiers require redundant multi-zone designs.
Nothing is encrypted by default. AWS and Azure support MACsec on select dedicated options, and Google suggests layering HA VPN or MACsec on top. If your content is pre-release or under studio security requirements, plan encryption in from the start.
A sensible way to approach it
Start by auditing current utilization and growth, then identify the colocation facility nearest your existing cross-connect. Pre-order the physical connection, configure the virtual interface or VLAN and validate BGP, and run in parallel with your existing path before staging the cutover. Decommission the old connection only after one clean billing cycle.
Once it's live, treat it like any other FinOps line item. Tag the spend by workload rather than lumping it under “networking,” and revisit sizing annually. Newer high-capacity single ports can sometimes beat older redundant pairs on cost.
The bottom line
There's no universal winner. At redundant 10 Gbps, AWS and Google are close on port cost, with AWS slightly lower, and Azure costs more but bundles redundancy and fits Microsoft-centric environments. For archives, the decision that moves the budget most is usually not which port you buy but how much data you plan to pull back out, and how fast you need to move it.
If you want help modeling connectivity, egress, and storage together for your own archive, get in touch.
Source: Tech Insider, “Direct Connect vs ExpressRoute vs Interconnect Comparison [2026]”, October 3, 2026. Transfer-time and restore-cost examples are ArchiveEconomics.com calculations based on the published rates and line-rate math.
← Back to all posts