Your data centre colocation conversation
Hey there, thanks for watching!
Colocation + Connectivity: The backbone of global enterprise infrastructure
The rack is the easy part. The neighbours are the point.
A colocation facility is an interconnection venue as much as it is a data centre. What makes one site more valuable than another is who else is in the building and how directly you can reach them. Floor space and power density matter, but they're rarely the deciding factor once you've been through this a few times.
A rack at a major facility in Singapore, Frankfurt, or Ashburn sits within a few metres of cloud on-ramps for AWS, Azure, Google, and Oracle. Cross-connect to any of them and that traffic never touches the public internet. The same building typically houses major SaaS providers, security platforms, and content networks, all reachable through private connections that stay inside the facility.
Most infrastructure decks skip over this part. The useful comparison for colocation isn't running your own data centre. Think instead about the cost of every workload taking the long route through the public internet to reach the cloud, the SaaS platform, and the security stack it depends on. That cost is invisible until you remove it.
The CFO’s guide to reducing cloud costs without sacrificing performance
Public cloud pricing pages are clear enough on storage and compute. Egress is where the bill gets interesting.
A direct cross-connect from a colocation facility into a cloud on-ramp typically costs a fraction of what equivalent egress over the public internet would, and the pricing is predictable rather than variable. For organisations running multi-cloud setups or moving large volumes between regions, that one line item often pays for the colocation footprint several times over. It rarely appears in the original business case because nobody thought to look for it.
The eProcurement company in the video cut its infrastructure costs by close to half. A meaningful portion of that came from rebalancing where their data lived and how it moved between clouds, not from the racks themselves.
AI and your network: what the numbers actually say
1 in 3 global network traffic is AI-enhanced
Mostly from features inside everyday SaaS apps like Microsoft 365, Salesforce, and Oracle, not from dedicated AI platforms.
4x AI bandwidth
This is how much AI bandwidth over two-thirds of current AI users expect to need within 18 months.
10 to 100x is the throughput spike AI workloads create
Compared to standard web traffic, AI workloads breaks the planning assumptions most networks were built around.
8 in 10 AI models are publicly hosted
This means every inference call your organisation makes is already a cloud round-trip your network has to handle.
Colocation puts your data adjacent to all those clouds at once, rather than committed to any one of them. That's the difference between AI that scales at a predictable cost and AI that quietly runs up egress charges every time it fires.
Source: Omdia commissioned research, 2025. Survey of 419 large global enterprises.
Ensuring data sovereignty in a global market
Where your data travels is becoming as regulated as where it lives.
A few years ago, data sovereignty meant proving that customer data was stored inside a permitted jurisdiction. That part hasn't gone away. What's been added is the question of where data travels between processing locations, and which networks it crosses on the way.
Regulators in several APAC and European markets are now asking organisations to show routing transparency, not just storage location. If your traffic between two in-region clouds transits a third country because that's the cheapest path, your compliance team may now need to account for it in a way they didn't before.
Colocation paired with carrier-neutral connectivity gives you both the routing control and the audit trail to answer those questions. Public internet routes don't give you either.
The colocation conversation, by industry
Financial services
Payments, trading, and core banking systems are moving toward colocation footprints adjacent to major exchanges and cloud regions. The trading firm in the video moved onto Ethernet Private Line connections terminating inside Telstra's cable landing stations for its core backbone routes. For payment authorisation and high-frequency workloads, the round-trip delay that saves is the entire commercial case.
Manufacturing
Quality control, OT systems, and ERP are increasingly distributed across sites with AI inference at the edge. Colocation gives manufacturers a regional anchor close to production hubs, with private connectivity into the clouds for workloads that need it and on-premises edge for the ones that can't tolerate a cloud round-trip at all. An idle automotive production line now costs around $2.3 million an hour. That number has made network reliability a board conversation, not just an IT one.