Hybrid Cloud Colocation
How to Connect Colocation to AWS, Azure and Google Cloud, What It Costs, and Which Providers Fit
Hybrid cloud colocation keeps steady workloads on your own hardware and connects them privately to the cloud, cutting egress from $0.09 a gigabyte to $0.02. We find the facilities with the on-ramps you need.
Price a Hybrid DeploymentHybrid cloud colocation is the architecture where a company runs its steady, predictable workloads on its own servers in a colocation data center and uses AWS, Azure or Google Cloud for what changes: burst capacity, managed services, development and the workloads that are cheaper on demand. The two halves are joined by a private connection, and that connection is the whole point. Over the public internet, moving data out of AWS costs $0.09 per gigabyte and travels an unpredictable path; over a private on-ramp such as AWS Direct Connect, Azure ExpressRoute or Google Cloud Interconnect, the same gigabyte costs $0.02 and the round trip is a few milliseconds. A facility with the cloud on-ramp in the building turns that into a cross-connect; a facility without it needs a circuit to one that has it. So a hybrid deployment is decided by three choices, in this order: which workloads stay on your hardware, which building hosts them, and how it reaches the cloud.
MCA calculation from AWS Direct Connect pricing and AWS internet data transfer tiers ($0.09 per GB to 10 TB, $0.085 to 50 TB, $0.07 to 150 TB, $0.05 beyond); prices checked September 2026. Azure and Google Cloud follow the same pattern with their own port and egress rates. Free to cite with a link to this page.
Once you have asked a provider for pricing directly, it will only work with you on its own terms. Send it to us first and you'll hear back within 24 hours from the person who will run your search, with every facility that has the on-ramps you need quoted at once, cross-connects and circuits included, and benchmarked against what comparable deployments pay. The provider you choose pays us, so it costs you nothing.
What Goes Where: Colocation or Cloud
The split is not by application but by behavior. Workloads that run all the time on hardware you can predict belong on your own servers; workloads that spike, change shape or depend on a managed service belong in the cloud.
| Workload | Where it usually belongs | Why |
|---|---|---|
| Databases and steady application tiers | Colocation | Predictable load, large storage, licensing that favors owned cores; the cloud charges most for exactly this profile |
| Large data sets and backups | Colocation | Storage at rest and egress are the two cloud bills that never stop; owned storage behind a private link is a fraction of either |
| GPU inference at steady load | Colocation, in a dense hall | Owned or reserved GPUs running around the clock beat on-demand cloud instances by a wide margin. See AI and GPU colocation |
| GPU training and bursts | Cloud, or a rented cluster | Weeks of peak demand, then nothing; rent it. See GPU cluster rental |
| Web front ends, development, test | Cloud | Elastic, short-lived, and cheap when idle |
| Managed services: analytics, queues, identity | Cloud | The value is the service, not the server |
| Disaster recovery | Either, depending on the primary | Colocation primary with cloud DR, or cloud primary with a colocation replica; the private link makes both work. See disaster recovery colocation |
Three Ways to Connect Colocation to the Cloud
| Method | How it works | Cost shape | Best for |
|---|---|---|---|
| 1. On-ramp in the buildinga cross-connect | The cloud's Direct Connect, ExpressRoute or Interconnect port is in the same facility; a fiber cross-connect from your cage to the meet-me room reaches it | Cross-connect ($200 to $500 a month) plus the cloud's port fee and $0.02-per-GB egress | The lowest latency and the fewest parties; the reason to choose an on-ramp building |
| 2. Cloud exchange or fabricvirtual connections | One physical port into the operator's interconnection platform (CoreSite Open Cloud Exchange, Equinix Fabric, Digital Realty ServiceFabric), then virtual circuits to each cloud, sized and changed by software | Port fee plus a per-virtual-connection fee by bandwidth; hosted cloud connections from about $22 a month at 50 Mbps | Multi-cloud, changing bandwidth, and reaching clouds whose on-ramp is elsewhere on the operator's footprint |
| 3. Network-as-a-service or carrier circuitMegaport, Lumen, PacketFabric and the carriers | A provider with ports in many facilities and at the cloud on-ramps carries your traffic from a building with no on-ramp to the cloud, as a virtual circuit or a dedicated wave | Port plus circuit by bandwidth and distance; competitive with method 2 and available in far more buildings | Deployments in buildings without on-ramps, secondary markets, and multi-region designs |
How to Design a Hybrid Architecture Between Colocation and AWS
Sort the workloads by behavior
Steady and data-heavy to colocation, elastic and managed to the cloud, using the table above. Measure a month of actual usage first; cloud bills tell you what is steady.
Measure the traffic between the two halves
Gigabytes per month out of the cloud, and the latency the applications can tolerate. That number sizes the connection and decides whether an on-ramp building pays for itself.
Choose the region, then the building
Pick the AWS region the cloud half runs in, then a colocation facility with a Direct Connect location in the building or one metro hop away. In the New York metro that is the Manhattan carrier hotels and Secaucus; in Northern Virginia, most of Ashburn.
Size the connection and build two paths
A dedicated 1 Gbps port for most mid-market deployments, 10 Gbps for data-heavy ones, with a second connection through a different path (a second on-ramp, a fabric or a NaaS provider) for resilience. A VPN over the internet is the last-resort backup, not the design.
Put the address plan and identity in place before the move
Non-overlapping private address space, a transit gateway or equivalent on the cloud side, one directory, one set of monitoring.
Price the whole bill, then negotiate it
Cabinet or per-kW rate, cross-connects at the number you will need, the cloud's port and egress, the second path, and remote hands, over the term. The pricing guide has what quotes leave out; cross-connect rates and expansion are negotiable before signature.
Which Providers Fit a Hybrid Cloud Architecture
Four operators built their platforms around cloud connectivity; the rest reach the clouds through cross-connects and network providers. The building matters as much as the brand: an operator's on-ramps are in specific facilities, not everywhere on its map.
| Provider | Platform | Clouds reached | Where the on-ramps are |
|---|---|---|---|
| CoreSite | Open Cloud Exchange | AWS, Azure, Google Cloud, Oracle, IBM and others | New York (32 Avenue of the Americas, Secaucus), Northern Virginia, Chicago, Los Angeles (One Wilshire), Silicon Valley, Denver, Boston |
| Equinix | Equinix Fabric | Every major cloud, from the widest footprint | Direct Connect and ExpressRoute locations across the major campuses, including Secaucus, Ashburn, Chicago, Dallas, Silicon Valley |
| Digital Realty | ServiceFabric | AWS, Azure, Google Cloud, Oracle and others | The Manhattan carrier hotels, Ashburn, Chicago (350 East Cermak), Dallas and the other campuses |
| Flexential | FlexAnywhere | AWS, Azure, Google Cloud | On-ramps at several of its 19 markets, strongest in the Southeast, Texas, Colorado and the Northwest |
| DataBank, Cologix, TierPoint, Iron Mountain | Cross-connects to on-ramps in the same building where present, plus Megaport or PacketFabric ports | All, through the network providers | On-ramps in their larger markets; NaaS ports at most sites. Often the better economics when the compute is dense or the deployment is large |
Equinix vs CoreSite for cloud on-ramp access
Both have on-ramps in the building and a software fabric to reach the clouds. Equinix Fabric reaches more clouds from more locations and sits next to the deepest interconnection ecosystem; CoreSite's Open Cloud Exchange covers the major clouds from its markets at a lower price point, and in New York it does so from both a Manhattan carrier hotel and Secaucus. For a mid-market hybrid deployment whose cloud is AWS, Azure or Google, the two are close on function and the decision usually comes down to the building, the rate and the cross-connect terms. For multi-cloud with unusual clouds or global reach, Equinix.
Who offers the best pricing and terms for hybrid colocation-cloud architectures for mid-market companies?
Usually not the operator with the best-known fabric, because the fabric is a small part of the bill. The largest lines are the cabinets and power, which are cheapest at the mid-market and regional operators, and the cloud's own port and egress fees, which are the same whoever hosts you. The best combinations we place are a small connectivity footprint (or a single NaaS port) where the on-ramp is, with the compute in a value facility a cross-connect or metro circuit away. That structure also keeps two operators competing for the growth. The provider comparison covers the operators by group.
Is high-density colocation cheaper than renting GPU capacity from a cloud provider?
For steady inference, yes: an owned or reserved H100 or H200 server running around the clock in a dense hall costs a fraction of on-demand cloud GPU instances over three years, and the model and data sit next to it. For training bursts, no: rent the cluster. The numbers are on H100 rental, H200 rental and GPU rental, and the hybrid version, owned inference plus rented training, is increasingly the standard AI architecture.
Interconnected Colocation in New York: Where the On-Ramps Are
New York's cloud on-ramps are in two places: the Manhattan carrier hotels (60 Hudson Street, 111 8th Avenue, 32 Avenue of the Americas) and the Secaucus campuses (Equinix NY2 to NY6, CoreSite NY2 and NY3). Every other zone in the metro, Brooklyn and Queens included, reaches them over a metro circuit or a NaaS port, which is fine for most workloads and adds a few hundred dollars a month and a millisecond or two.
Manhattan carrier hotels
On-ramps in the building, the densest carrier mix in the country, and the highest rates. Right for the connectivity footprint of a hybrid design; rarely the right place for the compute. See Manhattan data centers.
Secaucus
On-ramps in the building at both campuses, at a lower rate than Manhattan, with the trading ecosystem next door. The most common home for a New York hybrid deployment. See NYC colocation.
Brooklyn, Queens and the outer boroughs
Edge facilities with no cloud on-ramps of their own; they connect to Manhattan or Secaucus over metro fiber or a NaaS port. Right when latency to operations in the boroughs matters more than latency to the cloud.
New Jersey and Orangeburg campuses
DataBank, Cologix and the New Jersey sites of the interconnection operators: the best economics and the highest density in the metro, a cross-connect or metro circuit from the on-ramps. Where the compute half of most New York hybrid designs belongs.
Five Mistakes in Hybrid Cloud Colocation
Choosing the building for the compute, not the connection
Cheap cabinets with no path to the cloud become expensive cabinets once the circuit and the latency are counted. Decide how you reach the cloud, then place the hardware.
Paying carrier-hotel rates for racks that never touch a carrier
The connectivity footprint belongs at the on-ramp; the compute belongs where power is cheapest. Splitting them is the single largest saving in most hybrid designs.
One path to the cloud
A single cross-connect or circuit is a single point of failure for the whole architecture. Two paths through two providers, priced from the start.
Leaving cross-connects out of the negotiation
They are negotiable before signature and rarely after. State how many you will need in year three, cap the monthly rate, and ask for installation to be waived.
Going to one provider first
Operators price differently when they know others are quoting the same requirement, and once you have gone to one directly, it will not work through an advisor on your deal.
How to Write an RFP for Colocation and Interconnection
A good hybrid RFP is short and specific. It states the power and cabinets, the density, the term and start date; the clouds and regions to reach, the bandwidth and latency needed, and whether the on-ramp must be in the building; the carriers required and the number of cross-connects over the term; compliance scope; and the second path. It asks for a five-year total including cross-connects, setup, remote hands and power, not a monthly rate. We write and run these for clients as part of the search, and the data center lease guide covers the terms the answers should be held to.
How We Place Hybrid Deployments
You tell us the workloads staying on your hardware, the clouds and regions they connect to, the traffic between them, and the power, density, compliance and term. We shortlist the facilities with the on-ramps or the network paths you need, in the right zone, and have them quote the same requirement at the same time, cross-connects, circuits and the second path included. We benchmark the quotes against what comparable deployments pay and negotiate the rate, the cross-connect terms and the contract before you sign. If the better answer is a different structure, connectivity in one building and compute in another, we say so and price both. The provider you choose pays us from the channel budget it would otherwise spend on its own sales team, so it costs you nothing and the rate is not marked up. How it works.
Frequently Asked Questions
What is hybrid cloud colocation?
An architecture that runs steady, predictable workloads on a company's own servers in a colocation data center and uses public cloud (AWS, Azure, Google Cloud) for elastic capacity, managed services and development, with the two joined by a private connection such as AWS Direct Connect, Azure ExpressRoute or Google Cloud Interconnect rather than the public internet.
How much does it cost to connect colocation to AWS?
A dedicated 1 Gbps Direct Connect port is $0.30 an hour, about $219 a month, and 10 Gbps about $1,643; hosted connections through partners start around $22 a month at 50 Mbps. Data out of AWS costs $0.02 per gigabyte over Direct Connect against $0.09 over the internet, and data in is free. Add a cross-connect at the facility, typically $200 to $500 a month, or a circuit if the on-ramp is in another building.
When does a private cloud connection pay for itself?
At roughly 3 terabytes a month of outbound traffic on a 1 Gbps port, where the $0.07 per gigabyte saving covers the port fee. Most companies running real workloads in colocation pass that in the first month, and the latency and reliability benefits arrive regardless of volume.
Which colocation providers are best for hybrid cloud architectures?
The four with their own cloud connectivity platforms: CoreSite (Open Cloud Exchange), Equinix (Fabric), Digital Realty (ServiceFabric) and Flexential (FlexAnywhere). The mid-market and regional operators reach the clouds through cross-connects and network providers such as Megaport, often at better economics for the compute. The best designs frequently use both: a connectivity footprint at an on-ramp and the compute in a value facility.
How do I design a hybrid architecture between colocation and AWS?
Sort workloads by behavior (steady to colocation, elastic and managed to cloud); measure the traffic between them; choose the AWS region and then a facility with a Direct Connect location in the building or one metro hop away; size a dedicated port and add a second path through a different provider; put non-overlapping addressing, a transit gateway and one identity system in place; then price and negotiate the whole bill including cross-connects and egress.
What is the difference between Lumen, Equinix and Megaport for cloud on-ramps?
Equinix sells the facility with the on-ramp in the building and its Fabric to reach clouds. Megaport sells a virtual network that reaches the clouds from ports in most major colocation facilities, useful when your hardware is in a building without an on-ramp or you need several clouds from one port. Lumen sells carrier circuits and managed networking with contracted service levels. Many hybrid deployments combine two of them for the primary and second path.
Is Equinix or CoreSite better for cloud on-ramp access?
Both have on-ramps in the building and a software fabric. Equinix reaches more clouds from more locations alongside the deepest interconnection; CoreSite covers the major clouds from its markets at a lower price point. For a mid-market deployment on AWS, Azure or Google they are close on function, and the decision usually comes down to the building, the rate and the cross-connect terms.
Where are the cloud on-ramps in New York?
In the Manhattan carrier hotels (60 Hudson Street, 111 8th Avenue, 32 Avenue of the Americas) and the Secaucus campuses (Equinix and CoreSite). Facilities in Brooklyn, Queens, Westchester, northern New Jersey and Orangeburg reach them over metro fiber or a network-as-a-service port, which adds a few hundred dollars a month and a millisecond or two.
Is high-density colocation cheaper than renting GPUs from a cloud provider?
For steady inference, yes: owned or reserved GPU servers running around the clock in a dense hall cost a fraction of on-demand cloud instances over three years. For training bursts, renting a cluster is cheaper than owning idle hardware. The common hybrid answer is owned inference plus rented training.
What should a colocation and interconnection RFP include?
Power, cabinets, density, term and start date; the clouds and regions to reach, bandwidth and latency, and whether the on-ramp must be in the building; carriers and cross-connect count over the term; compliance scope; a second path; and a request for a five-year total including cross-connects, setup, remote hands and power rather than a monthly rate.
Can I use colocation as disaster recovery for a cloud primary?
Yes, and it is common: a colocation replica behind a private connection gives a recovery site outside the cloud provider's control at a predictable cost, and the same link carries the replication traffic at $0.02 per gigabyte. The reverse, cloud DR for a colocation primary, works the same way.
What does it cost to place a hybrid deployment through Metro Colo Advisory?
Nothing. The provider you choose pays us from its channel budget, the same way it pays its own sales team, and your rate is not marked up. If the better answer is a different structure or the provider you already use, we say so.
Price a Hybrid Deployment
Tell us the workloads staying on your hardware, the clouds and regions they connect to, the traffic between them, and the power, density, compliance and term. You'll hear back within 24 hours from the person who will run your search, with every facility that fits quoted at once, cross-connects and circuits included, benchmarked against what comparable deployments pay, and the terms negotiated before you sign. No cost, no obligation, and the provider you choose pays us.
That is what keeps every provider competing for it.
The move: cloud repatriation, the cloud vs colo calculator, data center migration and disaster recovery colocation.
Providers and buildings: the provider comparison, CoreSite, Equinix NY4, Digital Realty, DataBank, Cologix, Manhattan data centers, NYC colocation and data center connectivity.
The numbers: the colocation pricing guide, data center cost, the data center lease guide and bare metal.
AI in a hybrid design: AI and GPU colocation, GPU rental, GPU cluster rental, H100 rental, H200 rental, B200 and B300 rental, neocloud providers and CoreWeave competitors.
