When you sign a contract for a colocation cabinet at Equinix, CyberRealty, Switch, or any other data center operator, you're buying three things: physical rack space, power, and cooling. You are not buying connectivity. That decision—how your colo connects to your offices, your cloud workloads, and the broader internet—is separate, and it's where most infrastructure teams either overpay, under-architect, or both.

The distinction is critical. A major data center in Portland might quote you $8,000/month for a 10 kW cabinet. The facility handles the physical infrastructure flawlessly. But if you let them also sell you the "convenient" internet access from their network at $5/Mbps, you've just committed to paying 2–3x the market rate for bandwidth. And if you don't plan for redundancy in your connectivity circuits, your expensive, redundant colo equipment is useless when a single carrier has an outage.

This guide covers what you're actually buying in colocation, the connectivity options available to you, the hidden costs that turn a $8k/month cabinet into a $12–14k/month deployment, and how to architect connectivity that actually serves your business instead of your data center provider's margin sheet.

What Colocation Is—and What It's Not

Colocation is, at its core, renting physical space, power, and cooling from a third-party facility. The major providers in North America include Equinix (largest, carrier-neutral, global presence), CyberRealty (strong on mid-market, Midwest and West), Switch (Las Vegas-centric, known for density), Iron Mountain (traditionally archival, expanding into active colo), and QTS (regional, strong in Eastern US). There are also hundreds of smaller regional carriers operating their own facilities—AT&T, Lumen, Comcast, and others—who colocate equipment in their own data centers.

What's included in the colo rent: a physical cage or cabinet, measured power allocation (billed per kW), cooling capacity, physical security (card access, CCTV, security personnel), fire suppression, redundant backup power (usually), and UPS systems. The facility maintains the building; you maintain your equipment.

What's not included: any network connection to anywhere else. Not to your office. Not to AWS, Azure, or Google Cloud. Not to the public internet. The phrase "connectivity in the building" is key here—Equinix can provide a cross-connect (a direct fiber patch) from your cabinet to another party inside the same building. But "connectivity to the world" (your offices, cloud providers, the internet) comes from carriers and cloud providers, not the colo facility itself, even though facilities will often sell you add-on services at premium prices.

This is the first cost trap: accepting the facility's "managed internet" or "colo connectivity" service. You're paying for the facility's convenience, not for competitive market pricing.

The Four Types of Colo Connectivity

1. Internet Access from the Facility's Own Network

Most major data centers operate as both colo providers and network operators. Equinix runs its own AS (Autonomous System). CyberRealty contracts with multiple carriers. They offer internet transit as an add-on: you rent a cabinet, they provision a port to their network, and you get internet access.

This is the easiest option—you're done in a week, single SLA, single support queue. But it's also usually the most expensive and carries the highest risk: if the facility's network has an outage, your colo is offline. Pricing is typically $3–8/Mbps (committed), compared to $0.50–2/Mbps for the same bandwidth from a carrier via a diversity-aware procurement process.

When it makes sense: Small footprints (one cabinet), non-critical workloads, short-term deployments, or situations where capital is more valuable than monthly OpEx.

2. Dedicated Private Circuits from Office to Colo

A carrier (AT&T, Lumen, Zito, local fiber carriers) provisions a dedicated circuit—MPLS VPN or Ethernet over fiber—from your office to your colo. The circuit is yours; the bandwidth is guaranteed.

Pricing: $500–2,000/month for a 50 Mbps circuit within 50 miles, up to $4,000–5,000/month for 1 Gbps over longer distances. Distance, fiber availability, and the carrier's market presence drive cost. A major metropolitan area (Portland, Seattle, San Francisco) will have lower pricing than rural areas.

This circuit is not internet; it's a private tunnel. Your office connects to your colo, and your colo connects to the internet (or to cloud providers) via separate paths. For redundancy, you'd provision a second circuit from a different carrier on a physically diverse path.

When it makes sense: Organizations with 5+ Mbps sustained bandwidth between office and colo, or those needing consistent low-latency access to equipment in the colo.

3. Cloud Provider Direct Connections

If your workloads are in AWS, Azure, or Google Cloud, you can establish a dedicated connection from your colo directly to the cloud provider's network via a cross-connect in the data center.

For AWS Direct Connect: The facility often provides a port at no charge or charges $150–500/month. AWS charges ~$200–400/month for a 1 Gbps connection, plus $0.02–0.05/GB for data transfer (vs. ~$0.02/GB for internet transit, but with lower latency and no cap on simultaneous connections). If you're moving 5–10 TB/month or more to AWS, Direct Connect pricing is typically competitive or cheaper.

Azure ExpressRoute and Google Cloud Interconnect follow similar models. A 1 Gbps Dedicated connection on ExpressRoute runs $200–300/month from Microsoft, plus data transfer charges. Google Cloud Interconnect is priced in 10 Gbps increments, starting around $300–400/month.

When it makes sense: Hybrid architectures (primary workloads on-premises in the colo, secondary or burst workloads in cloud), high-bandwidth cloud analytics jobs, or disaster recovery failover to cloud.

4. SD-WAN Overlays

SD-WAN platforms (Cisco Meraki, Fortinet Fortigate, Arista CloudVision, Versa) allow you to multiplex traffic across multiple circuits to the colo (internet, private circuits, Direct Connect). The SD-WAN appliance intelligently routes traffic, provides application-level QoS, and fails over in milliseconds if one circuit fails.

Cost: platform licensing ($200–500/month depending on throughput), plus the cost of the underlying circuits. The advantage: you can use cheaper internet alongside expensive private circuits, and the SD-WAN layer makes sure mission-critical apps go over the reliable circuits while non-critical traffic uses internet.

When it makes sense: Organizations with multiple locations, cost-conscious operations that can tolerate variable latency for some workloads, or those migrating from MPLS to SD-WAN.

Economics of Cloud Connectivity from Your Colo

The decision to establish AWS Direct Connect, Azure ExpressRoute, or Google Cloud Interconnect from your colo is a pure cost-benefit calculation.

Internet transit: Equinix and most carriers will sell you internet bandwidth at $1–3/Mbps (oversubscribed, subject to fair use policies). A sustained 1 Gbps link (which you'll never fully use) costs $1,000–3,000/month. Data transfer over this link to AWS is charged at AWS's internet egress rate, roughly $0.02/GB.

AWS Direct Connect: Port fees are $0–500/month (many Equinix facilities waive this; smaller facilities may charge). Dedicated 1 Gbps connection is $200–400/month from AWS. Data transfer is $0.02–0.05/GB (cheaper for higher commit levels). No per-Mbps billing; you pay for what you use.

At 5 TB/month (roughly 50 Mbps sustained), Direct Connect costs roughly $300 (connection) + $0 (port) + $100 (transfer) = $400/month, vs. $2,500 (internet) + $100 (transfer) = $2,600/month on internet. Direct Connect wins by $2,200/month.

At 1 TB/month (10 Mbps), internet at $1,500 + $20 (transfer) = $1,520 might still beat Direct Connect at $300 + $20 = $320—until you account for the latency benefits and the reduced contention of a dedicated connection.

The break-even is typically between 5–10 TB/month. Above that, dedicated cloud connectivity is cost-effective.

Getting Direct Connect: The pathway is straightforward. (1) Identify a data center with AWS Direct Connect ports; most major Equinix, CyberRealty, and Switch facilities have them. (2) Request a port from the facility (often free or low-cost cross-connect). (3) Create a dedicated connection in AWS's console, which generates a LOA-CFA (Letter of Authorization and Connecting Facility Assignment). (4) Give the LOA to the facility; they patch your equipment to the AWS port. (5) Configure BGP peering on both sides. Total time: 2–4 weeks, assuming the port is available.

Building Redundancy into Colo Connectivity

A redundant colo deployment with single-path connectivity is a contradiction. You need to plan for carrier and path diversity.

Diverse Carrier Cross-Connects: Two carriers, entering the data center at different physical points. Equinix Portland, for example, has multiple carrier entry points (fiber conduits). You'd cross-connect to Carrier A's port from your cabinet, and to Carrier B's port from your cabinet, using two different cross-connects. Cost: $200–500/month per cross-connect (depending on facility and carrier). If Carrier A has an outage, traffic fails over to Carrier B, automatically or via BGP.

A word of caution: Ask the facility which carrier entry points are physically diverse. Some facilities claim two carriers but they both enter via the same conduit. If that conduit is cut, both carriers go down. Real diversity means separate physical paths, separate fiber bundles, and separate riser chambers. Equinix and CyberRealty typically design for this; smaller facilities sometimes don't.

Active-Active vs. Active-Standby: Active-active means both circuits carry traffic simultaneously, using ECMP (Equal-Cost Multi-Path) or BGP traffic engineering. Active-standby means one circuit is hot, the other is cold, and traffic switches only on failure. Active-active is more complex (you need to engineer load balancing) but extracts more value from your investment. Active-standby is simpler but leaves capacity on the table.

Power and Cooling Redundancy: This is the facility's responsibility, not yours, but you need to verify it. Facilities offer configurations: N+1 (if one power feed fails, the backup takes load), 2N (two independent power infrastructures; either can run the entire facility), or 2(N+1) (two independent 2N systems). For critical workloads, demand 2N. It costs more in facility rent, but it eliminates single points of failure in power and cooling.

Carrier-Neutral vs. Single-Carrier Data Centers

A carrier-neutral facility (Equinix, CoreSite, DataBank, CyberRealty) hosts multiple carriers and cloud providers. You can cross-connect to dozens of network operators, cloud providers, and peers. You're not locked in.

A single-carrier facility (AT&T, Lumen, Comcast facilities) is operated by the carrier. You can use that carrier's network easily, but you're limited if you want to cross-connect to a competitor's carrier or to AWS Direct Connect ports.

The cost premium for carrier-neutral: Rack pricing are typically 20–40% higher than in a carrier's own facility. At Equinix, a 10 kW cabinet in Portland might run $10,000–12,000/month. At an AT&T facility, it might run $7,000–8,000/month. But the flexibility often pays for itself: you can shop for connectivity from multiple carriers, you can establish direct cloud connections, and you're not captive to the host carrier's pricing.

For most organizations, carrier-neutral is worth the premium. You're buying long-term flexibility.

Hidden Costs in Colo Contracts

The quoted rack rate is rarely the true cost. Plan for:

  • Power Overage: Facilities bill power in increments (e.g., 2 kW, 5 kW, 10 kW). If your equipment draws 5.2 kW, you pay for the next tier. Overages are charged at $150–400/kW/month, 2–4x the base rate. Estimate your power draw conservatively; factor in headroom for future growth.
  • Remote Hands Labor: Need a technician to replace a power supply? Remote hands labor costs $75–200/hour, billed in 15-minute increments. A simple 30-minute task can run $300. Plan for quarterly or bi-annual hands-on maintenance; budget $2,000–5,000/year.
  • Cross-Connect Fees: $100–500/month per cross-connect, depending on distance and facility. A diverse, dual-carrier setup costs $300–1,000/month in cross-connect fees alone.
  • Cage/Cabinet Upgrades: Installing a new breaker panel, upgrading your PDU (Power Distribution Unit), or adding cooling infrastructure: $1,000–5,000 per upgrade. Budget for at least one upgrade during a 3-year lease.
  • Shipping and Receiving: Facilities charge $200–500 per shipment to receive, inspect, and load equipment into your cage. A server deployment with 20 boxes can cost $1,000–2,000 in receiving fees.

Total cost of deployment: A nominal $8,000/month 10 kW cabinet with dual-carrier connectivity, diverse cross-connects, and modest hands-on support often runs $12,000–14,000/month in year one, settling to $11,000–13,000/month after year one (no capital equipment shipping). That's a 40–60% premium over the rack rate. Plan accordingly in your budget.

Don't Let the Data Center Sell You Connectivity

Data center operators profit from selling you internet or connectivity as an add-on. They quote you the colo rack, then offer "convenient" connectivity at above-market rates. Get competing quotes from carriers who have cross-connects in the facility—you'll typically save 30–50% on bandwidth costs. A carrier like Zito or a regional provider will happily cross-connect to your cage for less than half what the facility charges.

ITG Perspective

The colo itself is usually the easy decision. The hard decisions are: which carriers connect to your colo, how do you get from your offices to your colo, and what does that network architecture look like at 3x your current bandwidth needs? We design colo footprints for 5–7 year horizons, accounting for Moore's Law, business growth, and carrier landscape changes. The connectivity layer requires constant attention and re-evaluation.

Frequently Asked Questions

Is colocation better than the cloud?

Neither is inherently better. Colocation is best for workloads requiring high bandwidth efficiency (financial services, media streaming, high-frequency trading), strict data residency requirements (European GDPR, healthcare HIPAA), or organizations with significant capital equipment investment (large databases, specialized hardware). Cloud (AWS, Azure, Google Cloud) is better for variable workloads, rapid scaling, and organizations that want to minimize infrastructure management. Many organizations use both: cloud for test/dev and burst capacity, colo for stable, high-bandwidth production workloads.

What's a cross-connect?

A cross-connect is a direct fiber connection between two network operators (or between your equipment and a carrier/cloud provider) inside the data center. It's typically a short run of dark fiber, patched at both ends, costing $100–500/month depending on the facility and distance. It's not shared; it's point-to-point. Equinix, for example, operates a cross-connect marketplace where you can order a connection to AWS, Zito, or another provider, and the facility handles the patching.

How do I get my offices connected to the colo?

Three main options: (1) A dedicated private circuit from a carrier (MPLS VPN or Ethernet over fiber, $500–5,000/month depending on distance and speed). Contact your local carrier or a carrier broker. (2) Internet access from the colo facility (easier, but more expensive and higher latency). (3) An SD-WAN overlay using multiple circuits (internet + private circuits, with intelligent routing). For redundancy, always use at least two carriers on two physically diverse paths. A single carrier to a single colo facility is a single point of failure.

What's AWS Direct Connect and do I need it?

AWS Direct Connect is a dedicated network connection from your office or colo directly to AWS, bypassing the public internet. Benefits: lower latency, consistent performance, lower egress costs ($0.02–0.05/GB vs. $0.02–0.12/GB over internet), and higher reliability. Costs: a 1 Gbps dedicated connection is $200–400/month from AWS, plus port fees ($0–500/month from the colo) and data transfer. If you're moving more than 5–10 TB/month to AWS, Direct Connect typically beats internet pricing. For smaller deployments or variable workloads, internet is fine. Azure ExpressRoute and Google Cloud Interconnect offer similar value propositions.

What should I look for in a colo SLA?

(1) Power redundancy: N+1 minimum, 2N preferred. (2) Uptime guarantee: 99.99% or better (no more than 4.38 hours/year of unplanned downtime). (3) Diverse carrier entry points: ensure at least two carriers can cross-connect from different physical entry points. (4) Transparent power and cooling measurement: real-time monitoring, capacity planning tools. (5) SLA credits: automatic, not claimed; credits should be large enough to hurt if the facility misses them (e.g., 10% of monthly rent per 9 of uptime missed). Don't accept SLAs with "best effort" language; demand hard numbers.

Getting Started

If you're evaluating a colo footprint or rearchitecting connectivity to an existing colo, start with these steps:

  1. Identify your peak bandwidth needs (office-to-colo, colo-to-cloud, colo-to-internet) over the next 5 years, not just today.
  2. Map out carrier availability in your target facilities. Which carriers have cross-connects? What's their pricing?
  3. Design for redundancy from day one: two circuits, two carriers, two entry points. Single-path colo is indefensible for any production workload.
  4. Get a detailed quote from the facility that breaks out rack, power, cross-connect, and support costs separately. Don't accept bundled pricing.
  5. Negotiate cross-connect pricing. Most facilities have room to move, especially if you're committing to a long-term lease and multiple cross-connects.
  6. Plan for hidden costs: remote hands, power overage, equipment upgrades. Budget 40–60% above the quoted rack rate.
  7. Review the SLA carefully. Hard uptime guarantees, power redundancy, and carrier diversity matter more than a low monthly rent.

The colo itself is straightforward infrastructure. The connectivity decisions are where real engineering happens—and where you'll save (or waste) significant money over the life of your deployment.