How traffic reaches AWS over an Equinix Fabric, why the Hosted-vs-Dedicated choice matters, and when each one is the right call.
The one question underneath everything
Every decision here comes down to a single question:
Whose infrastructure carries your traffic to AWS — Equinix’s shared backbone, or a physical path that’s entirely yours?
- Hosted = your traffic rides Equinix’s existing, shared connection to AWS. Equinix carves you a slice.
- Dedicated = AWS assigns you a physical port. You reach it over your own path. No shared middleman.
Everything else, visibility, cost, provisioning time, whether you need new ports, falls out of that one fact.
Two things people confuse
Separate two questions that get tangled together constantly:
| Question | Answer options | What it controls |
|---|---|---|
| How is the connection provisioned? | Hosted vs. Dedicated | Who owns the AWS-facing physical port |
| How much bandwidth does it get? | 2G, 5G, 10G, up to 25G | The contracted speed of that connection |
These are independent. “Hosted” is not a speed, and “10G” is not a provisioning model. You can have a 2G Hosted connection and a 10G Hosted connection on the same port.
The bandwidth model (the highway and the lanes)
Your 100G interface is the physical Equinix Fabric access port, the highway. It is not a “100G Hosted connection.” Each connection to AWS is a lane carried over that highway, and each lane has its own contracted speed:
HundredGigE0/1/0 — 100G physical Equinix Fabric port (the highway) | +-- VLAN 10 — AWS East 1 Prod — 10G (a lane) +-- VLAN 20 — AWS West 1 Prod — 10G (a lane) +-- VLAN 30 — Azure East Prod — 10G (a lane) +-- VLAN 40 — Azure West Prod — 10G (a lane)
The parent Cisco interface stays configured as 100G because that’s its physical Ethernet speed. Each virtual connection on it can have a lower contracted bandwidth.
“Hosted 25G maximum” means: one individual AWS Hosted connection can’t exceed 25 Gbps. It does not mean your Fabric port must be 25G or smaller, a 100G Fabric port can aggregate many connections, subject to Equinix’s capacity and oversubscription rules.
Cisco doesn’t automatically know that subinterface .10 is commercially capped at 10G. That limit is enforced in the Equinix/AWS service unless you also configure local QoS shaping.
What Hosted vs. Dedicated actually means
Think about who owns the physical port on the AWS side.
Hosted — Equinix already has large physical ports plugged into AWS. When you want a connection, Equinix carves out a slice of their port and hands it to you as a VLAN. You never touch AWS hardware. You’re a tenant in Equinix’s connection.
Dedicated — AWS assigns you a whole physical port. You order it directly from AWS, and you (or the facility) run a cross-connect to it. It’s yours.
That’s the entire difference. VLANs, IP addressing, BGP peers, routing policy, and your Cisco config are identical between the two.
| Area | Hosted DX | Dedicated DX |
|---|---|---|
| AWS-facing connection provisioned by | Equinix (a DX Partner) | You, directly from AWS |
| Physical AWS port associated with | Partner infrastructure | Your company |
| How the connection is created | Partner provisions it; you accept | You order it in your AWS account |
| Physical-layer troubleshooting | Often needs Equinix involvement | More direct visibility + AWS support |
| Port-level telemetry | More partner-dependent | Direct metrics under your AWS account |
| Cisco VLAN / BGP config | Essentially the same | Essentially the same |
The two traffic flows
The top of both paths is identical — same router, same subinterfaces, same BGP. That’s why the config never changes. The paths split in the middle: Hosted rides infrastructure that already exists; Dedicated requires net-new physical construction. 
Hosted rides the port and backbone whose handoff you already own — nothing new gets built. Equinix carves you a slice and you accept it.
Dedicated requires three net-new physical things:
- A new physical port on your side
- An LOA-CFA (Letter of Authorization – Connecting Facility Assignment) from AWS, telling the facility exactly which AWS port to wire you to
- A cross-connect to execute that wiring to the AWS-assigned port
You choose per physical connection, not per VLAN
This is the point most conversations miss.
Your existing 100G Fabric ports are cabled into Equinix’s infrastructure. Every VLAN on those ports, a 2G Non-Prod, a 10G Prod, rides into Equinix’s Fabric. That’s what makes it Hosted: the far end of the cable is Equinix’s equipment.
You cannot make one VLAN on that Fabric port “Dedicated.” Dedicated means the cable lands on an AWS-owned port assigned to you and the Fabric port doesn’t land there. It lands on Equinix.
Your 100G Fabric port -> Equinix Fabric -> AWS = everything HOSTED
(all VLANs)
A separate new port -> cross-connect -> your own AWS port = DEDICATED
(separate physical path)
So can you run both? Yes — but not on the same trunk. Keeping 2G Hosted Non-Prod while adding 10G Dedicated Prod would require a separate physical connection (new port + cross-connect to an AWS-assigned port). Two separate physical paths.
The rule: the Hosted-vs-Dedicated choice is made per physical connection, not per VLAN. All VLANs on a given port inherit whatever that port’s far end is.
Why choose Dedicated over Hosted?
The reasons, roughly in order of how often they actually matter:
- Visibility / troubleshooting ownership — the big one for Prod. With Dedicated, the AWS port is yours, so you get direct port telemetry (state, error counters, optical readings on supported speeds) and a direct AWS support line. With Hosted, if the fault is inside Equinix’s shared backbone, you see only symptoms and wait on Equinix.
- Not sharing a backbone — Dedicated is your own isolated path, which can matter for contention, performance, or compliance reasons.
- Large, stable bandwidth — if you need a full 10G or 100G port to yourself, a Dedicated port can be more economical at scale than stacking large Hosted connections.
And the two factors people ask about first:
- Time — Dedicated is slower to stand up. Equinix’s AWS ports are already live, so Hosted provisions quickly; Dedicated means ordering a port from AWS and waiting on a cross-connect install. Time argues for Hosted.
- Cost — For adding capacity to links you already run, Hosted is almost certainly cheaper: no new ports, no cross-connect, no build effort. Dedicated port pricing can win only at very high sustained bandwidth.
Scorecard
| Factor | Winner |
|---|---|
| Time to provision | Hosted (Equinix ports already live) |
| Cost | Hosted (no new ports / cross-connect) |
| Physical-layer visibility | Dedicated |
| Not sharing backbone | Dedicated |
| Fits existing 100G Fabric | Hosted (Dedicated needs new ports) |
The only column where Dedicated clearly wins is visibility/ownership. So the real question is: is that visibility gain worth new ports, a cross-connect, more time, and more cost, just to add Prod?
The visibility tradeoff, stated precisely
With Hosted, the end-to-end path contains infrastructure you don’t own and can’t fully inspect:
Cisco router -> Your 100G Equinix Fabric port -> Equinix Fabric -> Equinix AWS interconnect -> AWS Direct Connect
You can troubleshoot your side: Cisco interface state, optics, CRC/drops, subinterface and VLAN config, ARP, BFD/BGP state, route advertisements, VIF-level metrics, and your customer-facing Fabric port (depending on the Equinix portal).
But if the fault is inside Equinix’s Fabric or on its AWS-facing interconnect, you may see only symptoms: BGP down, bidirectional loss, intermittent latency, reduced throughput, VIF/connection state changes. You generally can’t inspect the partner-owned AWS-facing port’s optics, errors, utilization, or internal Layer 2 forwarding. Equinix must investigate that portion and may escalate to AWS.
With Dedicated, AWS assigns the physical port to you, giving a clearer troubleshooting boundary and direct AWS-side telemetry (including optical readings on supported speeds):
Cisco router -> Cross-connect -> Your assigned AWS Dedicated DX port -> AWS
The honest ceiling: even Dedicated does not give you command-line access to AWS’s internal switching, and you still can’t inspect AWS’s internal Layer 2 tables. You get substantially better direct telemetry and a clearer support relationship for the physical connection — not unlimited visibility.
| Layer | Hosted DX | Dedicated DX |
|---|---|---|
| Cisco interface and optics | You troubleshoot | You troubleshoot |
| Customer-facing Equinix port | You / Equinix | You / facility provider |
| Partner AWS interconnect | Equinix must investigate | Not part of the path |
| AWS DX physical-port metrics | Limited / partner-dependent | Tied to your connection |
| BGP and VIF troubleshooting | Available | Available |
| AWS-side optical / error telemetry | Generally partner-owned | Available on supported ports |
| Support path | You → Equinix → possibly AWS | You → AWS (+ facility) |
What stays the same either way (the config)
Hosted and Dedicated look identical at the Cisco VLAN and BGP layer. Adding a new Region is just another tagged subinterface on either model:
interface HundredGigE0/1/0 description EQUINIX-FABRIC-100G no ip address interface HundredGigE0/1/0.10 description AWS-EAST-DX-10G encapsulation dot1q 10 ip address 169.254.x.x 255.255.255.252 interface HundredGigE0/1/0.20 description AWS-WEST-DX-10G encapsulation dot1q 20 ip address 169.254.y.y 255.255.255.252
Two real differences hide behind “same config,” though, if you move to a Dedicated port:
- You create the VIFs. With Hosted, Equinix provisions each connection and you accept it. On a Dedicated port, AWS gives you the physical port and you create the virtual interfaces yourself for each Region / VGW / Direct Connect Gateway.
- Bandwidth is shared, not per-lane capped. A Dedicated port is a flat physical pipe (e.g. 10G or 100G); every VIF you add draws from that same total, with no partner-enforced per-connection cap. Stack several Regions on one Dedicated 10G port and they share 10G, you manage contention yourself with QoS/shaping.
Bottom line
- If you’re already on Equinix Fabric with 100G uplinks and Non-Prod already runs Hosted over them, adding Prod to those same links means it is physically Hosted too — those ports land on Equinix, and everything on them is Hosted by definition.
- Going Dedicated for Prod isn’t a config change or another VLAN; it’s a separate build (new port + LOA + cross-connect to an AWS-assigned port).
- Hosted wins on time, cost, and fit. Dedicated wins on physical-layer visibility.
- The one thing to make a conscious, documented decision: choosing Hosted for Prod means accepting dependence on Equinix if a fault sits inside their backbone. That’s a defensible trade — just put it on record so nobody’s surprised during a murky Prod incident.
A clarifying note on units: these are 2G / 10G connections — gigabit per second (Gbps), not gigabytes (GB).