The short version
Choose the connection.
Understand the path.
Dedicated Internet Access, often called DIA, gives a business an Internet connection with defined service characteristics. The contract defines where capacity is reserved, what the provider promises, and where responsibility changes hands.
A dedicated access circuit can be the right choice for predictable business connectivity. It does not make the entire Internet private, exclusive, or immune to congestion.
This guide is written for business owners, IT teams, and anyone comparing proposals. It explains common concepts in plain language, with enough technical detail to ask better questions.
01 / The fundamentals
What does “dedicated” actually mean?
Generally, dedicated access means the customer has a specified amount of capacity on its access connection to the provider. Ask the provider exactly which portion is dedicated, in both directions, and whether that capacity is committed or a maximum.
Beyond the access circuit, traffic travels through networks and interconnections used by many organizations. Performance to a particular destination also depends on those networks and the destination itself.
02 / Compare the options
Dedicated or shared?
Both can be useful. The right choice depends on the application, budget, and the actual terms offered.
| Consideration | Dedicated access | Shared access |
|---|---|---|
| Capacity | Contract describes capacity on the access circuit; confirm its scope. | Capacity may be shared at one or more points; advertised speed is often a maximum. |
| Upload & download | Often symmetrical; verify both rates. | May be symmetrical or asymmetrical, depending on the service. |
| Commitments | Often includes an SLA and business support; terms vary. | Business support and SLAs may be available; check the contract. |
| Cost | Often more expensive, depending on location and terms. | Often more economical and suitable for many needs. |
Neither label alone tells you how a particular service performs. Compare the actual access design, support, installation, and contract.
03 / Capacity planning
How much bandwidth do you need?
Begin with what happens at the same time: video meetings, cloud applications, large file transfers, backups, and customer-facing services. Measure busy periods rather than relying only on employee count.
Measure both directions
Look at upload and download separately. Cloud backups and off-site replication can make upload capacity important.
Allow for growth
Consider peak demand, future locations and applications, and how easily the circuit can be upgraded.
Check the equipment
A 10 Gbps circuit will not deliver 10 Gbps through a 1 Gbps interface or an undersized firewall.
Test the full path
Performance varies with server capacity, latency, client hardware, test method, and other traffic.
A speed test is one observation, not a complete service assessment. For reliable comparisons, document the test endpoint, direction, time, and concurrent traffic.
Find your starting point.
Move the sliders to see an estimate for your business. You can also enter an exact number.
How is this estimated?
We add a busy-period allowance per employee (5 Mbps download and 2 Mbps upload), simultaneous video call (4 Mbps each way), and concurrent guest (2 Mbps download and 0.5 Mbps upload). We add the selected server and AI inference allowances and 30% headroom to those workloads. We then add the selected training and data transfer capacity as entered, without extra headroom. Employee and guest allowances exclude the separately counted video calls.
Server allowances (download/upload) are light: 50/50 Mbps; moderate: 250/250 Mbps; heavy: 1,000/1,000 Mbps. AI inference allowances are light: 25/10 Mbps; moderate: 100/50 Mbps; heavy: 500/250 Mbps. AI training and transfer allowances are 0, 50, 250, 1,000, 5,000, 10,000, 40,000, or 100,000 Mbps in each direction, based on the selected level. These are illustrative, not measurements of your actual workloads.
Workloads and peak concurrency vary. Training compute requirements cannot be converted directly into Internet bandwidth. Large datasets and model transfers can be much larger than these allowances; measure your traffic and review your timeline and equipment capacity before ordering.
For time-sensitive AI, the estimate alone is not enough. Consider data volume, desired transfer time, round-trip latency, and whether the AI service processes data locally or remotely. Faster links can shorten large transfers and reduce delay from congestion; they do not automatically reduce propagation delay or model processing time.
AI & network performance
Fast answers need more than bandwidth.
For interactive AI, the time to receive an answer can include sending a request, network round trips, waiting for the service to process it, and receiving the response. The importance of each part depends on the application and where the AI system is running.
Latency
Round-trip time measures how long a signal takes to travel to a destination and back. Distance, routing, and congestion can affect it. A higher-speed link does not by itself shorten the physical distance or guarantee a lower round-trip time.
Bandwidth
More available capacity can shorten large data and model transfers and reduce queueing when a link would otherwise be busy. For a small interactive request on an uncongested connection, a faster link may make little difference.
For planning, measure round-trip latency to the actual AI service, expected data volume, the time allowed for transfers, and peak concurrent demand. Training time also depends on compute, storage, and how the training system exchanges data; Internet bandwidth is only one part of that picture.
04 / Service commitments
Read past the percentage.
An SLA describes a provider’s contractual commitments and remedies. It is not a promise that outages cannot occur. The details determine what a percentage means in practice.
Illustrative downtime in a 30-day month, calculated from the stated percentage. Actual SLAs can define availability, exclusions, measurement and credits differently.
- What service and endpoints are measured?
- Does the SLA cover the access circuit, latency, packet loss, or repair time?
- How are scheduled maintenance and third-party issues treated?
- Are credits automatic, or must the customer request them?
05 / Continuity
Two circuits are not always two paths.
Redundancy only helps when you understand what is independent. Two services can share a building entrance, conduit, power supply, switch, or upstream path, even when purchased from different companies.
Consider the whole chain: diverse building entries where feasible, separate equipment and power, a tested failover plan, and the routing design needed to use it. The right level of redundancy depends on the cost of an interruption.
06 / Equipment & routing
Know where responsibility changes hands.
The handoff is the interface where the provider’s service connects to your equipment. Confirm the location, interface speed, connector or optics, and whether the provider supplies any equipment.
For the IT team
Check port speed, interface type, expected traffic, and firewall throughput with the security features you intend to enable. Confirm IPv4 and IPv6 addressing, static routes or BGP, and any address assignments.
For the buyer
Ask who owns, configures, monitors, and supports each device. Clarify whether managed firewall or router service is included, optional, or your responsibility.
A high-speed port alone does not establish that a device can inspect, encrypt, or route at that rate. Review realistic performance specifications for the intended configuration.
07 / From order to service
Plan for the building, not just the circuit.
Installation can involve a site survey, building authorization, access to the equipment room, a path for fiber, and coordination with the customer’s IT team. Timelines vary based on the building and construction required.
- 01
Confirm the location
Identify the building, floor, suite, handoff location, and site contact.
- 02
Review the scope
Ask about access, risers, construction, pricing, permits, and any conditions that could change the schedule.
- 03
Test and document
Confirm the handoff, addressing, agreed service characteristics, and support contacts. Record the results and any open items.
08 / Take this with you
Questions for any provider.
Use this checklist to compare proposals on the same terms.
- What capacity is committed, and where does the commitment apply?
- What are the upload and download rates, and can they be upgraded?
- What does the SLA measure, exclude, and provide if a commitment is missed?
- What are the installation scope, total charges, contract term, and renewal terms?
- What physical diversity is available and independently verifiable?
- What interface, optics, addresses, and routing options are included?
- Who provides support, how is it reached, and what information is needed?
- What testing and documentation are provided at service acceptance?
About this brief
Built to be useful.
Prepared by Stealth Communications, a provider of business connectivity. This guide helps readers evaluate service from any provider. Technical features and contractual commitments vary; confirm them in the applicable agreement.
For further reading: RFC 2330: Framework for IP Performance Metrics RFC 6349: Framework for TCP Throughput Testing, and Cisco: AI/ML network delay and performance.
Published October 31, 2026.