There is a lot of marketing at the moment suggesting that AI adoption requires businesses to rebuild their network infrastructure. For most offices, that is not true, and it is worth being clear about why before anyone spends money on it.
What is true is more specific: AI changes where demand concentrates, and it changes the comms room more than it changes the desk. Understanding that distinction is what separates a sensible infrastructure decision from an expensive one.
What most AI use actually does to a network
The dominant pattern for AI in ordinary offices is cloud-based. Staff use hosted services through a browser or an application, and the processing happens elsewhere.
From a network perspective, that traffic looks unremarkable. Text-based interactions with a hosted service consume very little bandwidth — considerably less than a video call, and vastly less than a large file transfer. A hundred employees using cloud AI tools all day does not stress a well-designed office network.
What it does stress is the internet connection, not the internal cabling. If AI adoption creates a bottleneck in a typical office, it will be at the WAN link, and the fix is bandwidth or a better connection, not recabling.
So the first honest answer: if your organisation’s AI usage is cloud services accessed by staff at desks, your horizontal cabling requirement has not materially changed. Do not let anyone tell you otherwise.
Where the requirement genuinely changes
There are specific situations where the picture is different, and they are worth identifying because they are real:
On-premises AI workloads. Organisations running models locally — for data sensitivity, latency, cost, or regulatory reasons — have a genuinely different infrastructure profile. This is concentrated in the comms or server room rather than distributed to desks.
AI-assisted creative and media work. Video production, 3D rendering, large image datasets, engineering simulation. The AI element increases file sizes and iteration frequency, and the bandwidth requirement is at specific workstations, not everywhere.
Large local datasets. If the data being processed lives on-premises and is substantial, moving it between storage and compute is the demanding operation.
Increased video and collaboration density. Somewhat adjacent to AI, but real: meeting rooms with higher-quality video, more simultaneous streams, and AI-assisted transcription and analysis running alongside. This concentrates demand in meeting spaces.
Notice the pattern. In every case, demand concentrates at identifiable points — a server room, a handful of workstations, meeting rooms. It does not spread evenly across every desk. That is the design insight, and it argues for targeted provision rather than uniform over-specification.
The comms room is the real constraint
This is the part that gets least attention and causes the most problems.
Power. Higher-performance networking and any on-premises compute draw substantially more power than the equipment they replace. Many existing office comms rooms were provisioned for a switch stack, a firewall, and perhaps a small server. Adding meaningful compute can exceed the available circuit capacity.
Check what your comms room circuit actually supports, what it currently draws, and what headroom exists. This is a straightforward measurement and it frequently produces a surprise.
Cooling. Every watt consumed becomes heat. In Dubai this is a year-round load, not a seasonal one, and comms rooms in commercial towers are often cooled by whatever the base building provides rather than by dedicated cooling sized for the actual load.
An under-cooled comms room does not fail dramatically. It degrades — equipment runs hot, lifespan shortens, and thermal throttling reduces performance in ways that get misdiagnosed as network problems for months.
UPS capacity. Higher power draw means shorter runtime on existing UPS capacity. A UPS sized for fifteen minutes of runtime three years ago may deliver four now. Worth actually measuring rather than assuming.
Physical space. Rack units, depth clearance, and cable management. Higher-performance equipment is often physically larger and needs more clearance for airflow.
The honest summary: for the large majority of offices, if AI drives an infrastructure investment, that investment belongs in comms room power, cooling, and UPS rather than in horizontal cabling. That is a less exciting recommendation than recabling the floor, but it is the one that addresses the actual constraint.
What to specify in a new fit-out
If you are fitting out new space, a few decisions cost little now and preserve options later:
Backbone with headroom. Fibre between comms rooms and floors, specified generously. Backbone is where aggregate demand lands, it is the most disruptive element to upgrade later, and the incremental cost of higher-capacity fibre at installation is small relative to replacing it.
Comms room provision sized beyond current need. Power circuit capacity, cooling, and floor space. This is the single most valuable piece of future-proofing available, and the hardest to retrofit. If you specify one thing generously, specify this.
Targeted high-capacity provision. Identify the positions that genuinely need it — media workstations, engineering positions, lab equipment — and provide for them specifically rather than uplifting the whole floor.
Spare containment capacity. Tray and riser space you are not using. Adding cable through existing spare capacity is straightforward; adding containment through a finished ceiling is not.
Spare drops and conduit. The cheapest drop is installed during first fix. Adding twenty per cent spare capacity during the fit-out costs a fraction of adding one drop afterwards.
Wireless density planning based on a survey. Access point placement should follow a proper survey of the actual floor plate, materials, and expected density — not a uniform grid. Modern wireless standards deliver meaningful capacity, but only with correct placement and adequate backhaul to each access point.
What you generally do not need: uniformly higher-specification horizontal cabling to every desk position on the assumption that AI will require it. That is the expensive answer to a problem most offices do not have.
Wireless deserves specific attention
For AI-era offices, wireless is often more consequential than horizontal cabling, because that is where most staff devices actually connect.
Three things determine whether wireless performs:
Access point placement, based on a survey rather than assumption. Dubai office fit-outs frequently involve glass partitioning, which behaves differently from drywall and is routinely underestimated in coverage planning.
Backhaul to each access point. A high-capacity access point on an inadequate uplink is limited by the uplink. This is a cabling decision, and it is where cabling specification genuinely does interact with modern requirements.
Power delivery. Higher-capability access points draw more power, which affects PoE budget and switch selection. Confirm your switching can actually deliver the required power class to every access point simultaneously, not just to some of them.
Network security implications
Worth raising because it interacts with design decisions.
AI adoption typically means more data moving between the organisation and external services, and often more devices and services connecting outward. That has implications for segmentation, traffic visibility, and data governance — particularly where the data being sent externally includes anything sensitive or personal.
Under the UAE’s data protection framework, sending personal data to external AI services raises questions about lawful basis, processing location, and cross-border transfer that should be resolved by whoever handles compliance rather than settled by default when someone signs up for a service.
From an infrastructure perspective, the practical implications are segmentation so that different classes of traffic are separated, and visibility so you know what is actually leaving the network.
A realistic sequence
If you are trying to decide what to do:
- Establish what your AI usage actually looks like. Cloud services at desks, or on-premises workloads? These have entirely different infrastructure implications.
- Measure your internet link utilisation. For most cloud-AI offices this is the real constraint, and it is measurable rather than theoretical.
- Assess the comms room — power headroom, cooling adequacy, UPS runtime, physical space. This is where investment is most likely warranted.
- Identify the specific positions with genuine high-bandwidth requirements, and provide for those rather than uniformly.
- Survey your wireless if staff work primarily on wireless devices.
- Only then consider horizontal cabling, and only where the assessment above identified a need.
Running this sequence usually produces a much smaller and better-targeted scope than starting from “we need AI-ready infrastructure.
The summary worth remembering
AI does not require most offices to rebuild their cabling. It requires them to look honestly at their internet link, their comms room power and cooling, their wireless design, and the handful of positions doing genuinely demanding work.
An organisation that spends its budget on comms room capacity and a proper wireless survey will be better positioned than one that spends the same money uplifting horizontal cabling to every desk. The first addresses real constraints; the second addresses a constraint that mostly exists in marketing material.
Planning network infrastructure?
KNH IT designs and installs structured cabling and network infrastructure for offices across Dubai. We will tell you what your requirement actually is — including when the answer is that your existing cabling is fine and the investment belongs elsewhere.
Contact: Al Raffa Building, Office 07, Bur Dubai , info@knhit.com , contact us
FAQ
? Usually not. Most office AI usage involves cloud services accessed by staff, which consumes relatively little bandwidth — less than video calls. If a bottleneck appears it is typically at the internet connection rather than the internal cabling. On-premises AI workloads are a different case and concentrate demand in the server room.
Primarily the comms room: power circuit capacity, cooling load, UPS runtime, and physical space. Higherperformance equipment and any on-premises compute draw substantially more power than what many existing comms rooms were provisioned for.
Generous fibre backbone between comms rooms and floors, comms room power and cooling sized beyond current
need, spare containment capacity, spare drops during first fix, targeted high-capacity provision at specific demanding positions, and wireless access point placement based on an actual survey.
For most staff devices, wireless carries the traffic, making access point placement, adequate backhaul cabling to each access
point, and sufficient PoE power budget more consequential than uplifting horizontal cabling to every desk.
Cooling is a year-round load rather than seasonal, and comms rooms in commercial towers are often cooled by base building
provision rather than dedicated cooling sized for the actual equipment load. Under-cooled rooms cause shortened equipment life and thermal throttling that is frequently misdiagnosed as a network fault
