AI Server Power & Cooling Requirements for Your Office
Before you buy an AI server, the practical question is whether your building can physically run it. The good news for most Texas small businesses: a single-GPU server is fine on a normal office circuit. The boxes that need real planning are the multi-GPU and data-center-class builds. This is the plain-English guide to power, amps, circuits, and cooling — and TIS plans all of it on the on-site visit, before a single part is ordered.
How much power an AI server actually draws
The GPU is the big number. A current pro card like the RTX PRO 6000 Blackwell has a 600W power rating at full load, and that one figure dominates the math. Add the CPU, then roughly 20–30% overhead for the motherboard, memory, drives, and fans, and you have the server's real continuous draw.
So a single 600W-GPU build sits in the rough neighborhood of 900–1,100W at the wall under sustained load — meaningful, but not dramatic. Stack two or four of those cards and the picture changes fast, which is where dedicated power and serious cooling enter the conversation. These are rules of thumb, not a stamped spec; TIS measures the real numbers for your exact build and plans around them.
Want the part driving all of this? See how the card you choose sets the draw on GPU AI servers, or the whole spec on custom AI servers.
Build size, watts, amps, and circuit
A rough planning map from build size to the circuit it wants. All figures are continuous-load ranges, not guarantees — final numbers come from your real spec. A 20A/240V circuit gives roughly 1,920W continuous to compare against.
| Build size | Rough watts (continuous) | Rough amps | Circuit it wants |
|---|---|---|---|
| Single GPU (one 600W card) | ~900-1,100W | ~8-9A @ 120V | Normal 15/20A 120V circuit |
| Dual GPU (2 x 600W) | ~1,600-2,000W | ~13-17A @ 120V | Dedicated 20A circuit; 240V preferred |
| Quad GPU / workstation-class | ~2,400-3,200W | ~10-13A @ 240V | Dedicated 240V circuit |
| Data-center / 8-GPU box | ~4,000W and up | ~17A+ @ 240V | Dedicated 240V / 30A+ (often rack PDU) |
The takeaway: a single 600W-GPU build runs on a normal circuit, while multi-GPU and data-center boxes need dedicated 240V power. Form factor plays in too — see rack vs tower.
PSU sizing and redundancy
Size with headroom
The power supply should have comfortable headroom above the server's peak draw, not just barely cover it. GPUs spike above their rated wattage in brief bursts, and a PSU run flat-out runs hot and short-lived. A 2×600W build, for instance, wants a supply rated well past the cards' combined peak so it loafs rather than strains.
Why redundancy matters
For multi-GPU servers a business depends on, redundant dual PSUs are worth it: if one supply fails, the second keeps the server running instead of dropping it mid-workday. It is the same logic as RAID for drives — a single point of failure on a box the whole team uses is a risk worth designing out. We spec single or redundant power to match how critical the server is to your operation.
Cooling — the part people forget
Here is the rule that catches people out: heat equals watts. Almost every watt a server pulls comes back out as heat into the room. A server drawing 2.4kW is dumping roughly 2.4kW of heat into whatever closet it sits in, and ambient office air conditioning is not built to carry that away.
The plain rule of thumb: a single roughly 2.4kW server can want about a 10,000 BTU mini-split dedicated to its space. That is far short of a data center, but far more than "it'll be fine in the corner." For a single-GPU build the load is lighter and sometimes a well-ventilated closet handles it; for anything denser, dedicated cooling is the difference between a server that runs for years and one that thermal-throttles or fails early. TIS plans the cooling on the on-site visit, against your real space and the real heat your build makes.
Airflow: tower in a closet vs rack with front-to-back
Cooling capacity only helps if the air actually moves through the machine. A tower in a closet needs clearance and a path for warm air to leave — a sealed cabinet with no ventilation will cook a server no matter how cold the building is. A few inches of breathing room and an open or vented door usually does it.
A rack is built for density with front-to-back airflow: cool air in the front, hot air out the back, ideally into a return path rather than recirculating. That is what makes racks the right call once you have several GPUs in one place. When ambient AC is not enough — and for any multi-GPU build it usually is not — that is the signal to plan dedicated cooling and deliberate airflow rather than hope. See where a tower or rack fits your space in rack vs tower.
Pre-install electrical & cooling checklist
Walk this before the server arrives. TIS runs the same list on the on-site visit, but it helps to know what we will be looking at.
Know the server's real draw
Total GPU + CPU + overhead so you are planning against a real wattage, not a guess.
Identify the circuit
Find the exact circuit and breaker the server will live on, and what else shares it.
Confirm dedicated vs shared
A single-GPU build can share a normal circuit; multi-GPU needs a dedicated 240V line.
Check panel capacity
Make sure the electrical panel has room for a new circuit if one is needed.
Pick the location
Choose the closet, rack, or room — with clearance and a path for hot air to leave.
Plan the cooling
Match cooling to the heat: a mini-split for a closet that ambient AC cannot carry.
Plan airflow
Front-to-back for a rack; clearance and ventilation for a tower in a closet.
Consider a UPS
Protect against dips and brief outages so the server shuts down cleanly, not abruptly.
When you need an electrician (and when you don't)
When you usually don't: a single-GPU server plugging into an existing, appropriately-rated dedicated circuit is typically a plug-and-go install. If the outlet and breaker already suit the load, there is nothing electrical to change.
When you do: any time a new dedicated circuit is needed, a 240V line has to be run, or the panel is near capacity, that is licensed-electrician work — full stop. Multi-GPU and data-center-class builds almost always fall here. We do not freelance the electrical; we identify exactly what is required during planning so you can line up a licensed electrician before install day, with no surprises.
This is precisely the on-site planning TIS handles end to end. See how we run the install on AI server installation.
We plan the power and cooling on-site, across Fort Bend County
Before any hardware is ordered, we walk your space in Katy, Fulshear, Sugar Land and across the Houston metro — read the panel, size the circuit, and plan the cooling for the build you actually need. The team that plans it is the team that installs it. See our Texas service areas.
Power & cooling questions
Will a standard office circuit run an AI server?+
Usually yes for a single-GPU build. A normal 20A/120V circuit delivers roughly 1,920W of continuous power, and a server built around one 600W GPU draws well under that. Multi-GPU and data-center-class boxes are the ones that often need a dedicated 240V circuit at 30A or more — we confirm which camp you are in during the on-site visit.
Do I need special cooling or a server room?+
Often a closet plus a mini-split is enough, not a full data center. The rule of thumb is that watts become heat, so a single roughly 2.4kW server can want about a 10,000 BTU mini-split to stay in range. We plan the cooling around your actual space on the install visit.
How many amps does an AI server draw?+
It depends on the build. A single 600W-GPU server plus CPU and overhead lands in the low single digits of amps on 240V, comfortably inside a normal circuit. Multi-GPU builds climb quickly and are why dense boxes get a dedicated circuit. We size the exact amperage to your final spec.
Do I need a 240V circuit for an AI server?+
Not for most single-GPU office builds, which run fine on a standard 120V circuit. Dedicated 240V at 30A or more is for multi-GPU and data-center-class servers that pull too much for a shared 120V line. We tell you which you need before anything is ordered.
Will an AI server need an electrician?+
A single-GPU build on an existing dedicated circuit usually does not. You want a licensed electrician when a new dedicated circuit, a 240V line, or panel capacity is involved. We flag exactly what is needed during planning so there are no surprises on install day.
Next, pick the form factor in rack vs tower, or back to AI Servers.
Let's check if your office can run it
Tell us your space and your workload — we'll plan the power, the circuit, and the cooling on-site before a single part is ordered.