What Electricity Costs to Run Local AI
private ai

What Electricity Costs to Run Local AI

Put a computer in the office that replaces a cloud service and someone sensible will ask what it does to the power bill. It is the right question, and for modern local AI hardware it has a surprisingly small answer.

Run the box at its absolute ceiling every hour of the year and the electricity comes to about $285. Here is that math, with nothing hidden in it.

The Whole Box Draws 240 Watts

NVIDIA's DGX Spark, the class of hardware this work runs on, ships with a 240-watt power supply, and that is the ceiling for everything inside it.1 The GB10 chip doing the actual AI work is rated at 140 watts flat out.1

For scale, 240 watts is two bright incandescent bulbs. It is a fraction of what a single gaming desktop is built to draw.

The Worst Case Math

Assume the impossible: the box pinned at its full 240 watts, 24 hours a day, 365 days a year. That is about 2,100 kilowatt-hours.2

At the U.S. average commercial electricity rate of roughly 13.5 cents per kilowatt-hour, the year comes to about $285, or under $24 a month.2 That is the number to write into the budget, because reality cannot exceed it.

Real Usage Runs Lower

The 240-watt figure is the power supply, not the workload. Even the compute chip running at its full 140-watt rating every hour of the year would come to about $165.1, 2

Real systems breathe: chat is bursty, overnight jobs grind and finish, and change-gated ingestion skips every file that has not been touched. The worst case is the budgeting number; the actual bill lands beneath it.

No Special Wiring Needed

A 240-watt appliance plugs into a standard wall outlet on an ordinary office circuit. There is no server room, no dedicated 240-volt line, no cooling retrofit, and no electrician in the project plan.

Power and cooling are real line items for datacenter GPU racks. A desktop-class AI box simply is not in that category, which is part of why the hardware stays useful for years without facility upgrades.

Power Against the Alternative

One Microsoft 365 Copilot seat runs $30 per user per month, or $360 a year.3 Powering the entire box, worst case, for every user in the office, is less than that one seat.

Electricity is the classic gotcha raised against owning hardware, and this is why it belongs in the total cost of ownership math rather than in the argument. As an operating expense, it rounds to a coffee budget.

Owning the box has real considerations, from depreciation to the occasional queue at peak. If the power bill was the objection holding the decision back, what is actually left?

Recent Articles

How AI Hardware Depreciation Works

A $50,000 AI appliance is 5-year MACRS property that Section 179 or permanent bonus depreciation lets most businesses expense in full in year one.

How Long AI Hardware Stays Useful

The fear that an AI appliance goes obsolete in eighteen months has the mechanism backwards: better models keep arriving as software for the same silicon.

Does Private AI Need an IT Department

The objection to on-prem AI is staffing, not price. What a well-built system runs itself, the minutes a week a human spends, and what still needs a plan.

Budgeting for Private AI Growth

An honest private AI budget has two numbers: a defined, all-in entry and a growth path that is neither free nor unlimited, planned without surprise invoices.

Calculating ROI on Private AI

ROI on private AI is two columns: cost to own, and return in time, risk, and capability. A worked look at why honest math favors ownership.

How Many Staff One AI Box Supports

An on-premise AI box comfortably serves ten to twenty active users at once. The math, the workload mix, and how the queue keeps peaks soft.

CAPEX Versus OPEX for AI

AI lands on your books two ways: a rented subscription or an owned asset. Why the accounting, not the features, decides the real cost over time.

Total Cost of Ownership Local AI vs Cloud

Most AI TCO models are built by vendors selling subscriptions. This one is built from the buyer's side, with five years of numbers on both lines.

Popular Articles

CAPEX Versus OPEX for AI

AI lands on your books two ways: a rented subscription or an owned asset. Why the accounting, not the features, decides the real cost over time.

Budgeting for Private AI Growth

An honest private AI budget has two numbers: a defined, all-in entry and a growth path that is neither free nor unlimited, planned without surprise invoices.

Calculating ROI on Private AI

ROI on private AI is two columns: cost to own, and return in time, risk, and capability. A worked look at why honest math favors ownership.

Total Cost of Ownership Local AI vs Cloud

Most AI TCO models are built by vendors selling subscriptions. This one is built from the buyer's side, with five years of numbers on both lines.

How Many Staff One AI Box Supports

An on-premise AI box comfortably serves ten to twenty active users at once. The math, the workload mix, and how the queue keeps peaks soft.

How AI Hardware Depreciation Works

A $50,000 AI appliance is 5-year MACRS property that Section 179 or permanent bonus depreciation lets most businesses expense in full in year one.

Does Private AI Need an IT Department

The objection to on-prem AI is staffing, not price. What a well-built system runs itself, the minutes a week a human spends, and what still needs a plan.

How Long AI Hardware Stays Useful

The fear that an AI appliance goes obsolete in eighteen months has the mechanism backwards: better models keep arriving as software for the same silicon.

Other Categories