AI for Electrical Load Calculations: The Method and Where AI Helps

Last updated: August 2026

An electrical load calculation decides how big your service and feeders need to be: you total the connected loads, apply the code’s demand factors, and size the service to what the installation will actually draw. It is structured, repetitive arithmetic against a code book, which is exactly why engineers ask whether AI can do it. The honest answer is that a language model can genuinely speed up AI for electrical load calculations as a method assistant, but it must never be trusted to supply the code values themselves. This guide walks the method, shows where AI helps, and marks the places where a hallucinated demand factor could undersize a service.

It sits under our roundup of AI tools for electrical engineers, and it is the companion to our guide on AI for cable sizing and voltage drop: that one covers conductor ampacity and voltage drop, while this one covers the load and demand calculation that sizes the service in the first place.

What an electrical load calculation is

A load calculation sizes branch circuits, feeders, and the service by summing the loads they carry and applying the code’s demand factors. The key idea is that the connected load, the sum of every appliance nameplate, is almost always larger than the demand load, the amount the installation is realistically expected to draw at once, because not everything runs simultaneously. Demand factors capture that, so the service you size is smaller than the raw total of the nameplates.

In the United States this is governed by the National Electrical Code, NFPA 70, Article 220. For dwelling units the code gives two paths: a standard method and an optional method. The standard method builds the load up from the general-lighting demand table and separate appliance, range, dryer, and HVAC calculations; the optional method is a simpler alternative with its own demand structure. Which one you use, and the exact table values, depend on the edition your jurisdiction has adopted, so always work from the code in front of you rather than from memory.

The optional method, concretely

The optional dwelling method, NEC 220.82, is a good example of how specific these values are, and why they must come from the code rather than a model. It applies to a dwelling served by a single 120/240-volt or 120/208-volt 3-wire service rated 100 amperes or greater. The general load is built from 3 VA per square foot of floor area for general lighting, 1,500 VA for each of the two required small-appliance branch circuits, 1,500 VA for the laundry branch circuit, and the nameplate ratings of the fastened-in-place appliances. The general load then gets a demand factor: the first 10,000 VA is taken at 100 percent and the remainder at 40 percent, with the heating and air-conditioning load added per its own rule (per NEC 220.82).

Those are exact, edition-specific numbers. A language model may state them with total confidence and still be wrong, or be quoting a different code edition, which is precisely why the values have to be read from the adopted NEC rather than trusted from an AI answer.

The workflow, step by step

Here is a defensible process, with AI’s genuine role called out at each step.

• 1. Build the load inventory. List general lighting, the small-appliance and laundry circuits, fastened appliances, HVAC, and motors, and tag each as continuous or noncontinuous. AI can organize this list from a plan or schedule into a clean table and flag likely continuous loads; it does not know the actual installed nameplates, so you confirm them.
• 2. Choose the method. Decide between the standard and optional method for a dwelling, or the appropriate commercial method. AI can explain the eligibility rules, for instance that the optional method needs a single 120/240 or 120/208 3-wire service of at least 100 amperes; it must not pick the legally correct method for your jurisdiction, and applying the wrong method to the wrong occupancy is a common failure.
• 3. Apply the demand factors. Bring in the general-lighting demand, the small-appliance and appliance demands, and the optional-method split. AI can set up the spreadsheet formulas once you supply the code values; it must not supply the demand-factor percentages or VA values from memory, because that is the highest hallucination risk in the whole process.
• 4. Handle continuous loads and motors. Continuous loads, those expected to run for three hours or more, are taken at 125 percent, and the largest motor is also increased. AI can apply a 1.25 multiplier you specify; it cannot reliably recall the exact governing sections across code editions, so verify them.
• 5. Sum to the service size. Total the adjusted VA, divide by the service voltage to get amperes, and round up to the next standard size. AI can do the summation and division in a spreadsheet you can audit, and draft the load schedule; it must not own the sizing decision, and its inline mental arithmetic is unreliable, so keep the math in a tool.
• 6. Check and document. Record the assumptions and the schedule for the reviewer and the inspector. AI is genuinely useful here, drafting a clean write-up from your inputs, which a qualified person then verifies.

Where AI must be checked

The failures are specific, and each one can undersize a service.

• It hallucinates code values. Demand factors, VA-per-square-foot figures, and continuous-load rules are exactly the kind of specific numbers a model will state confidently and get wrong, or quote from the wrong edition. Every value comes from the adopted NEC edition and jurisdiction.
• It applies the wrong method. Using the optional method where it is not permitted, or a dwelling method for a commercial occupancy, produces a plausible but invalid result.
• It misses the multipliers. The continuous-load 125 percent and the largest-motor rule are easy for a model to drop, and hard to spot in a clean-looking answer.
• The edition trap. Some Article 220 sections were renumbered between the 2020 and 2023 code, so an answer trained on one edition can cite a section that no longer says what it claims. Confirm section numbers against your adopted edition.
• It does not own the stamp. A load calculation is reviewed by a licensed engineer or qualified person and signed off by the authority having jurisdiction. Practitioners on electrical forums report that AI tools are frequently wrong on code-table values, and that verifying them erased any time saved, which is the right instinct: the model assists, it does not certify.

A note on non-US codes

Outside the United States the principle is the same but the mechanism differs. Under IEC 60364 and the UK’s BS 7671, you size the supply from the maximum demand after applying diversity, rather than from NEC-style demand factors. Maximum demand is the total connected load reduced by a diversity allowance, so it is always lower than the connected load, and it must be assessed before the supply and protective devices are sized. The diversity allowances come from published tables, and as with the NEC, those values belong to the standard, not to an AI answer.

The tools, and how much AI is really in them

The rigorous work belongs to purpose-built software. Pricing is mostly quote-based, so check each official site.

• Power-system analysis engines such as ETAP, SKM PowerTools, and EasyPower model the network and perform demand-load, load-flow, short-circuit, and coordination studies with the code libraries built in. They are analysis platforms, not language models.
• Panel-schedule and NEC-calculator software encode the code’s demand factoring into rule-based calculators. They are reliable for the specific job when the rule set matches your adopted edition, so confirm the edition.
• AI-branded NEC assistants are useful as a reference and lookup layer, and many are rule-based underneath a chat front end. Treat them as an assistant, not the source of a final number.
• General language models such as ChatGPT, Claude, and Gemini are method assistants: strong for explaining Article 220, organizing the inventory, and building spreadsheet formulas, and unreliable for recalling exact code values or doing the arithmetic in their head.

Frequently asked questions

Can ChatGPT do electrical load calculations?

It can help with the method, but it should not produce the final numbers alone. Use it to explain NEC Article 220, organize the load inventory, and build a summation spreadsheet from values you provide. It hallucinates demand factors and VA figures, can apply the wrong method, and does not carry code responsibility, so you supply the code values from the adopted edition, keep the arithmetic in a tool, and have a licensed engineer and the authority having jurisdiction review the result. For the broader question of whether AI can do engineering calculations at all, see our dedicated guide.

What is NEC Article 220?

Article 220 of the National Electrical Code contains the rules for calculating branch-circuit, feeder, and service loads. It sets out how to total the connected loads, apply demand factors so the calculated demand reflects realistic simultaneous use, and size the service accordingly. It offers a standard method and, for dwellings, an optional method. The exact table values depend on the code edition your jurisdiction has adopted, and some sections were renumbered between the 2020 and 2023 editions.

What is the difference between the standard and optional method?

Both are ways to size a dwelling service under Article 220. The standard method builds the load up from the general-lighting demand table plus separate appliance, range, dryer, and HVAC calculations. The optional method, NEC 220.82, is a simpler alternative available for a dwelling on a single 120/240 or 120/208 3-wire service of at least 100 amperes; it takes the general load at 100 percent for the first 10,000 VA and 40 percent for the remainder, with heating and cooling added per its own rule. Use the method your jurisdiction permits and read the values from the adopted edition.

What is a continuous load, and why 125 percent?

A continuous load is one expected to operate for three hours or more, such as store lighting. The code requires the conductor and overcurrent device to be sized for 125 percent of a continuous load, which is the same as sizing to 80 percent of the device rating, to account for sustained heating. It is an easy step for an AI answer to omit, so confirm it and the exact governing sections against your adopted code edition when it applies to your calculation.

What is the difference between connected load and demand load?

The connected load is the sum of every appliance and circuit nameplate, the theoretical maximum if everything ran at once. The demand load is the amount the installation is realistically expected to draw at the same time, found by applying the code’s demand factors to the connected load. Because not everything runs simultaneously, the demand load is lower, and it is the demand load, not the connected load, that sizes the service.

Sources

• Dwelling service calculation, optional method NEC 220.82 (ExpertCE)
• Dwelling load calculations per the NEC (EC and M)
• Feeder and branch-circuit load calculation basics (EE Power)
• Maximum demand and diversity under BS 7671 (Pro Certs)

Written by the CognitiveFuture editorial team. We build our guidance from published standards and official product documentation, and we label vendor statements as such. We do not independently benchmark any tool, and we do not treat AI-generated code values or results as reliable. Every demand factor and code value must be read from the adopted code edition, and an electrical load calculation is reviewed by a licensed professional and the authority having jurisdiction before it is used.