How Much Electricity and Water Does TheraPro Use?

Ben Toronto

Co-Founder & CEO

image 11/18
2026

Recently we’ve been hearing questions from our therapists about AI data centers and the electricity and water they consume. Good questions that deserve to be asked. So we thought, why not estimate AI resource usage on a per-session basis for TheraPro?

Here’s where we landed: for a 53-minute TheraPro session, we estimate that AI usage for transcription and note drafting consumes about as much electricity as an electric kettle uses in seven seconds, and about two-thirds of a teaspoon of water. Let’s look at the details.

One session, three watt-hours

We started with the two jobs we’re asking our AI to do. First, it has to turn 53 minutes of audio conversation into text. Then it has to work through that conversation and draft a note for you to review.

For the first job, we used a published test of speech recognition software running on a cloud server. In that test, the chip used about 2 watt-hours of electricity to transcribe an hour of audio. For our 53-minute session, that comes to about 1.8 watt-hours. For the second job, we adapted a published estimate for a language model processing a long document and producing a much shorter response. Assuming the model can reuse the parts of the transcript it has already processed, we get another 1.1 to 1.3 watt-hours. Call it about 3 watt-hours of electricity for the AI processing. [Speech recognition benchmark] · [Language model estimate]

I don’t know that many people have an intuition for what a “watt-hour” means. I know I don’t. I do, however, charge my MacBook and make myself a cup of coffee in the morning. If my computer is drawing the full 90 watts from its adapter, it consumes 3 watt-hours in two minutes of charging. When I switch on my 1,500-watt kettle to boil water for a cup of coffee in the morning (French press, please), the kettle uses that much power in about seven seconds. I have to wait a fair bit longer than that for my coffee to be ready (and now I’m thinking about my resource usage there).

The 3 watt-hours is a benchmark estimate. We don’t have a meter reading for an individual TheraPro session, and the AI chips are one part of the overall service. The rest of the servers, the cooling systems, the network, and storage all use electricity too. Manufacturing the hardware and training the models also take resources. So while we can put a rough number on how much electricity usage is attributable to AI in a TheraPro session, the full footprint is still a bigger question.

One session, two-thirds of a teaspoon

The water question is a bit trickier. A data center’s cooling system can circulate the same water several times. From the perspective of someone who lives near a data center, it makes sense to ask how much freshwater the facility takes in and how much it returns. The difference is net water consumption. Much of this consumption comes from evaporation during cooling.

Google publishes a figure for this across its data centers, and we believe the data centers we use are comparable enough to make it a reasonable starting point for TheraPro. In its study, Google reported consuming an average of 1.15 liters of freshwater on site for each kilowatt-hour used by its computing equipment. Apply that rate to our 3 watt-hours of chip work and you get about 3.5 milliliters of water, or roughly two-thirds of a teaspoon. [Google’s study]

That teaspoon comparison covers the share of on-site water we’d attribute to the AI usage in our 53-minute session example. Cooling the rest of the equipment and generating the electricity can also consume water. And location matters. A community drawing on a scarce water supply has good reason to care about the combined demand of an entire data center. As a resident of Utah, a high desert where water is relatively scarce, I care about this, too.

Is it worth it?

One thing I like hearing in the broader conversation around AI is that people are asking what we get in return for the resources we use. We should ask that of any company building with AI, including TheraPro.

For a therapist, the alternative to using TheraPro is sitting down at a computer and writing the note yourself. With TheraPro, you still need to read the draft, make any edits needed, and sign off on the note. Our hope is that you can then close your laptop and go live your life. If the laptop goes to sleep while you go do something else, that offsets some of the electricity used on our end. We haven’t measured that saving, so we’ve left it out of the math.

Our therapists report saving an average of 5.3 hours per week from using TheraPro, but we all have to answer the trade-off question for ourselves. If writing your own notes works well for you and you’re done when your last client leaves, that’s great (and congratulations). If Friday’s notes regularly find their way into Saturday morning, you know exactly what those notes cost you. And if TheraPro helps you finish them on Friday, close the laptop, and have Saturday to yourself, with AI that uses less electricity and water than it takes to make your morning coffee, we hope that trade-off is worth it to you.

Get started with TheraPro AI by checking out our Plans and Pricing, finding out more about How TheraPro Works, or seeing our approach to Security and Privacy. You can Contact Our Team to talk with a real person if you have questions, or Sign Up to create a free account and start using TheraPro AI today!