Phrasly estimate
≈0.19Wh
GPU electricity per 300-word response
Covers GPU processing only, using an assumed average power draw rather than meter readings. The full calculation is below.
Phrasly hosting information [1]Our company
Sustainability at Phrasly
Better writing starts with thoughtful engineering. Phrasly combines optimized AI models with renewable-powered hosting and air-cooled servers.
Sustainability
Our approach
How we run Phrasly
From the models we run to the facilities that host them, here’s what sits behind your writing.
GPU computing
Advanced computing, focused on writing.
Professional GPUs power Phrasly’s writing tools. Phrasly tunes the models and the infrastructure together to make efficient use of that computing power.
By handling many calculations at once, GPUs give Phrasly’s writing models the performance they need, and that tuning keeps operating costs down.
Electricity
Powered by renewable electricity
Phrasly’s AI tools run in data centers powered by renewable electricity.
That supply powers the whole hosting environment, from the GPUs processing your writing to storage, networking and cooling.
Cooling
Server cooling without water
Phrasly’s servers are air-cooled, so no water is used directly for server cooling. That setup runs in facilities powered by renewable electricity.
This describes cooling at the server. The facilities hosting those servers still reject that heat, and facility cooling, electricity generation and equipment manufacturing can all involve water.
Air-cooled infrastructure
Energy use
The energy behind a Phrasly response
Phrasly’s per-response energy estimate, alongside published research on longer AI reasoning queries.
Long reasoning queries
3.91Wh
Modeled median per long reasoning query
Based on 500 input tokens and a median 5,000 output tokens. Includes server electricity and data center overhead.
Oviedo et al. · Joule, 2026 [4]Writing and long reasoning are different tasks. The figures cover different response lengths and equipment.
How Phrasly calculates the estimate300 words ≈ 0.19 Wh · GPU processing
From words to electricity
A 300-word response is roughly 400 output tokens. At 0.48 Wh per 1,000 output tokens, that works out to about 0.19 Wh.
OpenAI’s token guide [3]The basis for the estimate
Phrasly calculates this from its own writing workload and an assumed average GPU power draw, not from metered per-query readings.
Phrasly hosting information [1]What is included
The estimate covers the GPUs processing your writing, excluding supporting servers, networking and facility operations.
Read the measurement reference [5]Efficiency & value
More room to write
Phrasly optimizes its writing models and infrastructure together to use computing resources efficiently and keep operating costs down.
Those savings go back into affordable subscriptions with unlimited humanization, giving you more room to revise and keep writing.
After the trial. Fair use applies. Plan terms
Questions & answers
Sustainability, explained
Yes. Renewable electricity powers the data centers hosting Phrasly’s AI tools, including their processing, storage and cooling. This comes from energy information published by the provider that operates those facilities.
Our energy information [1]Have another question?
Contact our teamThe values behind Phrasly
Sources & further reading
Phrasly: Hosting and infrastructure
Company information · September 2026
Phrasly’s AI tools run on renewable-powered hosting and air-cooled servers, with models optimized alongside the infrastructure that runs them. The renewable electricity information comes from hosting provider materials reviewed by Phrasly. These electricity and direct cooling-water statements describe hosting operations: facility cooling, electricity generation and equipment manufacturing sit outside that scope.
NVIDIA: GPU processing
Technical reference
NVIDIA’s introduction to parallel GPU processing and the calculations behind deep learning.
OpenAI: What are tokens?
Token and word estimates
The approximate English conversion used in our calculation: 100 tokens is about 75 words.
Oviedo et al.: Energy Use of AI Inference
Joule, 2026 · Manuscript revised June 9, 2026
Research modeling AI inference: 3.91 Wh for long reasoning queries with 500 input and a median 5,000 output tokens, and 0.31 Wh for short queries. Includes server electricity and facility overhead.
Google: Measuring AI inference
Published August 21, 2025
Google’s approach to measuring AI inference energy, including processors, memory, idle capacity and facility overhead.
Phrasly: Plans and pricing
Current product offer
Current features and trial limits: unlimited humanization, up to 5,000 words per request. Other tools have separate allowances.
Phrasly: Terms of Service
Subscription and usage terms
Section 6 covers unlimited humanization after the trial, fair use, the Acceptable Use Policy and trial access.
Images on this page are AI-generated illustrations. They are not photographs, and they do not show Phrasly’s facilities, hardware or energy supply.
Reflects Phrasly’s operating setup as of September 2026.