Stockertim Dictionary · Artificial intelligence · Technology
Compute
Also called compute resources
In short
Compute is the processing power, usually from specialised chips in data centres, used to train and run AI models, and a large driver of their energy use and cost.
- Beginner
- Concept
- Checked
In detail
Compute is the shorthand for the raw processing work that AI needs. Training a large model means running enormous numbers of calculations over huge data sets, often on thousands of specialised chips working together for weeks. Running the finished model, every time someone asks it a question, also needs compute, though far less per request.
The amount of compute depends on the size of the model, how much data it is trained on and how many times it is used. Larger models and longer outputs need more calculations. Because each request is cheap but requests are counted in millions, the energy for everyday use can add up to more than the one-off training effort over time.
Compute matters for three reasons. It is a large part of the cost of AI. It drives electricity use, and in some data centres water use for cooling, which has an environmental footprint that depends on how the power is generated. And access to large amounts of compute is concentrated among a small number of providers, which competition authorities have studied.
At work, choices affect compute: picking a smaller model when it does the job, avoiding unnecessary reruns, and not generating long outputs or many images when one will do.
Worked example
An illustrative energy calculation for a team's AI use, using round, made-up figures.
- Requests per day: 50 x 40 = 2,000
- Assumed energy per request of 0.003 kWh: 2,000 x 0.003 = 6 kWh a day
- Over 250 working days: 6 x 250 = 1,500 kWh a year
- Halving reruns and using a smaller model for simple tasks cuts requests by 30%: 1,500 x 0.7 = 1,050 kWh
On these assumed figures, simple habits save 1,500 − 1,050 = 450 kWh a year; real figures vary widely by model and data centre.
In practice
IT and sustainability teams include compute in the cost and footprint of AI projects, choose model sizes that fit the task, and ask providers for information on energy sources and efficiency.
Watch out
- Energy per request varies widely by model, task and data centre; treat any single figure as illustrative.
- Compute is not the same as data: a model can be data-hungry and compute-hungry for different reasons.
Learn it properly
- Environmental and social impacts15 min lesson Free course: AI ethics and responsible AI
- Lab 2: computing returns in code15 min lesson Free course: Financial Data Lab
- Indicators: what RSI and MACD compute, and nothing more15 min lesson Free course: Charts in context: technical analysis and its limits
- Hyperscalers and cloud computing15 min lesson Free course: AI stocks & technology investing: Intermediate
- What are AI companies?5 min lesson Free course: AI stocks & technology investing: Beginner
Official sources
- International AI Safety Report 2025 · UK Government
- AI foundation models: initial report · Competition and Markets Authority
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.