Training vs. inference
Training is the one-off, extremely compute-intensive learning of an AI model from data; inference is every subsequent use of the finished model, such as an answer in ChatGPT.
Explained simply
During training the model sees huge amounts of data and adjusts billions of internal parameters. That takes weeks to months on tens of thousands of chips and costs providers hundreds of millions.
During inference the model is finished and answers requests. Each one is cheap, but there are billions per day. In total, computing demand for inference now exceeds that for training.
Both phases need different infrastructure: training concentrated in a few giant data centres, inference distributed and close to the user.
From the familiar world
Medical school and practice: the education takes years and is expensive (training). Afterwards the doctor treats patients daily (inference). The education happens once, the treatments millions of times.
An example
A language model is trained over three months. Afterwards it answers billions of requests per week worldwide, each within seconds.
Why it matters
The distinction explains why demand for computing does not fade after the first investment wave: inference grows with every new application. That matters when assessing chip and data centre companies.
To pass on
„Training is the degree, inference the daily practice. The degree is expensive, but the practice runs millions of times.“
The content of this wiki is for general information and educational purposes only. It does not constitute investment advice, an investment recommendation or tax or legal advice. Digital assets are subject to strong price fluctuations and can lead to losses up to a total loss. Content as of 2026, no guarantee of completeness or timeliness.
21 Oaks Capital GmbH acts as a tied agent exclusively for the account and under the liability of AHP Capital Management GmbH, Weißfrauenstraße 12-16, 60311 Frankfurt am Main, Germany. This wiki is a general information service provided by 21 Oaks Capital GmbH. Its content does not constitute individual investment advice and is not a recommendation by AHP Capital Management GmbH. 21 Oaks Capital GmbH, its affiliated companies, the funds it advises or its employees may hold positions in assets explained in this wiki. The explanations are independent of this; the disclosure is made for reasons of transparency.