Nvidia's graphics processing units provide the key computing power for the majority of servers used to train today's most advanced artificial intelligence models. The company has already secured a leading position in this training market.
Once an AI model is trained, it must be deployed to perform tasks such as answering questions, writing code, generating images or searching for information. This phase, known as inference, requires computing resources each time the model is used. Unlike training, which occurs infrequently, inference workloads can happen millions or billions of times.
The analogy often used is that training is like attending school, while inference is akin to applying what was learned in a job. After the initial training investment, the need for computing power does not end; it may actually grow as AI models are put to work.
If AI systems become a common part of everyday life and business, the volume of inference workloads could become enormous. The emergence of AI agents that perform multi‑step tasks—such as researching a company, analysing financial data and producing a report—could further increase demand for inference computing.
Nvidia's current advantage in training hardware may therefore position it to capture a significant share of the expanding inference market, though the exact scale of that opportunity remains uncertain.
