On-premises

On-premises means a company buys computer servers or AI chips outright and installs them inside its own building, rather than renting that computing power. Think of it like buying a DVD player and discs to watch movies at home, instead of streaming from Netflix. The upfront cost is high, but you manage everything yourself, and your data never leaves your hands.

Cloud computing, by contrast, means renting computers sitting in someone else's data center over the internet. You only pay for what you use, which is convenient, but some organizations are uncomfortable with their data being stored on another company's servers. That's why banks, government agencies, and companies handling sensitive data often prefer on-premises setups.

In the AI era, this distinction matters more than ever. Running large AI models requires massive numbers of expensive computer chips, and deciding whether to rent that capacity from the cloud or buy it outright for your own facility has become a major strategic choice for companies.

In the article, NVIDIA breaks down its data center revenue and explains that it wants to "distinguish between demand from AI-native cloud providers, sovereign AI, and on-premises enterprises." Here, on-premises refers to demand from companies that buy AI chips directly and install them in their own facilities, rather than cloud providers. It helps to remember that this is the opposite concept of renting from the cloud.