You might be sitting on a tiny or considerable fortune in wasted computing power right now.
Nvidia just put a number on that waste at IFA 2026, but that’s not the big story. The company has announced free software designed to put that idle computing power to good use.
How does PAIR actually work?
The software is called PAIR. It’s an acronym for Personal AI Router. The open-source software pools the unused computing power from all the PCs in your house and divides a complex AI task among their respective GPUs (which are sitting idle).
So, instead of your main computer tackling the workload, the PCs connected to your local network use their own graphics chips to complete delegated tasks and send the results back to the main computer. This is how the company promises to turn downtime into usable GPU compute.
All of this happens over a secure local network, without requiring an additional purchase or new hardware. Once you install the software, it automatically finds idle PCs on your home network and pairs them for seamless, encrypted communication.
From there, PAIR watches which PCs are actually free and routes AI tasks to those with computational room, while skipping anything you’re actively using. PAIR is a smart traffic manager, if you will (hence the name “AI Router”), not a tool that combines those GPUs into a single powerful processor.
How much compute is actually sitting idle in a typical home?
According to Nvidia, most American households already have two or more PCs. However, those machines are used for just 17% of an average day; around four hours.
Multiply those idle hours across all the computers in a typical home, and Nvidia estimates become genuinely concerning. Families are wasting about 165 teraflops of processing power, roughly equivalent to leaving a high-end gaming PC sitting idle all day.
Put a dollar figure on it, and the pitch behind PAIR gets even more compelling. Nvidia claims that running those idle machines, even at 60% utilization, could process around 120 million tokens a day with a mid-sized open model, or several million tokens an hour with a larger model. Outsourcing this type of AI workload would otherwise cost roughly $1,200 a month through cloud-based AI services, plus up to $120 in added electricity costs.
Starting today, the beta is live for Windows, macOS, and Linux, covering NVIDIA RTX 20-series GPUs and newer models, including RTX PRO workstation cards, DGX Spark, and Apple M4 or newer Macs.
