Nvidia patents system to predict AI workload needs before they start
Nvidia describes a method where a system studies past AI tasks to predict how much computing power future requests will need. It attaches these predictions to new requests so the server can assign resources efficiently before the work begins.
In the filing’s own words
A system is provided for predicting and optimizing resource utilization in artificial intelligence (AI) agent workflows. The system trains forecasting models using offline profiling data to predict future behavior patterns and resource utilization. During runtime execution, the system generates predictive metadata for incoming requests, including estimates of expected workflow execution patterns and anticipated resource utilization. This predictive metadata is then attached to the requests and sent to an inference server, which makes resource allocation decisions based on the predictive information.
What exactly they claim (claim 1)
1. One or more processors comprising processing circuitry to perform operations comprising: training, by an artificial intelligence (AI) agent system, one or more forecasting models based on offline profiling data collected from one or more AI agent workflows, the one or more forecasting models predicting future workflow behavior patterns and resource utilization characteristics for the one or more AI agent workflows; generating, by the AI agent system during runtime execution of the one or more AI agent workflows, predictive metadata for one or more requests using the one or more forecasting models, the predictive metadata including estimates of future work comprising expected workflow execution patterns and anticipated resource utilization; associating, by the AI agent system, the predictive metadata with the one or more requests; and causing, by the AI agent system, an inference server to perform resource allocation decisions using the predictive metadata to execute the one or more requests.
Facts
- Company
- Nvidia (NVIDIA Corporation)
- Official title
- ONLINE INFERENCE OPTIMIZATION AND CONTROL OF AGENT WORKFLOWS
- Publication number
- US20260278406A1
- Application number
- 19358820
- Filed
- Oct 15, 2025
- Published
- Sep 17, 2026
- Inventors
- 2
- Claims
- 20
Classified under
G06N 5/022 Physics & computingG06N 5/043 Physics & computingMore from Nvidia in AI & software
- Nvidia patents AI coach that analyzes your gameplay to suggest strategies Oct 13, 2026
- Nvidia patents AI system that writes and voices character conversations Sep 10, 2026
- Nvidia patents system that uses AI to find bugs in game videos Sep 17, 2026
- Nvidia patents AI method to make digital characters move realistically Sep 8, 2026
- Nvidia patents AI system that builds image generators from text prompts Sep 15, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.