NVIDIA DAQIRI: Real-Time AI Inference on High-Speed Data Acquisition

NVIDIA has unveiled DAQIRI, a framework for embedding AI inference directly into high-speed data acquisition pipelines. The goal: AI should work where data is born, rather than wait for it to be written to storage and handed off for processing.

Why real-time AI in the stream matters

When AlphaFold2 revolutionized drug discovery in 2020, its success depended entirely on ~170,000 protein structures that scientists had collected since 1971 and that are stored in the Protein Data Bank. It is a telling example: good data is the foundation of any AI model. But the classic "collect, store, train" scheme has a fundamental flaw: time passes between the moment data appears and the moment AI reacts to it. In scientific experiments, industrial control, or medical diagnostics, that delay can cost you the result.

What DAQIRI is

DAQIRI (Data AcQuisition with Intelligent Real-time Inference) is a software layer from NVIDIA that connects GPU-accelerated inference directly to the high-speed stream coming from instruments and sensors. The framework tackles several tasks at once:

• Lossless ingestion and buffering of streaming data
• Running an AI model on the GPU in real time
• On-the-fly filtering and tagging of events, before anything is written to disk
• Integration with scientific instruments through standard interfaces
• Support for workflows where the volume of raw data exceeds storage throughput

The last point is especially important: in physics experiments and genomic research, detectors generate terabytes per second. Recording everything is impossible, so you have to choose what to keep. DAQIRI makes that choice in real time, using AI as a filter.

Where it applies

The framework targets primarily scientific and industrial scenarios with high-speed data sources: particle accelerators and physics detectors, genomic sequencing, industrial quality control on production lines, and medical imaging in streaming mode.

"Measured data is the foundation of all AI models and workflows that

process data at the moment of creation, react to what matters in real time, and analyze data to derive deep insights," from the NVIDIA Developer blog.

In each of these cases, the value of an event declines over time. DAQIRI moves AI as close as possible to the signal source.

What it means

NVIDIA has consistently pushed the "AI everywhere" concept, not only in data centers but at the edge of infrastructure, inside measurement instruments themselves. DAQIRI is another step in that direction: inference moves to where data originates, not to where it is convenient to store it.