NVIDIA has introduced DAQIRI (Data Acquisition for Integrated Real-time Instruments), a software-defined architecture engineered to resolve the data processing bottlenecks inherent in modern scientific and industrial sensors. Traditional hardware-centric acquisition systems struggle with the immense output rates of high-bandwidth detectors, forcing a slow collect-store-analyze workflow. DAQIRI shifts this paradigm by enabling real-time, in-stream data processing directly at the source, seamlessly integrating high-speed detectors into the NVIDIA software ecosystem without requiring instrument hardware modifications. Built as a high-performance networking library within the NVIDIA Holoscan Platform, DAQIRI bypasses the Linux kernel to achieve line-rate processing of hundreds of Gbps Ethernet, UDP, and RoCE v2 traffic. Leveraging the Data Plane Development Kit, the system provides zero-copy access that routes data directly from network interface cards to GPU Direct Memory Access buffers. This architecture drastically reduces latency and CPU overhead. Developers interact with the system through C++ and Python APIs alongside configuration files, which automate the batching of incoming network packets into GPU tensors and handle on-the-fly data format conversions so applications can immediately execute filtering, inference, or compression routines. The architecture is currently being deployed to address extreme data challenges at major research facilities. At the European Organization for Nuclear Research, scientists are integrating DAQIRI into the High-Luminosity Large Hadron Collider upgrade for the ATLAS detector. The A-GHOST research initiative, a collaboration between CERN Openlab, the University of Chicago, and University College London, utilizes the framework to apply advanced artificial intelligence models directly to data streams typically discarded during initial event selection. By connecting custom field-programmable gate array hardware to GPU processing farms, researchers can conduct real-time analyses using convolutional auto-encoders, temporal neural networks, and transformer architectures on millions of collisions per second. By eliminating the traditional storage bottleneck, DAQIRI empowers instrument builders to monitor experiments continuously, adapt to changing conditions, and trigger actions instantaneously. Edge supercomputing systems can process raw streams and forward only high-value data to downstream high-performance computing facilities. This software-centric approach accelerates the pathway from raw data collection to scientific discovery, establishing a new standard for real-time AI integration across high-speed data acquisition workflows.