NVIDIA has shown that a new workflow, XANI, can cut the time needed to analyze nanoscale imaging data from nine months to less than four hours. The company says the system relies on its Grace Blackwell Superchips to handle massive datasets that once required months of processing.
The scale of the speedup
Nanoscale imaging produces enormous amounts of raw data. Researchers often spend months sifting through it to extract meaningful structures. NVIDIA's XANI workflow compresses that timeline dramatically. The nine-month figure represents what could be done with conventional computing setups; the new workflow brings that down to under a single workday.
The company hasn't released full benchmarks or named early adopters, but the claim points to a major leap in how electron microscopy and similar techniques might be used in materials science, biology, and semiconductor inspection.
What makes XANI different
XANI — short for something like 'X-ray and nanoscale imaging' though NVIDIA hasn't formally spelled out the acronym — is built on a pipeline that combines AI models with the parallel processing power of the Grace Blackwell architecture. The Grace CPU and Blackwell GPU are designed to handle both memory-intensive and compute-heavy tasks, which is exactly what nanoscale image reconstruction requires.
The workflow uses neural networks to denoise, segment, and reconstruct three-dimensional structures from two-dimensional slices. Normally, each step has to be run sequentially with manual tuning. XANI automates much of that process, letting the hardware chew through the data in parallel.
Nanoscale imaging is used to study everything from battery materials to protein complexes. A nine-month analysis delay often means that findings lag behind experiments, slowing down R&D cycles. Cutting that to hours could let researchers iterate faster, spot defects in real time, or run larger studies that were previously impractical.
NVIDIA hasn't announced when XANI will be broadly available or if it will be integrated into existing imaging tools. The company is likely to showcase the workflow at upcoming conferences, but for now the results are based on internal demonstrations.
The next question is how quickly labs can adopt the hardware. Grace Blackwell Superchips aren't cheap, and the software stack may need tweaks for specific microscope setups. If the speed holds up in real-world tests, XANI could become a standard part of nanoscale data pipelines.
