Argonne's ALCF adds Tara supercomputer to expand AI inference service

According to one outlet's report, Argonne National Laboratory's Argonne Leadership Computing Facility is widening its inference service around a new large-scale system named Tara. The move places a broader set of AI models and agentic workflows inside scientific research.
What the expansion adds
The Argonne Leadership Computing Facility is extending its inference service with Tara, a system built on supercomputing architecture. The expansion brings a wider selection of AI and super intelligence models into scientific work, along with agentic workflows, meaning AI that carries out multi-step tasks rather than answering a single question.
Tara was developed alongside Hewlett Packard Enterprise and rests on that company's exascale-class HPE Cray Supercomputing EX4000 architecture. NVIDIA, a semiconductor company based in Santa Clara, is associated with the system, and the headline cites 2,688 NVIDIA chips.
What changes for researchers
Beyond simple prompts, scientists can now hand autonomous AI agents tasks such as planning experiments, producing code and operating laboratory instruments at scale.
Venkat Vishwanath, SI team lead at the ALCF, said the facility has long supplied the computing resources researchers require to pursue science at scale.
Timing for American science
The report describes the upgrade as arriving at a pivotal moment for American science.
Open questions
- Which specific scientific projects will be the first to use the agentic workflows?
- Have Argonne or NVIDIA directly confirmed the 2,688-chip figure?