NASA will occupy Booth #4038 at Supercomputing 2026, running a three-day tech demo schedule from November 17-19. The lineup is narrow — seven distinct presentations by six speakers — but it maps the agency's current HPC priorities with unusual clarity. The heaviest emphasis falls on Fortran modernization. Tom Clune's triple-slotted presentation on type-safe templates in Fortran 2028 opens the conference and signals that NASA's enormous legacy Fortran codebase isn't going anywhere — it's being upgraded in place. Fortran 2028's template system is the language's most ambitious new feature in decades, and NASA clearly wants to demonstrate that its scientific codes can adopt modern generic programming without a full rewrite. Brad Richardson follows with a dependency-management talk — "Mixing and Matching and Multiple Versions" — addressing one of the oldest pain points in scientific computing: how to build reproducible software stacks when libraries ship conflicting versions. This is infrastructure plumbing, not glamorous, but it's the kind of work that determines whether large codebases remain maintainable at scale. Jeff Becker's presentation on porting the NAS Parallel Benchmarks to OpenMP Target Offload is a direct GPU-migration story. The NAS benchmarks are a canonical HPC performance suite, and porting them to offload directives is both a proof of concept and a public signal: NASA is moving workloads to GPU accelerators using portable, standards-based approaches rather than vendor-locked frameworks. Wednesday brings the most forward-looking entry. Sujay Kumar's presentation on high-resolution hydrology simulations paired with "agentic SI interfaces" for mapping water availability and extremes combines Earth science simulation with AI-driven query interfaces — a concrete example of how agentic systems are entering operational scientific workflows, not just research demos. Chris Tanner's modular supercomputing data center talk and Jeff Triplett's joint NASA-bp presentation on Open OnDemand and ColdFront round out the schedule with infrastructure and cross-sector HPC management. The bp collaboration is notable: oil-and-gas and space science share enough HPC architecture that tooling transfers directly. The schedule's structure — each speaker repeating their demo three times in 30-minute slots — is standard conference-booth logistics, designed to maximize attendee throughput. What's telling is what's absent: no AI-training showcase, no quantum computing demo, no flashy visualization. This is an agency investing in the unsexy infrastructure that keeps decade-old simulation codes running on next-generation hardware.
