Where NCAR software can lead — and get paid for it

NCAR has no direct federal budget line; it is a Federally Funded R&D Center run by UCAR under a cooperative agreement with NSF. That structure made it acutely exposed when, in December 2025, OMB Director Russell Vought announced NSF would "break up" NCAR. A January 2026 Dear Colleague Letter proposed transferring the NCAR-Wyoming Supercomputing Center — home to the Derecho supercomputer — to another steward, divesting research aircraft, and closing or narrowing the Mesa Lab.

UCAR sued NSF, NOAA, OMB and Commerce in March 2026. On June 1, 2026, Judge R. Brooke Jackson granted a preliminary injunction blocking the supercomputing-facility transfer, finding the action "arbitrary and capricious" and citing evidence of political retaliation and of a "brain drain" of supercomputing experts already occurring. The injunction blocks only the first concrete step; the broader restructuring review is unresolved.

Underneath the headline fight is a budget reality: NSF core funding to NCAR was roughly $123M in FY2025 — about half of NCAR's ~$311M total, with the rest from NOAA, NASA, DOD, FAA, DOE and EPA. The FY2026 request proposed a ~40% cut to NCAR (~$50M) inside a much larger proposed NSF reduction; Congress rejected the deepest cuts, but NSF still trimmed hundreds of individual program budgets by up to 30%.

Between 2023 and 2026, data-driven weather models moved from demos to operations. The striking pattern for strategy: the flagship models are largely open. When the model is a commodity, competitive advantage shifts to the things around it that are hard to fake — and several of those are natural public-institution roles.

While the models opened up, private companies moved into the observation and delivery layer — often straight into gaps left by NOAA cuts. WindBorne fed NOAA balloon data at no cost during 2025 radiosonde suspensions; Tomorrow.io and Spire sell microwave-sounder and radio-occultation data into AWIPS2 under multi-million-dollar contracts; Brightband is under a NOAA CRADA to make the NOAA-NASA archive AI-ready (the "NNJA-AI" dataset); Climavision's private X-band radar feed reached operational AWIPS2 in January 2026. NCAR already has live technical relationships in this ecosystem (WindBorne; The Weather Company running NCAR's MPAS in its GPU "GRAF" system).

Each of these is backed by recent literature or community signal (see sources). They are the raw material for the project portfolio in the next section.

Each idea is scored 1–5 on five criteria. Because "which idea wins" depends on what you value, the ranking is live: drag the weight sliders and the table re-sorts. Defaults weight fundability highest, reflecting the moment. Click any column header to sort by it.

Criteria — Impact: community + operational value · Feasibility: can a RAL software team realistically do it (skills, data, compute) · Fundability: breadth of non-NSF sponsors & revenue potential · Strategic fit: leverages NCAR's unique public-good advantages · Ease: speed to a useful MVP (5 = easiest). Composite is a weighted average scaled to 100.

The point of diversification is that no single sponsor gates the whole portfolio. Each source below fits a different slice.

• The restructuring is live. Facts about the breakup, injunction, and budgets were true as of mid-2026 and can change materially. Treat the institutional picture as a moving target.
• Some signals are conference-reported, not primary. NOAA's "10-year MPAS unified system" direction rests on AMS-talk coverage more than a funded roadmap. The RAL reorg's causal link to the funding threat is inference, not stated.
• The scores are judgment, not measurement. The 1–5 ratings encode my read of the evidence; that's why the weights are yours to change. Feasibility especially depends on internal team composition I can't see.
• A few product details need internal confirmation — e.g., whether AQ-WATCH is genuinely an RAL product vs. an adjacent EU project, and current GDEX capacity figures. Verify before citing externally.
• No confirmed NCAR–NVIDIA/Microsoft AI partnership was found — an absence of evidence, not evidence of absence.