Swiss researchers pull NASA climate archives onto a supercomputer, in case the originals disappear

On 20 July 2026, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) in Lugano confirmed they had begun copying large volumes of publicly available NASA climate and environmental data onto the Alps supercomputer, citing both AI-training demand and concerns about continuity at the source. The move, reported by Physics Today (PHYS), is part academic infrastructure, part strategic reserve: a copy of the public record held by a foreign power now sits, in triplicate, on neutral territory.

The act is small in diplomatic terms and large in what it implies. NASA has, for decades, operated the de facto global archive of satellite-derived climate observations: sea-surface temperature, ice mass, atmospheric composition, land-use change. If access to that archive narrows, the models downstream of it narrow too. Building a foreign mirror is the kind of precaution institutions take when they no longer trust that the original will remain open on its current terms.

What's actually being copied

ETH Zurich's project, dubbed "Earth-2," runs on CSCS's flagship Alps system and is built around the Pangeo open-source stack, a Python-based toolkit for geoscience data analysis. According to the Physics Today report, the immediate targets are atmospheric reanalyses, satellite-derived sea-surface temperature records, and land-use change datasets that already sit in NASA's public archives. The point is not to supersede NASA but to train large AI models on data the Swiss team can be confident will still be there in six months.

The technical reasoning is straightforward. Modern climate-AI models need petabyte-scale training corpora; round-tripping queries against NASA servers on every training epoch is slow, expensive, and brittle. A local copy changes the economics. The political reasoning is less often stated aloud but is doing real work in the room: Swiss researchers are hedging against a future in which US federal data policy turns the spigot off, or slows to a trickle, on datasets that the rest of the world depends on.

The framing is preemptive rather than reactive. No NASA dataset has, to date, been withdrawn in any documented case covered by the sources Monexus reviewed on 20 July 2026. The Swiss move is insurance against a contingency, not a repair of a confirmed break.

The other reason it's happening now

The Swiss decision sits inside a wider pattern of academic institutions quietly backing up US federal datasets that they treat as the global common. The habit hardened after January 2025, when deletions at US government health and science agencies prompted a volunteer mirroring effort nicknamed "Endangered Data Week" that copied CDC, EPA, and NOAA datasets onto university servers. Climate and Earth-system data were swept up in that wave. ETH Zurich's project is the institutionalised, well-funded version of the same impulse: don't ask Washington, just have your own copy.

The pattern raises an awkward structural question. For three decades, the United States has underwritten, at public expense, the global atmospheric and oceanic observing system and the open data products derived from it. That arrangement was never strictly bilateral; European, Japanese, and Indian satellites contributed, and European centres like ECMWF did much of the heavy lifting on reanalysis. But the distribution of those products ran through US servers, on terms set by US policy. A mirror on Swiss soil does not change the underlying science. It does change where the choke points sit.

The stakes, in plain terms

If the Swiss archive holds, two things follow. First, European climate science and the AI industry training on top of it gain a degree of operational independence from US federal decisions. Second, the implicit subsidy the United States has extended to the rest of the world, by hosting open climate data on favourable terms, becomes visible as a subsidy, which means it can be priced, renegotiated, or reciprocated.

For a Swiss university this looks like prudent stewardship. For a US administration that wants to keep the open-data bargain intact, it is a quiet warning. The mirror will do its job whether the original remains open or not. That is precisely why it is being built.

The remaining uncertainty is what triggers the contingency the mirror is being built against. The sources Monexus reviewed on 20 July 2026 do not document a specific NASA data withdrawal; they document a research group's stated concern that one could occur. The project is therefore better read as a load-bearing acknowledgement that, in 2026, the durability of US open-data policy is itself a research variable.

Desk note: Monexus framed this as a data-infrastructure story with a geopolitical edge, not as a science-and-AI story with a politics footnote. The lead centres the action (the copy), the nut graf names the structural pattern (a foreign mirror of US public archives), and the closing uncertainty is left standing rather than resolved.