ETH Alps Supercomputer Opens Apertus 1.5 as On-Demand AI Service
ETH Zurich reported on September 15, 2026 that the Alps supercomputer at the Swiss National Supercomputing Centre is evolving beyond raw FLOPS into a service platform. Software layers known as vClusters let researchers and partners such as MeteoSwiss shape dedicated environments, while AI applications—including the Apertus 1.5 large language model offered since July 2026—draw compute automatically from Alps.
Filed under Research and dated September 15, 2026, this AI4Switzerland briefing situates the update as Swiss research-infrastructure news—not a commercial cloud competitor. Alps underpins much of the Swiss AI Initiative’s foundation-model work linking ETH, EPFL, and hundreds of researchers across climate, health, and language projects.
Why it matters: Swiss labs need reproducible access to large models without every group standing up private GPU farms. Service-style Alps access can widen participation—but only if fair-share policies, data classifications, and energy accounting stay transparent for cantonal and federal sponsors.
What it means in practice
Swiss principal investigators should inventory which projects need Apertus versus specialised models; confirm vCluster quotas and storage rules; assign an owner for research-data governance; run time-boxed comparisons against commercial APIs on cost and auditability; and prefer open evaluation logs. Cross-check applied robotics momentum via Swiss physical-AI robotics startups and disaster-forecast AI on Alps.
Caveats come first. A corporate communications feature is not a capacity guarantee; service SLAs for research differ from hyperscale clouds; and foundation-model roadmaps remain multi-year. AI4Switzerland therefore presents the Apertus-on-Alps story as directional research context until published user metrics appear.
What to watch next: expanded model catalogue beyond Apertus 1.5; published energy-per-token figures; and SME access pathways. Readers following Exclaim’s data-centre robots can continue on the AI4Switzerland homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Swiss AI compute is becoming more service-like and still research-first. Organizations that benefit most will design for shared infrastructure, publish evaluations, and refuse to confuse a press feature with unlimited GPU entitlement.