Norway's National Library Uses 2PB Huawei Flash Storage for LLM Training
Original: Norway's 2 petabytes of Huawei flash storage and LLM training
Why This Matters
Highlights sovereign AI development challenges and infrastructure needs for national language models
Norway's National Library is developing a Norwegian language LLM using 2 PB of Huawei OceanStor Dorado flash storage. The library leverages its 20 PB digital collection and exclusive newspaper content agreements for sovereign AI development.
Marius Husnes, Head of IT Platform at Norway's National Library, presented at Huawei's ID Forum 2026 about developing a sovereign Norwegian language LLM. The Ministry of Culture tasked the library with this project, citing disadvantages of relying on English-trained commercial LLMs that lack knowledge of local history and culture. The library holds Norway's largest digital collection, including books, newspapers, and web content totaling 20 PB stored in 3-2-1 format (60 PB overall). The AI pipeline uses Nvidia DGX H200 systems, 384-core CPU clusters, and 2 PB of Huawei OceanStor Dorado flash arrays for low-latency data processing. After pipeline processing, data goes to Norway's Sigma2 Olivia supercomputer for training. Husnes noted the main challenge was data quality and pipeline throughput, not compute power.