Release 2026.11QUIC HTTP/3 Multipart Acceleration & PostgreSQL 17 Logical Slot Failover are now GA across ARN-1 & HEL-2.
Release Notes →
PostgreSQL 14–17 PITR Kubernetes & Velero Veeam & Restic WORM AI & Iceberg Lakes Case Studies
Solutions · AI Infrastructure & Analytics

Zero-Egress AI Checkpoints & Apache Iceberg Data Lakes

Training GPU clusters in one facility while hosting object lakes in another usually incurs crippling egress bills. Stream multi-terabyte model weights, Parquet tables, and genomics cohorts across Europe at €0.00/GB egress.

PyTorch & Safetensors Checkpoints

Burst-write 500 GB distributed FSDP / DeepSpeed checkpoints every 1,000 steps onto Gen5 NVMe journals at multi-gigabyte/s aggregate throughput.

Apache Iceberg, DuckDB & Polars

Execute serverless SQL pushdown queries directly against Parquet and Iceberg tables over HTTP/2 & HTTP/3 byte-range requests with conditional ETag catalog commits.

Genomics, Bioimaging & Earth Observation

Host Cloud-Optimized GeoTIFFs (COG), Zarr arrays, CRAM/FASTQ genomes, and hospital DICOM archives inside sovereign EU hydro-powered vaults.

DuckDB / Polars Direct S3 Analytics Example

Query partitioned Parquet datasets directly from Stockholm ARN-1 with zero egress charges:

INSTALL httpfs;
LOAD httpfs;

CREATE SECRET nordic_arn1 (
    TYPE S3,
    KEY_ID 'NDS_AKIA_ANALYTICS_RO',
    SECRET 'nds_secret_analytics_token',
    REGION 'eu-north-1',
    ENDPOINT 'se.nordicdatasync.com',
    URL_STYLE 'path'
);

SELECT event_date, country_code, count(*) AS events, sum(amount_eur) AS volume
FROM read_parquet('s3://lakehouse-prod/events/year=2026/*/*.parquet')
GROUP BY 1, 2
ORDER BY 1 DESC;