The full catalogue of Prowys research. Quarterly briefings, standalone papers, sector notes, and work co-authored with partner universities and research labs.
Quarterly market briefings, standalone papers, sector notes, and co-authored work with partner universities and research labs. Some papers are open; others are released on request under NDA.
A working framework for token cost, covering serving stack, batch dynamics, utilization, and the four line items most buyers undercount.
A side-by-side of three reference fabrics deployed in the past 18 months. Networking choice now drives effective tokens-per-watt more than GPU SKU.
How three national regulators are converging, and diverging, on model residency, audit, and cross-border inference. Practical implications for buyer timelines.
A field study across nine facilities. Power envelope, water budget, and the operational gap between rated and deliverable rack density.
Mid-cycle reading of the GPU market. What is moving, what is mispriced, what is now obsolete for new deployments.
Eval criteria, datasets, and reproducible scoring for enterprise multilingual LLM selection. Released as an open benchmark.
Why year-two cost-per-token is set by FinOps and scheduler choice, not the original GPU contract. Five reference clusters benchmarked.
Five use-cases ranked by realised ROI across six Tier-1 banks. What is shipping, what is shelved, and why.
Co-authored papers are released jointly under Prowys' and the partner institution's marks. Bespoke buyer-side research is commissioned on a per-engagement basis.
Commission research