AI BriefingShopify Cuts Inference Costs by 90% with NVIDIA Open Models as European Retailers Adopt Physical AI
·2026.10.22 14:00
Key point
European retailers like Shopify, Zooplus, and Otto Group use NVIDIA-accelerated open models and digital twins to reduce costs, improve search accuracy, and simulate warehouse operations.
Details
- Shopify post-trained NVIDIA Nemotron 3.5 Lightning to surpass a leading frontier model in accuracy while cutting serving costs by 90% and delivering responses 7x faster.
- Zooplus achieved 86% search accuracy in a production-ready prototype using NVIDIA Nemotron Retriever and Blueprint for retail catalog enrichment.
- Otto Group advanced its autonomous warehouse initiative to its largest fulfillment center in Haldensleben, Germany, using NVIDIA Omniverse to design and test operations in simulation before physical implementation.
- CATCHES reported a 55% conversion increase and 56% gross sales increase for AMIRI using its RealFit virtual try-on deployment.
- Photoroom standardized approximately 40,000 images daily for Vestiaire Collective, resulting in a 5% higher rate of liquidity and 3% to 7% more unique views per product page.
- Omi’s ProductDrop workflow uses NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs to speed up inference by 10x and reduce pipeline latency by 4x for L’Oréal’s CreAItech program.
- L’Oréal integrated NVIDIA ALCHEMI to reduce ingredient experimentation time from months to hours, and tested OrbMol-v2, which is five times faster and delivers 21% lower error than comparable models.
- Nebius is showcasing three NVIDIA Blueprints for retail at GTC Berlin, including Agent Harness for Employee Productivity, Inventory Simulation and Optimization, and Agentic Finetuning Flywheel, all available in early access.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
RFM requires GPUs simultaneously across three stages: training, simulation, and edge inference