Open AI models transform retail commerce and product discovery

Retail leaders and technology companies across Europe are using open frontier models accelerated by NVIDIA to advance agentic commerce, digital product discovery and retail research and development.

Shopify reduces AI inference costs by 90%

Shopify post-trains open models to achieve frontier-level accuracy at a fraction of the cost. Its engineers use NVIDIA Megatron Bridge libraries for faster fine-tuning and NVIDIA NeMo RL for reinforcement learning.

Shopify post-trained NVIDIA Nemotron 3.5 Lightning for the GraphQL subagent integrated with Sidekick, Shopify’s AI assistant for sellers. The subagents capture store data for several Sidekick features, including Sidekick Pulse, which inspects a merchant’s store and recommends ways to improve it.

In Shopify’s internal benchmarks, the trained model outperformed leading frontier models in accuracy, reduced service costs by 90% and delivered responses seven times faster. Beyond Sidekick, Shopify used NVIDIA CCluster with vLLM inference optimization software to achieve 4.5x faster performance and up to 12x lower inference costs across five additional use cases:

• Catalog enrichment
• Merchant assistance
• E-commerce shopping simulation
• Product classification
• Merchant and buyer recommendations

Zooplus and Infosys improve AI-powered product search

Zooplus, a leading European online pet retailer, and Infosys are rebuilding the product discovery stack for the agentic commerce era. A production-ready prototype deployed in Zooplus’ Amazon Web Services environment achieved 86% search accuracy.

The system uses the NVIDIA Nemotron Retriever library for semantic search and the NVIDIA Blueprint for Powering Your Retail Catalog to generate searchable product attributes.

Photoroom standardizes images for Vestiaire Collective

Photoroom’s image editing API standardizes approximately 40,000 featured images per day for Vestiaire Collective across 70 countries at the time of listing. According to Photoroom, the listing process produced 5% higher liquidity and 3% to 7% more unique views per product page.

The pipeline runs on NVIDIA GPUs with TensorRT and Triton Inference Server. NVIDIA Dynamo Snapshot helps users respond quickly to demand spikes.

Omi uses digital twins to scale accurate product imagery

Omi ProductDrop anchors generated images to an approved digital twin. This allows brands to scale creative output across their catalogs without introducing product errors, according to Omi.

Omi created a digital twin with a branded environment through the CreAItech program. The company runs TRELLIS 3D asset generation on NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, achieving 10x faster inference and 4x lower pipeline latency.

L’Oréal accelerates ingredient research with AI

L’Oréal integrates NVIDIA ALCHEMI into its research workflow, reducing computational screening from months to hours and accelerating the discovery of optimal ingredient combinations.

L’Oréal is also testing this approach with CurieOS, a science and engineering operating system from Orbital Industries that uses the open-source OrbMol-v2 molecular simulation model. Integrated with the NVIDIA ALCHEMI Toolkit and powered by NVIDIA accelerated computing, OrbMol-v2 is five times faster than comparable models and delivers 21% lower error than the next-best model of its kind.

This AI-based workflow helps researchers identify promising ingredient combinations for optimizing UV filter solubilization and performance before laboratory testing begins.

Mirakl uses open models to support technology sovereignty

Mirakl, an intelligent commerce software company, runs open models on NVIDIA GPUs on its own infrastructure and through European cloud providers, including Scaleway. This approach helps the company establish technology sovereignty, avoid vendor lock-in and deliver the speed and scale expected by its customers.

Mirakl’s Catalog Transformer combines a fine-tuned open model with proprietary services to onboard more than 70 million products and transform messy vendor data into accurate, well-organized catalogs.

Physical intelligence enables retail simulation and automation

NVIDIA physical AI, digital twins and open models are helping European retailers simulate environments, automate warehouse operations and deliver virtual try-on experiences for shoppers.

Otto Group simulates autonomous warehouse operations

Otto Group is advancing autonomous warehouse operations under its Human Oversight Initiative at its largest fulfillment center in Haldensleben, Germany. Built by Reply and running on the NVIDIA Omniverse library and Google Cloud, the system helps logistics teams design, test and optimize warehouse operations in a fully simulated environment before physical implementation.

AllSides creates accurate 3D digital twins

AllSides has launched a 3D digital twin factory with an automated pipeline that can convert physical objects into SimReady 3D digital twins in minutes.

The process captures sub-millimeter geometric accuracy, measured physical properties and fully relightable materials that maintain accuracy under different lighting conditions. These digital twins provide a foundation for simulating store layouts in 3D and tracking products during shopping and checkout, regardless of ambient light.

Built on NVIDIA CUDA, OptiX and TensorRT, the platform already serves brands including Adidas, Amazon, Nike, Zalando and Zara.

Catch expands physics-based virtual try-on

Catch achieved production results by implementing RealFit for Amiri. According to reported results, conversion rates increased by 55%, add-to-cart activity increased by 80% and total sales increased by 56%. The results were seasonally adjusted for similar styles.

Powered by open-weight models such as NVIDIA Nemotron 3, built on NVIDIA Omniverse and the Newton Physics Engine, Catch launched PredictiveFit. The technology extends physics-based virtual try-on to retailers. Catch also launched CASI, a personal fashion agent that provides customer shopping, style and size assistance through WhatsApp.

Explore NVIDIA retail blueprints at GTC Berlin

At GTC Berlin, Nebius introduced three NVIDIA blueprints for retail. The blueprints are available in early access and run on infrastructure accelerated by NVIDIA.

• Agent harness for employee productivity: Agent Assistant integrates scheduling, task management, inventory alerts and operational insights for store managers.
• Inventory simulation and optimization: This blueprint helps retailers reduce overstock and stockouts by modeling demand scenarios, testing replenishment strategies and optimizing inventory levels before implementation.
• Agentic fine-tuning flywheel: This self-improvement loop uses agent workflows to continuously generate training data, fine-tune open-source models and feed real-world performance into the next training cycle.

Request early access to the blueprints or learn more about NVIDIA technology for retail and consumer packaged goods.