At Cannes Lions 2026 (June 22–26 in Cannes), NVIDIA showcased a slate of production-ready AI tools that reshape advertising and marketing workflows, demonstrated alongside a roster of adtech partners. The demos ran on NVIDIA GPU and inference infrastructure built to meet the scale demands of modern digital advertising NVIDIA Blog. The event has since closed, but the partner announcements lay out where adtech infrastructure is heading for the rest of the year. This is a NVIDIA adtech-infrastructure story — Cannes Lions is the venue, not the subject — covering how Alembic, AWS, Criteo, Taboola, Higgsfield, and KERV.ai are shipping GPU-native marketing AI.
Causal marketing attribution runs on NVIDIA infrastructure for ad-tech teams
Proving which campaigns actually drive revenue remains a core challenge for marketing teams. Most existing tools only report post-hoc correlation rather than true causation between marketing spend and business outcomes.
Causal AI platform Alembic is addressing this gap using NVIDIA DGX Vera Rubin NVL72 systems. It is the first causal AI vendor to deploy NVIDIA DGX Vera Rubin SuperPODs for enterprise-scale causal modeling NVIDIA Blog. All inference runs on private supercomputing infrastructure housed in Equinix data centers where customer data already resides. The stack delivers a single, unified source of truth for marketing ROI, replacing fragmented reports from individual channel platforms.
Real-time bidding uses NVIDIA inference to cut latency for auction-scale workloads
Real-time digital ad auctions require low-latency inference to avoid missing bid windows. This barrier has kept most adtech firms reliant on rules-based decisioning systems that underperform against modern AI models.
AWS has built a production-ready reference stack for adtech customers combining cloud infrastructure, foundation models, and NVIDIA GPU-accelerated computing. NVIDIA Triton Inference Server delivers the low-latency inference needed to run AI-powered bid price optimization, audience activation, and deal scoring directly inside live auctions NVIDIA Blog.
Retail ad network Criteo serves product recommendations across one of digital advertising’s largest retail networks, processing billions of daily shopper interactions. The company achieved a roughly 2x speedup in model training using NVIDIA Blackwell GPUs and the open-source NVIDIA cuEmbed library. This efficiency cuts an estimated 17,000 annual GPU hours from Criteo’s model retraining pipeline, freeing resources to iterate on recommendation models faster and improve relevance for shoppers.
Taboola is leveraging NVIDIA GPU infrastructure to power its DeeperDive AI answer engine NVIDIA Blog.
Autonomous AI automates full marketing lifecycles for brands
Autonomous AI is moving from experimental pilot to production-ready for marketing workflows. Adoption requires built-in guardrails for safety, auditability, and role-based access control to meet compliance requirements.
AI video and image production platform Higgsfield AI is demonstrating its Higgsfield Supercomputer agent suite at Cannes Lions. The suite is built on NVIDIA Blackwell architecture and powered by the NVIDIA Agent Toolkit, including NemoClaw blueprints and the OpenShell secure runtime NVIDIA Blog.
The agents manage the full marketing lifecycle end-to-end: campaign ideation, planning, creative production, posting, and autonomous performance optimization. All functions are accessible from a single unified interface. The integration of NemoClaw and OpenShell provides an enterprise trust layer, letting marketing teams audit agent actions and enforce role-based permissions for sensitive campaign data.
Multimodal AI enables contextual ad matching at production scale
Contextual advertising matches ad creative to surrounding content by interpreting semantic meaning across video frames, images, and text. This task was previously too compute-intensive for widespread deployment.
AI-powered media measurement and ad targeting firm KERV.ai is demonstrating its Moment Match Engine at the event, built on NVIDIA’s multimodal AI stack NVIDIA Blog. The system identifies individual scenes, objects, and products in content to match ad creative based on its visual and textual elements.
The collective demos highlight a structural shift in adtech infrastructure. The industry is moving away from siloed, task-specific tools running on general-purpose CPUs toward unified, GPU-native stacks purpose-built for autonomous, full-funnel operations. Unlike earlier waves of AI adoption that focused on isolated optimizations like click-through rate prediction or basic audience segmentation, the tools on display are designed to handle end-to-end workflows with minimal human intervention. This is a direct response to pressure on marketing teams to deliver higher ROI with smaller headcounts.
For brands and agencies evaluating where to invest, the takeaway from Cannes Lions is concrete: the differentiator is no longer whether a vendor offers an AI feature, but whether the underlying stack can run causal attribution, real-time bidding, and full-lifecycle agentic campaigns on the same GPU-native foundation with auditable guardrails. Criteo’s roughly 2x training speedup and 17,000 GPU-hours saved on Blackwell is the kind of operational number teams can benchmark against when weighing a migration.
Bottom line: Adtech teams building or upgrading AI infrastructure should prioritize GPU-native stacks with low-latency inference and flexible colocation deployment options to support causal attribution, real-time bidding, and autonomous campaign management tools, as demonstrated by the NVIDIA partner ecosystem at Cannes Lions. For context, Criteo achieved a 2x model training speedup and eliminated an estimated 17,000 annual GPU hours using NVIDIA Blackwell GPUs and the cuEmbed library, while Alembic’s causal AI platform runs on NVIDIA DGX Vera Rubin NVL72 SuperPODs housed in Equinix data centers for enterprise-scale deployment.
