NVIDIA Ecosystem Brings Causal AI, Real-Time Bidding, and Agentic Workflows to Cannes Lions

At Cannes Lions (June 22-26), NVIDIA partners are demonstrating production-ready AI infrastructure purpose-built for the shift from digital-era speed to AI-era autonomous operations in advertising and marketing.

The core message from NVIDIA: enterprises no longer debate whether to adopt AI in marketing — the question is whether their stack can handle the data volume, latency, and control requirements at scale.

Causal AI at Enterprise Scale

Alembic’s causal AI platform is using NVIDIA DGX Vera Rubin NVL72 systems and DGX Vera Rubin SuperPODs to model true causation across every channel, market, and audience simultaneously.

This moves beyond correlation-based reporting to quantify actual drivers of growth and identify wasted spend. Alembic is the first Causal AI company deploying these SuperPODs for enterprise-scale work. Inference runs on private supercomputing infrastructure inside Equinix data centers (with World Wide Technology extending coverage to secure/regulatory environments), keeping sensitive marketing data local.

Real-Time Auction Bidding and Recommendations

AWS is delivering a production-ready reference architecture for adtech that combines cloud infrastructure, foundation models, and NVIDIA GPU acceleration. The stack uses NVIDIA Triton Inference Server to run AI-powered bidding, audience activation, and deal scoring inside live auction windows — moving adtech from rules-based systems to model-driven decisions at auction speed.

Criteo, running one of the largest recommendation networks, collaborated with NVIDIA on Blackwell GPUs and the open cuEmbed library. Result: roughly 2x speedup in model training on billions of shopper timelines. That efficiency already frees ~17,000 GPU hours per year, with further scaling underway.

Taboola is extending similar NVIDIA GPU infrastructure to its conversational AI answer engine (DeeperDive) and downstream AI platforms/chatbots for ad revenue generation.

Agentic AI Across the Full Marketing Lifecycle

Higgsfield AI’s Supercomputer agents handle the complete marketing automation loop — campaign ideation, planning, creative production (image/video/audio), posting, performance analysis, and autonomous optimization — in one interface.

The agents orchestrate leading LLMs plus 35+ image, audio, and video models, including Higgsfield’s proprietary Soul and Soul 2.0 models built on NVIDIA Blackwell architecture. NVIDIA Agent Toolkit (including NemoClaw blueprints and OpenShell secure runtime) supplies the enterprise controls: safety guardrails, auditability, and role-based permissions. Marketing campaigns for nearly 400 Fortune 500 companies are already created on the platform.

Content and Contextual Intelligence

KERV.ai’s Moment Match Engine analyzes every video frame and media asset to understand scenes, objects, and products, then recommends ad creative placements for better engagement. After optimizing its pipeline with the NVIDIA Nemotron 3 Nano Omni open model, KERV reported over 10x improvements in speed and efficiency. On the open MediaPerf benchmark for AI video understanding, Nemotron 3 Nano Omni posted the highest throughput and lowest inference cost (open or closed source). PYLER is deploying it on NVIDIA DGX B200 systems.

Actions to Take

• Adtech / DSPs / SSPs: Evaluate the AWS + NVIDIA Triton reference implementation for moving bid optimization and audience activation into live auctions. The latency fit is now production-viable.

• Large retailers and performance marketers: Test Criteo-style retraining pipelines on Blackwell + cuEmbed. The documented 2x training speedup and 17k GPU-hour annual savings are immediate infrastructure ROI signals.

• Enterprise marketing leaders accountable for ROI: Pilot Alembic’s causal modeling on DGX Vera Rubin infrastructure (via Equinix or World Wide Technology) to replace last-touch or correlation reporting with quantified growth drivers.

• Marketing automation and creative platforms: Integrate the NVIDIA Agent Toolkit (NemoClaw + OpenShell) for guardrailed, auditable agent workflows. Higgsfield’s full-lifecycle example shows what “autonomous operations” looks like when controls are built in from day one.

• Video/content intelligence teams: Benchmark Nemotron 3 Nano Omni on MediaPerf-style workloads. The 10x efficiency gain and top throughput/cost numbers make it a strong candidate for production video understanding pipelines.

The through-line across all the Cannes Lions showcases is the same: NVIDIA’s latest inference servers, training libraries, agent tooling, and open models are removing the infrastructure bottlenecks that previously limited AI from moving from pilots to autonomous, enterprise-scale marketing operations.