NVIDIA Partners Deploy GPU Infrastructure for Ad Tech at Cannes

GPU infrastructure moves from concept to production in advertising

NVIDIA announced collaborations with six advertising technology companies deploying GPU-accelerated infrastructure at the Cannes Lions International Festival of Creativity this week. The partnerships with Alembic, Amazon Web Services, Criteo, HiggsField, KERV.ai, and Taboola represent production-grade implementations addressing four core problems: causal measurement, real-time bidding, full-lifecycle marketing automation, and video content intelligence.

These are not demonstrations. Each deployment involves specific hardware choices—GPU generations, inference servers, open libraries—with direct consequences for speed, cost, and accuracy at enterprise scale.

Alembic brings causal AI to SuperPOD infrastructure

Alembic will become the first causal AI company to run NVIDIA DGX Vera Rubin NVL72 SuperPODs for enterprise-scale marketing measurement. The company focuses on modeling what actually caused business outcomes, not just reporting correlations in campaign data.

The NVL72 configuration uses NVIDIA's Blackwell-generation GPU architecture. Running causal models across every channel, market, and audience simultaneously requires processing enormous, constantly changing datasets. Alembic's inference runs on private supercomputing infrastructure inside Equinix data centers, addressing both latency and data governance requirements.

The infrastructure responds to a measurement crisis. Research published in February 2026 showed that up to 75% of senior planning and analytics decision-makers at US brands and agencies report that advanced measurement approaches fail to deliver the rigor, timeliness, and trust needed to justify spending.

Real-time auction optimization from AWS and Criteo

Amazon Web Services is presenting a production-ready reference implementation for AI-powered bidding inside programmatic auctions. The infrastructure uses NVIDIA Triton Inference Server to deliver deep learning inference fast enough to fit within millisecond auction windows. The system targets demand-side platforms, supply-side platforms, and independent software vendors moving from rules-based decisioning to AI-powered models.

Criteo achieved approximately 2x speedup in model training on NVIDIA Blackwell GPUs using the cuEmbed open library. That efficiency frees roughly 17,000 GPU hours per year. For Criteo, which reported activated media spend exceeding $1 billion in Q1 2026, the improvement reduces operational costs for maintaining recommendation quality across its network.

HiggsField orchestrates autonomous campaign workflows

HiggsField's Supercomputer handles the complete marketing workflow—from ideation through production, posting, and performance optimization—in a single interface. The platform orchestrates more than 35 image, audio, and video models, including proprietary Soul models built on NVIDIA Blackwell architecture.

NVIDIA's Agent Toolkit provides the control layer. The toolkit includes NemoClaw blueprints and OpenShell secure runtime, which provide safety guardrails, auditability, and role-based permissioning for autonomous agents in enterprise settings. Marketing campaigns for nearly 400 Fortune 500 companies are created on the platform.

Taboola and KERV.ai tackle content intelligence

Taboola uses NVIDIA GPUs to power DeeperDive, its AI answer engine that allows publishers to embed generative AI search directly on their websites. DeeperDive reached nearly 7 million monthly active users by April 2026. The economics model—publishers receive a share of advertising revenue without paying for the service—requires inference to be cheap enough to remain profitable at scale.

KERV's Moment Match Engine evaluates signals across every video frame, understanding individual scenes, objects, and products to match advertising to contextually relevant moments. The company recently achieved over 10x improvements in speed and efficiency using the NVIDIA Nemotron 3 Nano Omni open model.

Why it matters

US programmatic ad spending is expected to exceed $200 billion in 2026. The transition from rules-based to model-based decisioning requires infrastructure that can process billions of decisions in milliseconds. These deployments represent hardware vendors, cloud providers, and ad tech companies converging on GPU-powered, autonomous infrastructure as the primary competitive axis. Whether training speedups and causal modeling at SuperPOD scale translate to measurably better campaign outcomes will be answered in the coming quarters.

These details were first reported by PPC Land.