Nvidia is using Cannes Lions, the advertising industry's flagship June 22-26 gathering in France, to position Blackwell-class infrastructure as the new substrate for adtech. Six partners — Alembic, Amazon Web Services, Criteo, Higgsfield, KERV.ai and Taboola — are demonstrating production deployments, including a roughly 2x training speedup at Criteo that already frees 17,000 GPU hours a year. The message to ad buyers and platform operators is that AI-powered bidding, creative and measurement are shipping, not slideware.
The pitch lands at a moment when adtech is moving from rules-based decisioning to live, model-driven bidding inside auction windows measured in milliseconds. Nvidia's framing is that infrastructure, not strategy, now decides which advertisers and platforms can keep up.
“The digital era gave the advertising and marketing industry speed; the AI era is giving it autonomous operations.”— Jamie Allan, Nvidia
Causal AI platform Alembic is anchoring the enterprise-decisioning side of the showcase. The company is deploying Nvidia DGX Vera Rubin NVL72 systems and will become the first Causal AI vendor to run on DGX Vera Rubin SuperPODs, with inference hosted on private supercomputing infrastructure inside Equinix data centers so customer data stays local. World Wide Technology is packaging the stack for regulated environments.
Key facts
- 01Criteo achieved a roughly 2x speedup in model training on Nvidia Blackwell GPUs using the cuEmbed library, freeing roughly 17,000 GPU hours a year.
- 02Higgsfield's Supercomputer agent platform orchestrates 35+ image, audio and video models and runs campaigns for nearly 400 of the Fortune 500.
- 03KERV.ai logged over 10x improvements in speed and efficiency on its Moment Match Engine using Nvidia Nemotron 3 Nano Omni.
- 04Alembic will be the first Causal AI company to deploy Nvidia DGX Vera Rubin SuperPODs, running inference inside Equinix data centers.
- 05Cannes Lions runs June 22-26 in France, with the Nvidia-led adtech showcase announced June 18.
Amazon Web Services is contributing a reference implementation that lets demand-side platforms, supply-side platforms and independent software vendors run AI-powered bidding directly inside live auctions. The stack leans on Nvidia Triton Inference Server to deliver deep-learning inference fast enough to fit inside real-time auction windows across billions of daily transactions.
Criteo is the clearest performance story. Working with Nvidia, the recommendation network hit a roughly 2x speedup in model training on Nvidia Blackwell GPUs using the open cuEmbed library, freeing roughly 17,000 GPU hours a year that previously went to retraining models on billions of shopper timelines. That capacity is being redirected into further scaling rather than cost-cutting.
Taboola is applying the same infrastructure logic to conversational surfaces. The company is using Nvidia GPUs to power DeeperDive, its AI answer engine, and is extending the underlying infrastructure to third-party AI platforms and chatbots that want to monetize through advertising — a direct play at the emerging market for ads inside AI assistants.
On the agent side, Higgsfield AI is showcasing Higgsfield Supercomputer, a platform that runs the full marketing lifecycle — ideation, planning, creative production, posting and optimization — through autonomous subagents in a single interface. It orchestrates large language models alongside 35-plus image, audio and video models, including Higgsfield's proprietary Soul and Soul 2.0, built on Nvidia Blackwell. Campaigns for nearly 400 of the Fortune 500 already run on the platform.
The enterprise-trust layer comes from the Nvidia Agent Toolkit, which bundles NemoClaw blueprints and the OpenShell secure runtime, plus Nvidia Nemotron open models powering specialized subagents. The combination is meant to address the practical blockers — safety guardrails, auditability, role-based permissions — that have kept agentic systems out of regulated marketing workflows.
KERV.ai's Moment Match Engine handles the content-understanding side, analyzing every video frame and media asset to match ad creative to scenes, objects and products. KERV.ai logged over 10x improvements in speed and efficiency after adopting Nvidia Nemotron 3 Nano Omni, and on MediaPerf, an open benchmark for AI video understanding, Nemotron 3 Nano Omni delivered the highest throughput and lowest inference cost of any model evaluated, open or closed. Ecosystem partner PYLER is running the same stack on Nvidia DGX B200 systems.
The open questions are commercial rather than technical. Adtech margins are thin, programmatic supply chains are crowded with intermediaries, and "AI-powered" bidding has to clear measurable ROI bars before holding companies and brands shift budget. None of the partners disclosed lift figures from live campaigns at Cannes, and the claim that nearly 400 of the Fortune 500 run campaigns on Higgsfield says nothing about spend concentration or retention.
Nvidia's Cannes presence underscores how aggressively the company is moving past chips-as-product into vertical reference stacks. Advertising joins finance, healthcare, robotics and manufacturing as a beat where Nvidia is bundling silicon, inference servers, open models and an agent runtime into something close to a turnkey enterprise platform — and forcing every cloud, every adtech vendor and every creative-tool maker to decide whether to build on top of it or around it. For an industry whose holding companies have spent two years debating AI strategy, the infrastructure has now arrived, priced and benchmarked.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
