NVIDIA’s ad tech partners show what GPU automation looks like at Cannes Lions

Alembic will develop into the primary causal AI firm to run NVIDIA DGX Vera Rubin SuperPODs for enterprise-scale advertising and marketing measurement, as a cluster of promoting expertise corporations prepares to exhibit GPU-accelerated infrastructure at Cannes Lions in France this week.

On June 18, 2026, NVIDIA revealed particulars of a set of collaborations with six promoting and advertising and marketing expertise corporations – Alembic, Amazon Net Companies, Criteo, HiggsField, KERV.ai, and Taboola – which can be every bringing NVIDIA-powered infrastructure to the Cannes Lions Worldwide Competition of Creativity, scheduled from June 22 to 26 in Cannes, France. The bulletins span 4 distinct downside areas: causal measurement, real-time bidding, full-lifecycle advertising and marketing automation, and video content material intelligence.

The scope of the deployments is technical in nature. None of those are idea demonstrations. Every includes production-grade infrastructure choices – the selection of a selected GPU era, a selected inference server, an open library, or a runtime setting – which have direct penalties for the velocity, value, and accuracy of promoting programs operating at enterprise scale.

What Alembic is doing with DGX Vera Rubin

Causal AI sits on the middle of Alembic’s deployment. The corporate positions itself round a selected measurement downside: not simply reporting on what occurred in a advertising and marketing marketing campaign, however modeling what really induced enterprise outcomes. The excellence between correlation and causation in advertising and marketing information has occupied measurement groups for years, and causal modeling requires far higher computational assets than typical attribution or analytics.

In keeping with NVIDIA, NVIDIA DGX Vera Rubin NVL72 programs will allow Alembic to scale its causal AI fashions to investigate extra variables, run bigger simulations, and quantify the true drivers of development throughout advertising and marketing investments. The NVL72 designation refers to a selected configuration inside NVIDIA’s DGX Vera Rubin line – a system constructed round NVIDIA’s Blackwell-generation GPU structure. Working causal fashions throughout each channel, market, and viewers concurrently calls for the flexibility to course of huge datasets which can be continuously altering. Alembic’s inference runs on personal supercomputing infrastructure inside Equinix information facilities, which suggests AI workloads stay native to the place enterprise information already resides – a configuration that addresses each latency and information governance necessities. World Extensive Know-how extends this to safe and controlled environments.

The SuperPOD configuration – clusters of NVL72 programs working collectively – is what makes enterprise-scale causal modeling tractable. In keeping with NVIDIA, Alembic would be the first causal AI firm to make use of these programs for this goal. For advertising and marketing executives accountable for capital allocation choices, the sensible output is a single supply of what NVIDIA describes as “unbiased reality” on what drove enterprise outcomes and the place capital is being wasted.

The growing interest in causal measurement infrastructure displays a broader disaster in advertising and marketing analytics. Analysis revealed in February 2026 confirmed that as much as 75% of senior planning and analytics decision-makers at US manufacturers and companies report that superior measurement approaches – attribution, incrementality testing, and advertising and marketing combine modeling – fail to ship the rigor, timeliness, and belief wanted to justify spending. Causal modeling represents an try to repair that failure on the infrastructure stage.

AWS and Criteo: AI contained in the public sale

The second downside space is the public sale itself. Programmatic promoting auctions function in home windows measured in milliseconds. The constraint is bodily: any AI mannequin that runs too slowly misses the public sale fully. The programs that NVIDIA and its companions are demonstrating at Cannes are designed to suit inside these home windows.

Amazon Net Companies is presenting a production-ready reference implementation for AI-powered bidding inside auctions. In keeping with NVIDIA, AWS is bringing cloud infrastructure, basis fashions, and GPU-accelerated computing collectively right into a cohesive stack for the advert tech trade – one scaled for the period of AI brokers. The infrastructure targets demand-side platforms, supply-side platforms, and unbiased software program distributors, providing them a path to maneuver from rules-based decisioning to AI-powered fashions for bid worth optimization, viewers activation, and deal scoring, all working straight inside the reside public sale pipeline.

The important thing part enabling that is the NVIDIA Triton Inference Server, which in keeping with NVIDIA delivers deep studying inference quick sufficient to suit inside real-time public sale home windows. That technical element issues. Guidelines-based bidding programs make choices primarily based on pre-defined logic. AI-powered fashions rating every impression in opposition to a realized illustration of worth – a basically completely different calculation that requires considerably extra compute per resolution. Triton makes that compute possible at public sale velocity.

Criteo’s deployment addresses a special a part of the identical downside: preserving product suggestions related as buying conduct adjustments. Criteo operates one of many largest suggestion networks in digital promoting. Repeatedly retraining AI on billions of customer timelines is the mechanism by which these suggestions keep correct. Pace in retraining straight interprets to high quality in manufacturing.

Collaborating with NVIDIA, Criteo achieved roughly a 2x speedup in mannequin coaching on NVIDIA Blackwell GPUs, pushed by the NVIDIA cuEmbed open library. In keeping with NVIDIA, that effectivity already frees roughly 17,000 GPU hours per yr. The businesses at the moment are scaling the work additional. For Criteo, the development is consequential given the aggressive setting the corporate is navigating. Criteo’s Q1 2026 results confirmed activated media spend exceeding $1 billion in a single quarter for the primary time, whereas the corporate concurrently manages the downstream results of retail media shopper scope reductions. Effectivity on the mannequin coaching layer reduces the operational value of sustaining suggestion high quality throughout its community.

The trajectory Criteo has been on with AI infrastructure consists of the corporate’s emergence as OpenAI’s first and solely formal programmatic accomplice inside ChatGPT, the place AI-referred consumers have been changing at almost double the speed of conventional search referrals in classes like shopper electronics and residential and backyard. The cuEmbed-driven coaching enhancements reported at Cannes lengthen the identical infrastructure logic to suggestion high quality throughout the broader retail community.

HiggsField: the complete lifecycle in a single interface

The third downside space is the entire advertising and marketing workflow. Working a marketing campaign from ideation via manufacturing, posting, and efficiency optimization has traditionally required a number of instruments, a number of groups, and substantial guide coordination. HiggsField’s Supercomputer is positioned as a system that handles all of these steps autonomously.

In keeping with NVIDIA, HiggsField gives brokers that handle the complete advertising and marketing automation lifecycle – from marketing campaign ideation, planning, inventive manufacturing, and posting via autonomous marketing campaign optimization – in a single interface. The platform orchestrates greater than 35 picture, audio, and video fashions, together with HiggsField’s proprietary Soul and Soul 2.0 fashions constructed on NVIDIA Blackwell structure.

The agentic dimension is provided by NVIDIA’s Agent Toolkit. In keeping with NVIDIA, the toolkit consists of NVIDIA NemoClaw blueprints and the NVIDIA OpenShell safe runtime, which give the controls – security guardrails, auditability, and role-based permissioning – that make autonomous brokers deployable in enterprise settings. Inside HiggsField, NVIDIA Agent Toolkit software program, together with NVIDIA Nemotron open fashions, powers specialised subagents operating constantly inside each marketing campaign. NemoClaw and OpenShell are being built-in to offer what NVIDIA describes because the enterprise belief layer.

The sensible consequence is {that a} marketing campaign on HiggsField runs via a set of specialised subagents – one dealing with ideation, one other manufacturing, one other optimization – coordinated via a shared interface. In keeping with NVIDIA, advertising and marketing campaigns for almost 400 of the Fortune 500 corporations are created on the platform.

Why does the agent management layer matter? The ad tech industry’s broader push into agentic infrastructure has repeatedly surfaced the identical query: what stops an autonomous system from making expensive or brand-damaging choices with out human assessment? The NVIDIA OpenShell runtime is a direct reply to that query, establishing a safe boundary round what brokers can do and preserving a log of what they’ve performed.

Taboola: conversational AI with promoting in-built

Taboola’s utility of NVIDIA GPU infrastructure focuses on DeeperDive, its AI reply engine. The product permits publishers to embed a generative AI search and reply expertise straight on their very own web sites, drawing from the writer’s editorial archive somewhat than from a web-wide index. The industrial mannequin is determined by inserting contextually related promoting into AI-generated reply pages – turning reader questions into promoting stock. For that to work at scale, the inference layer powering DeeperDive needs to be quick and inexpensive sufficient to serve thousands and thousands of each day queries.

In keeping with NVIDIA, Taboola makes use of NVIDIA GPUs to energy DeeperDive and is extending that infrastructure to AI platforms and chatbots to allow them to generate income from promoting.

Taboola has been growing DeeperDive since launching it in June 2025, initially with Gannett’s USA TODAY Community and The Unbiased as launch companions. DeeperDive reached nearly 7 million monthly active users by April 2026 and expanded to French, German, Hebrew, Japanese, Korean, and Spanish, as the corporate constructed towards what CEO Adam Singolda has described as the biggest ad-supported massive language mannequin for the open internet. HuffPost UK adopted the system in April 2026 as search referral visitors continued to erode throughout the publishing trade.

The GPU infrastructure query for Taboola is value per question. Publishers integrating DeeperDive don’t pay for the service; they obtain a share of the promoting income generated. That economics mannequin requires inference to be low cost sufficient to stay worthwhile at scale throughout a community touching greater than 9,000 writer websites and roughly 600 million each day energetic customers. Taboola’s broader agentic trajectory, together with the April 2026 launch of Understand+ and its Claude Expertise integration, situates DeeperDive inside an organization that has been systematically shifting its infrastructure towards autonomous, AI-native operations throughout each the advertiser and writer sides of its enterprise.

KERV.ai: understanding each body of video

The fourth space is content material intelligence – particularly, the flexibility to know what is occurring inside a video on the stage of particular person frames, objects, and scenes, with a view to match promoting to contextually related moments.

KERV’s Second Match Engine evaluates alerts throughout each video body and media asset, understanding particular person scenes, objects, and merchandise, then gives content material suggestions primarily based on the visible and textual components of an commercial. In keeping with NVIDIA, KERV.ai lately optimized its processing pipeline, attaining over 10x enhancements in velocity and effectivity utilizing the NVIDIA Nemotron 3 Nano Omni open mannequin.

The benchmark that provides this declare context is MediaPerf, an open benchmark for AI video understanding. In keeping with NVIDIA, Nemotron 3 Nano Omni delivered the best throughput and lowest inference value of any mannequin evaluated on MediaPerf, open or closed supply. Ecosystem companions together with PYLER, which makes use of NVIDIA DGX B200 programs, additionally adopted the mannequin.

For advertisers, the sensible consequence is a tighter connection between inventive and context. KERV’s system analyzes what every advert or media transient accommodates, who it resonates with, and which precise second inside content material to focus on. That specificity – matching inventive to the proper body, not simply the proper program – is what KERV positions as “improved engagement.”

The infrastructure query behind all of it

What connects these six deployments is just not a single product or platform. It’s an infrastructure query: can the {hardware} operating AI fashions for promoting preserve tempo with the quantity, velocity, and complexity that trendy promoting requires?

US programmatic advert spending is anticipated to exceed $200 billion in 2026, in keeping with EMARKETER. Each public sale, each suggestion, each content material classification resolution inside that market includes an AI mannequin making a judgment in milliseconds. The transition from rules-based to model-based decisioning is nicely underway, however the infrastructure value of operating fashions at this scale is just not trivial. The bulletins at Cannes are, partly, NVIDIA’s argument that its GPU programs are the reply to that infrastructure downside.

The particular technical selections on show are revealing. Criteo’s 2x coaching speedup comes from cuEmbed, an open library for embedding operations – a slim however high-impact optimization for suggestion programs that course of billions of consumer timelines. AWS’s reference implementation makes use of Triton Inference Server, which manages the mannequin deployment layer in order that auction-speed inference turns into a configurable infrastructure selection somewhat than a customized engineering undertaking. Alembic’s transfer to DGX Vera Rubin NVL72 SuperPODs is a guess on scale: that causal modeling throughout each channel and market concurrently requires a stage of compute that solely devoted supercomputing infrastructure can present.

The agentic control layer is a second thread operating via all of those bulletins. HiggsField’s deployment of NemoClaw and OpenShell displays some extent that the IAB Tech Lab’s Agentic RTB Framework has been making since November 2025: autonomous AI programs operating inside enterprise promoting workflows want security guardrails, role-based permissioning, and auditability earlier than they are often responsibly deployed at scale. The NVIDIA Agent Toolkit is one reply to that requirement on the infrastructure stage.

WPP Media launched a midyear forecast on June 16, 2026, projecting world promoting to achieve $1.3 trillion in 2026, with AI funding named as the first driver. That forecast arrived within the week instantly previous the Cannes Lions competition, alongside a dense cluster of agentic product launches from Yahoo, Fox, Horizon Media, LiveRamp, Stagwell, and DoubleVerify – all inside 72 hours of one another. NVIDIA’s Cannes showcases are a part of the identical structural second: {hardware} distributors, cloud suppliers, and advert tech corporations converging on the identical argument that GPU-powered, autonomous promoting infrastructure is now the first axis of competitors within the trade.

Whether or not that argument holds on the margins – whether or not the 2x coaching speedup interprets to measurably higher marketing campaign outcomes, whether or not causal AI at SuperPOD scale adjustments what entrepreneurs really determine – is the query that the following a number of quarters of outcomes will reply.

Timeline

• June 4, 2025 – Taboola proclaims the complete industrial launch of Predictive Audiences with as much as 270% conversion enhancements for early adopters. PPC Land coverage
• June 11, 2025 – Taboola launches DeeperDive, an AI reply engine designed to reside straight on writer web sites, with Gannett and The Unbiased as launch companions. PPC Land coverage
• November 13, 2025 – IAB Tech Lab releases Agentic RTB Framework model 1.0 for public remark, standardizing containerized agent deployment in real-time programmatic promoting. PPC Land coverage
• January 5, 2026 – PubMatic launches AgenticOS, the primary working system constructed particularly for autonomous promoting execution. PPC Land coverage
• February 5, 2026 – Criteo proclaims an Agentic Commerce Advice Service attaining as much as 60% enchancment in suggestion relevancy. PPC Land coverage
• February 7, 2026 – IAB State of Information 2026 report finds as much as 75% of entrepreneurs say attribution, incrementality, and MMM underperform on rigor, timeliness, and belief. PPC Land coverage
• March 2, 2026 – Criteo introduced as the primary formal advert tech accomplice contained in the ChatGPT promoting pilot. PPC Land coverage
• April 10, 2026 – Taboola’s DeeperDive reaches almost 7 million month-to-month energetic customers and expands to 6 languages. PPC Land coverage
• April 23, 2026 – Taboola launches Understand+, an agentic AI system for open internet efficiency campaigns, alongside Claude Expertise integration. PPC Land coverage
• April 28, 2026 – LiveRamp provides NVIDIA GPU infrastructure to its clear rooms, enabling as much as 15x sooner AI mannequin coaching. PPC Land coverage
• Might 6, 2026 – Criteo studies Q1 2026 activated media spend exceeding $1 billion for the primary time; Taboola studies Q1 income of $466.4 million (+9.1%). PPC Land Criteo coverage – PPC Land Taboola coverage
• June 16, 2026 – WPP Media publishes midyear forecast projecting world promoting to achieve $1.3 trillion in 2026, naming AI funding as the first driver.
• June 17, 2026 – DoubleVerify launches DV Neura, its cognitive AI engine, with agentic execution capabilities and an almost 300x enhance in content material classification output. PPC Land coverage
• June 18, 2026 – NVIDIA publishes particulars of six promoting and advertising and marketing expertise partnerships to be showcased at Cannes Lions 2026 (June 22-26).

Abstract

Who: NVIDIA, alongside six promoting and advertising and marketing expertise companions – Alembic, Amazon Net Companies, Criteo, HiggsField, KERV.ai, and Taboola – are the central events. World Extensive Know-how and Equinix are named as infrastructure companions within the Alembic deployment. PYLER is known as as a accomplice adopting Nemotron 3 Nano Omni for video understanding.

What: NVIDIA revealed particulars of GPU-powered infrastructure deployments throughout 4 distinct promoting downside areas: causal advertising and marketing measurement (Alembic, utilizing DGX Vera Rubin NVL72 SuperPODs), real-time AI bidding (AWS, utilizing Triton Inference Server; Criteo, utilizing cuEmbed on Blackwell GPUs with a 2x coaching speedup and roughly 17,000 GPU hours freed per yr), full-lifecycle advertising and marketing automation (HiggsField, utilizing NVIDIA Agent Toolkit together with NemoClaw and OpenShell), and video content material intelligence (KERV.ai, utilizing Nemotron 3 Nano Omni with over 10x velocity enhancements). Taboola can also be named, making use of NVIDIA GPU infrastructure to its DeeperDive AI reply engine.

When: NVIDIA’s announcement was revealed on June 18, 2026. The Cannes Lions Worldwide Competition of Creativity, the place these partnerships are being showcased, runs from June 22 to 26, 2026, in Cannes, France.

The place: Cannes Lions 2026 in Cannes, France. Alembic’s inference runs on personal supercomputing infrastructure inside Equinix information facilities. HiggsField campaigns serve Fortune 500 corporations globally. KERV.ai’s MediaPerf benchmark is an open analysis platform. Taboola’s DeeperDive operates throughout its community of greater than 9,000 writer websites.

Why: The promoting trade’s shift from rules-based to AI-powered decisioning requires infrastructure able to operating advanced fashions at public sale velocity and enterprise scale. Demonstrating these deployments at Cannes Lions – the trade’s principal annual gathering for senior advertising and marketing and media executives – permits NVIDIA and its companions to place GPU-accelerated computing because the underlying requirement for autonomous promoting operations on the scale the trade now calls for.