Cannes Sells AI Transformation, Industry Grapples With Costs

Per reporting by Digiday, private conversations at the 2026 Cannes Lions Festival shifted from promotional AI hype to concrete worries about operational cost. Ian Maxwell, CEO of Converge Digital, is quoted saying "if you throw a whole code base at it the costs become absolutely astronomical" and that rising token costs can make AI "vastly more costly than simply having engineers." Digiday also cited "a senior marketer who was not authorized to speak to Digiday on the record" saying cost is "already becoming a conversation in the conversations we're having with media agencies."

Per a Publicis Groupe press release, the agency used Cannes to emphasise "real business value" from AI. The release says Publicis held a flagship session for 350 clients and 70 investors under Chatham House rules, held more than 60 closed-door sessions, and announced on-stage participation by executives from Mars Inc and The Coca-Cola Company. Publicis CEO Arthur Sadoun is quoted: "The compound effect of over promising on AI and unsustainable commercial offers in pitches to generate headlines is leading to massive jobs cuts in our industry."

Companies deploying generative workflows and large-scale inference commonly see costs driven by model size, inference frequency, and token billing for API-based models. Practitioners shifting production workloads from human-only to hybrid human+AI workflows often encounter higher-than-expected cloud and API expenses, and recurring inference costs that scale nonlinearly with usage.

Agencies and large enterprises frequently start with pilot deployments that understate ongoing operational and data costs. Public-facing pitches often emphasise productivity multipliers; trade reporting now shows practitioners reconciling those claims with line-item budgets and vendor billing realities.

• •Adoption signals for cost-management tooling: meterers, prompt optimization, batching, quantized or distilled on-prem models.
• •Contractual shifts in agency-client agreements, including performance-linked fees or clearer scope definitions in AI-enabled pitches.
• •Vendor pricing changes from major cloud and API providers in response to enterprise pushback on token and inference costs.

Teams moving from prototypes to production should instrument real usage, model-call patterns, and unit economics early. Tracking per-call costs and building conservative forecasts is a pragmatic risk-reduction step as agencies scale AI-enabled offerings.