Why Advertising Holdcos Are Treating AI Compute Tokens Like Media Inventory

Why Agencies Are Treating AI Tokens Like Media Inventory

Holding companies are increasingly adopting a playbook they know well: bulk-buying AI compute tokens upfront and embedding the cost – plus a margin – into principal media deals with clients. The model mirrors how agency groups have long managed media inventory, pricing the uncertainty that clients cannot forecast and profiting from the spread.

The economics are becoming unavoidable. After two years of building AI infrastructure and absorbing running costs on their own balance sheets, many agencies can no longer afford to keep those costs to themselves. As AI adoption accelerates and autonomous agent workloads drive token consumption higher, the capex must show up somewhere, or shareholders will start asking why it hasn’t.

Some agencies are already signing real contracts on this basis, not just floating the idea in pitches. IAB Europe chief economist Daniel Knapp summed it up: “No one really knows quite yet how to price that in, so you’re seeing the business sort of try and come up with solutions to try and figure that out.” In a Digiday poll, 42% of respondents said agencies shouldn’t become token futures markets, while only a minority saw it as a legitimate revenue opportunity – yet the model is gaining traction because it solves an immediate cash‑flow problem.

The client gets cost certainty for a line item they can’t forecast, and the holdco gets a shot at pricing power that its traditional time‑and‑materials fee model has never provided. But the arrangement rests on a fragile premise: that token prices will rise from today’s widely‑believed subsidized levels, making advance buying a hedge as well as a sales gambit. Whether that bet pays off depends on just how accurate agencies’ demand forecasts turn out to be.

Inside the Holdcos’ Token Arbitrage Strategy and Its Blind Spots

The Principal Media Playbook Meets AI Compute Costs

Principal media has long allowed agencies to put a number on media inventory uncertainty, buy in bulk and resell at a margin. AI tokens are now being fed through that same machine. The logic is straightforward: nobody can say with confidence where token prices will land next quarter, let alone next year, so an intermediary that aggregates demand and carries the risk can offer both sides a price. Holdcos are effectively turning themselves into a futures market for compute, profiting from the spread between bulk purchase rates and what clients are willing to pay to avoid volatility.

Why Agencies Need This Revenue Stream

For the past two years, most holdcos have quietly carried AI infrastructure costs on their own balance sheets, using it as a competitive differentiator. That arrangement had a shelf life, and it ran out this year. “What we saw in the last two years is that most agencies were carrying those costs entirely on their balance sheet, and they’ve been using it as a differentiator to win clients,” said Ruben Schreurs, CEO of Ebiquity. “But now, starting this year, agencies can’t afford to continue to subsidize all those costs.” Bundling AI costs into principal media margin sidesteps the difficult conversation about why bills aren’t shrinking as AI replaces headcount, while still funding the infrastructure agencies have already built.

The Oversupply Risk Nobody Is Discussing

The model has a critical failure mode. If an agency locks in three years of token volume at a fixed rate and then dramatically improves its prompting efficiency or agentic orchestration, it could end up needing a fraction of the tokens it committed to. Those excess tokens would have to be resold, potentially at a loss, or written off entirely. Schreurs called this “obviously the danger” sitting underneath the whole arrangement, and it’s a risk that few agency leaders are publicly acknowledging. The same efficiency gains that would delight clients could blow a hole in the balance sheet of an agency that bet too heavily on volume.

What This Means for Client‑Agency Dynamics

Clients outsourcing token procurement gain certainty but sacrifice any meaningful cost benchmarking. One executive near the negotiations told Digiday: “You just don’t know quite how much the agency pays for it versus you.” With no equivalent of a media rate card, procurement teams are effectively buying AI consumption on trust. Ana Milicevic of Sparrow Advisers described token‑heavy deals as a “big bargaining chip” that allows agencies to bundle compute and media discounts to win business – but the opacity also hides whether the client is genuinely getting a better deal or simply being charged for capacity the agency has committed to anyway.

What Advertisers and Agencies Should Watch For

• For advertisers: Request a transparent breakdown of AI token costs within principal media fees. Without one, you cannot verify that the bundled price reflects true consumption, not the agency’s own over‑commitments.
• For advertisers: Push for contractual safeguards, such as caps on token spend or clauses that allow adjustment if the agency’s unit cost is demonstrably below the markup. The present opacity leaves procurement with no benchmark.
• For agencies: Model the volume risk carefully. A multi‑year bulk purchase at today’s subsidized prices locks in a margin only if your clients consume the expected volume; any significant efficiency gain in prompting or orchestration could leave you holding expensive, underutilized capacity.
• For agency CFOs: Ensure that token commitments are matched by flexible resale or sub‑lease arrangements with AI providers, so that excess capacity can be offloaded without a full write‑off.
• For industry bodies: The lack of a common rate card or benchmark for AI work creates a transparency vacuum. Developing even a voluntary framework for disclosing AI cost components would help forestall trust crises with clients.