S4 Capital's AI Token Costs Are 'Exploding' — Now It Faces a Margin Squeeze

S4 Capital's AI Token Bill Threatens Hard-Won Margin Gains

S4 Capital’s digital production unit Monks is guzzling AI tokens at a rate that has co-founder and chief AI officer Wesley ter Haar calling it “exploding.” That surge, driven by heavy use of generative AI tools across coding and creative workflows, is colliding with the company’s renewed focus on profitability. Last week S4 reported first-half operating profit doubled to £35.2 million and margins widened to 12.3%, sending shares higher. The turnaround came after years of headcount cuts — from over 8,200 people to 6,150 — and strict cost discipline.

Now Sir Martin Sorrell, S4’s chair, admits the company has “probably been ill-disciplined” on token spending. With Goldman Sachs forecasting a 24-fold jump in enterprise token consumption between 2026 and 2030, the cost line is set to swell. Sorrell calls the situation an “open question” and hints at a tougher spending policy, even as he insists that heavy investment in capability may not be a bad thing.

The tension is stark: S4 needs to keep its AI advantage to win business from technology clients, but unfettered token usage could eat into those freshly restored margins. Meanwhile, rivals like Texas-based PMG have already imposed a $50 daily cap on employee token use, and Forrester research shows 60% of agencies still prioritise tool licenses over cloud compute costs — a model that will look expensive when third-party providers shift to consumption-based pricing.

Where S4 and the Agency Model Stand on AI's Cost Cliff

S4’s Margin Playbook vs. Its AI Identity

The profit jump and share price bump signal that investors are rewarding S4’s cost control. Any backsliding into “overindulgent” token usage — especially after Sorrell’s own use of the word “ill-disciplined” — could invite swift market punishment. Yet Monks’ identity as an AI-first production house means it cannot simply cap usage across the board without risking its competitive differentiator. The company is trying a “horses for courses” approach: caps on lower-value tasks, open throttle for high-value coding that drives client outcomes.

Token Economics Are Reshaping Agency Pricing

Ter Haar acknowledges that project-based billing, where token costs are bundled, already carries margin risk. For the company’s flagship Monks.flow platform, costs are passed through at cost — no extra margin. As Goldman Sachs’ projections play out, that model becomes commercially painful. S4’s answer is to pivot toward subscription and outcome-based pricing. Subscription already accounts for nearly a quarter of revenue and is growing fast; almost all new business lands there. But Forrester’s Jay Pattisall warns that investor subsidies for AI costs are masking the true expense. When the “fallacy of this business model” becomes obvious, the real cost of AI could make automated advertising more expensive than human labour — a scenario he calls potentially “devastating” for an industry that shed 8% of staff in 2025–26.

The Procurement Hurdle

CMOs may find outcome-based pricing intuitive, but corporate procurement teams prefer rate cards for easy comparison. Ter Haar admits it is “a difficult sell” in RFIs. With only 9% of agencies monetising generative AI, S4 is ahead of the curve — but that also means it must educate a market that isn’t yet comfortable buying on deliverables rather than hours or seats.

What S4’s Dilemma Means for Investors and Agency Chiefs

• Monitor S4’s subscription revenue progress against the 25% target by end-2026. The speed of conversion from project work will directly determine whether rising token costs are covered by predictable revenue or become a margin drag.
• Watch for token cost disclosure in the next S4 earnings call. If management begins breaking out AI-related spending or hints at soft caps, it signals that the “open question” is hardening into a concrete policy — which will affect margin guidance.
• For agency CFOs and CEOs: The days of passing token costs through as a bundled overhead are numbered. A pricing model that ties fees to outcomes or usage will become a competitive necessity, not a differentiator. The PMG example shows early movers are setting a benchmark clients may demand industry-wide.
• Hiring and capability planning must factor in Pattisall’s warning. If investor-subsidised AI costs unwind, the labour-cost equation could temporarily reverse. Agencies that have cut creative staff too aggressively may find themselves short-handed if a rebalancing occurs.