By Seb Joseph  •  January 13, 2026  •

Ivy Liu

In his telling, AI has yet to graduate from pilot programs into anything that materially reshapes how marketing departments are staffed, structured or paid. Until that changes, agency chiefs can stop bracing for a shift to outcome-based compensation. CMOs, he said, are still buying time and headcount. 

“Without naming names, I can remember a real situation where the marketer said ‘of course we’ll move to an output model’ and then you get into a conversation with their procurement people and they’re just worried about getting caught out,” he continued. 

Which, in plain terms, means they’re worried about paying agencies more than they have to. After all, outcome-based pay turns marketing from a predictable labor expense into a variable, performance-indexed cost that can swell when campaigns work, and in turn complicate budgeting, forecasting and cost control.

The only force that changes that is the CFO. And that moment arrives only when companies can no longer meet near-term performance expectations by trimming headcount, tightening scopes and layering AI onto existing workflows. 

For now, there is little evidence it has arrived. Sorrell pointed to a set of economic signals that reinforce that inertia: S&P 500 third quarter earnings per share were up 12% and up 9% when excluding the hyperscalers. Morgan Stanley’s internal forecasts also show double-digit earnings acceleration this year. Goldman Sachs has projected a similar trajectory.

“You don’t get change in legacy companies until there’s a real reason to do it,” said Sorrell.

Cynics might take his comments with a grain of salt. Last year, Sorrell worked to position S4 Capital as an early leader in the AI transition, That positioning has yet to translate into a financial windfall, and the company remains under pressure. Dialing back the hype now reads less like a reversal than a reset — a way to recalibrate expectations as much as to critique the market. 

“Clients are comfortable with models they understand,” said Sorrell. “When there’s economic pressure then they get pushed.”

If and when that outlook changes, he wants S4 Capital positioned to take advantage of it. Senior execs there are already in discussion with partners including Nvidia about how to do that. 

“Cost of inference will be a recurring conversation this year,” said S4 Capital’s chief AI officer Wesley ter Haar. “That’s where all of our conversations sit with Nvidia and some of the other inference providers.”

By “cost of inference”, ter Haar is talking about compute — the price of running AI systems at scale. In pilots, those costs barely register. But once AI moves into everyday use across large marketing organizations, compute becomes a core production cost — closer to media inventory or cloud infrastructure than labor.

That is the pressure point. When machines become the dominant cost of delivering marketing, agencies can no longer charge primarily for hours and headcount. They have to price what the technology produces. Until those costs become large and unavoidable, the old system holds. When they do, it does not.  

“If you start thinking about burning tokens at enterprise scale then your inference suddenly becomes a massive part of those commercial discussions,” said ter Haar. 

As those costs grow, the conversation shifts to where AI actually runs, and whether part of it moves in-house, into private data centers where computing power is cheaper and more predictable. At that point, the debate stops being about innovation. It is about economics. 

“That’s the takeaway for me this year: it’s how companies are going to buy their inference, and how much of that is going to be back on premise,” said ter Haar. “I think there’ll be some really interesting infrastructure conversations that are maybe a little against where the trend has been last year.”