As AI becomes increasingly central to modern marketing, the conversation is shifting from capability to economics. Falling token costs may prove far more important than the next breakthrough model, determining whether AI remains a specialist tool or becomes the infrastructure that powers every campaign, customer interaction, and business decision.
For marketers, the AI conversation has largely centered on capability. Every new model promises better reasoning, stronger creative outputs, faster workflows, and more sophisticated agents. What has received far less attention is the economics that determine whether any of those capabilities can be deployed at scale. That conversation is now becoming impossible to ignore.
Nikesh Arora, CEO of Palo Alto Networks, believes the AI industry still has a long way to go before enterprise adoption reaches its full potential. Speaking with CNBC, Arora argued that token costs need to fall dramatically, suggesting reductions of as much as 90% over the next two years if AI is to become economically viable for widespread business use. His comments followed OpenAI CEO Sam Altman’s announcement that the company’s latest model is 54% more token-efficient for agentic coding, an improvement Arora described as encouraging but ultimately insufficient.
For the advertising and marketing industry, this is about far more than infrastructure pricing. It is about whether AI remains an experimental productivity tool used by innovation teams or becomes the operating system that powers every customer interaction, campaign optimization, media decision, creative workflow, and consumer experience.
Much of the industry’s AI strategy has been built on the assumption that costs will naturally fall over time. While that has historically happened with computing power, AI introduces a different challenge because every prompt, workflow, recommendation, and automated decision carries an ongoing computational cost. Unlike traditional software licenses, which become relatively predictable operational expenses, generative AI scales with usage. The more organizations rely on AI, the more they pay.
That creates an uncomfortable reality for marketers pursuing AI-first strategies. The brands extracting the greatest value from AI are typically those deploying it across thousands or even millions of customer interactions every day. Personalized creative generation, dynamic media optimization, AI-powered customer service, automated campaign reporting, retail media optimization, and intelligent commerce all rely on enormous volumes of inference. If every interaction carries a meaningful cost, scaling becomes a budgeting challenge rather than simply a technology challenge.
The economics become even more significant as AI agents begin replacing traditional software workflows. Marketing organizations increasingly envision AI systems that coordinate media buying, generate campaign assets, analyze performance, update CRM records, produce reports, and interact directly with customers. Each task may appear inexpensive in isolation, but multiplied across global marketing operations, token costs can quickly become one of the largest technology line items within the department.
This may also explain why so many enterprises remain cautious despite the industry’s relentless AI enthusiasm. Surveys consistently show strong executive interest in AI adoption, yet relatively few organizations have successfully deployed AI across every part of the marketing organization. Governance, trust, and data quality all contribute to that hesitation, but economics are becoming an equally important factor.
The conversation is also shifting toward open-weight models, which several technology leaders increasingly view as a way to reduce long-term operating costs. Last week, Palantir CEO Alex Karp criticized the token-based commercial model used by several frontier AI providers, arguing that many enterprises are spending excessive amounts simply consuming proprietary AI services rather than building more sustainable internal capabilities.
For marketers, this reflects a broader strategic decision that extends well beyond model quality. Organizations must increasingly decide whether they want to rent intelligence through proprietary APIs or invest in building AI infrastructure that they can control more directly. The first option offers speed and access to state-of-the-art models, while the second may ultimately prove more economical as AI becomes embedded throughout the business.
The emergence of increasingly capable open-weight models, including rapidly improving alternatives from China, only intensifies that debate. As performance differences narrow, procurement decisions may increasingly be driven by economics rather than benchmark scores. Marketing leaders rarely choose technology based solely on having the absolute best capabilities. They choose platforms that deliver the greatest business return, and AI will eventually be evaluated by the same standard.
Ironically, falling token prices may accelerate AI adoption far more than the next breakthrough model release. History suggests that lower operating costs consistently unlock larger markets than incremental performance gains. Cloud computing expanded because infrastructure became affordable. Digital advertising scaled because distribution costs approached zero. Streaming overtook physical media because delivery became dramatically cheaper. AI is unlikely to follow a different trajectory.
That has important implications for advertising agencies as well. Many agency business models currently assume AI can increase productivity while maintaining existing pricing structures. If token costs remain high, agencies may find themselves absorbing significant infrastructure expenses that gradually erode margins. Conversely, if inference becomes dramatically cheaper, agencies gain the flexibility to embed AI throughout strategy, creative development, production, optimization, measurement, and client service without constantly calculating computational costs.
The industry’s obsession with model releases sometimes obscures the more important trend. AI’s future may depend less on which company builds the smartest model and more on which ecosystem delivers intelligence at a price that makes everyday business use economically invisible.
For marketers, the next competitive advantage may not come from having access to the most powerful AI. It may come from being able to afford using that intelligence everywhere.