How brands and agencies are grappling with rising enterprise AI costs

AI tools are getting even more expensive.

The Gemini app will now charge users based on a “compute-used model” that takes into account chat length, features used, and prompt complexity, the company announced at its annual Google I/O event in Mountain View, California, last week. It’s a departure from the daily prompt limits it previously used.

It’s a response to the staggering cost of providing AI to so many users, and the latest change affecting how major AI companies are charging for their products that could have broad business effects.

• Microsoft’s GitHub Copilot product is transitioning to usage-based billing in June;
• Anthropic’s Claude Enterprise made a similar switch in April.

As the cost of AI tools continues to climb, many brands and agencies are keeping an eye on how those changes could impact their budgets, according to brands and agencies that Marketing Brew spoke with. Some are also strategizing around how to charge their own clients for AI services they use in their work for clients, even as they seek to balance the value of leveraging AI internally with rising subscription costs.

The focus comes amid an increasingly public conversation about stomaching the cost of using AI tools at companies. Uber COO Andrew Macdonald recently said in an interview that high AI costs due to token consumption were becoming harder to justify.

Budget crunch: One such company, Full Glass Wine Co., has allocated new spend in its budget for enterprise AI that’s proven to be “very, very helpful for us,” Neha Kumar, co-founder and co-CEO, told Marketing Brew.

“We are building it onto the budget because it allows our team to be able to move so much faster,” Kumar said.

AI use has helped increase internal speed and efficiency, she said. The company uses both ChatGPT and Gemini across its organization, including for summarizing internal and external communications. Full Glass Wine Co. sommeliers, for example, use ChatGPT to dictate notes on wine pairings that they’d otherwise work with a copywriter on drafting.

“Instead of them sitting there and working with a copywriter for a day to craft out the messaging, you can essentially have ChatGPT now put all that together,” she said. “Would we want to factor that into [our] budget? Absolutely, and we have.”

Scaling for ambition: Pomo, a marketing technology startup, is keeping an eye on rising enterprise AI costs as it scales, Praneet Dutta, its co-founder and CEO, told Marketing Brew. Pomo leans on AI coding tools to build; some of the company’s engineers use OpenAI’s Codex, while others prefer Anthropic’s Claude Code. Pomo also builds on the enterprise AI platform Databricks, which is an investor in the startup.

“[Building] something of the scale and ambition of what we have would not be possible without the help of all of these amazing new AI-based technologies,” Dutta said.

Some brands that use tools from the major enterprise AI providers like OpenAI and Anthropic are charging their own clients based on usage. Pomo charges clients a retainer and then a percentage of their ad spend through the platform, which Dutta sees as a “proxy for usage.”

“Our pricing…scales with usage, so as token costs evolve, our revenue evolves with them,” he told Marketing Brew. “That’s a very different position from enterprises that are now retrofitting their cost structure for the AI era.”

It’s not just brands. Agencies are factoring enterprise AI costs into their invoices, too. The creative agency VML charges clients based on usage through its platform Open Action, according to Pablo Bertero, its chief innovation officer. Clients can choose multiple models through parent company WPP’s platform WPP Open.

How did we get here?: The usage-based billing that brands are now contending with wasn’t always the norm—there was a period some dubbed the AI “subsidy era.” However, expanding costly compute power to meet aggressive external demand has been a challenge for large enterprise AI providers.

Usage-based billing has been just one way for some of these companies to make more money as they scramble to amass more compute resources.

• Anthropic, for example, recently inked a deal with SpaceX in which the latter will provide data center access through May 2029, increasing Anthropic’s compute capacity.
• After the deal was announced, Anthropic doubled the five-hour rate limits on seat-based enterprise Claude Code subscriptions, among others.

The large enterprise AI providers also need substantial compute for their own internal use. Stratechery noted that Amy Hood, Microsoft CFO, touched on the company’s strategy around reconciling external demand, particularly around the cloud platform Microsoft Azure, in which users can build AI tools, with internal needs for AI-related assets like GPUs and CPUs, during the company’s earnings call in January.

Keep it together: Microsoft 365 Copilot now has over 20 million paid seats, CEO Satya Nadella told investors on Microsoft’s quarterly earnings call in April. While brands grapple with rising enterprise AI costs, demand is clearly still strong.