Companies Rethink AI Spending as Bills Soar
The era of cheap artificial intelligence is ending. Companies that rushed to adopt AI tools are now facing steep bills, forcing a reckoning about how and where they deploy the technology.
For the past three years, AI companies subsidized customer usage to build market share. OpenAI, Anthropic, and others charged minimal fees while investors covered the losses. That model is breaking down as the major players prepare to go public and answer to mainstream investors who expect profitability.
AI Agents Drive Costs Higher
The biggest cost driver is AI Agents & Automation. Unlike chatbots that answer questions, agents perform tasks-booking appointments, writing code, managing files. A single task can spawn dozens of agents working simultaneously, each consuming tokens, the unit AI companies use for billing.
One agent-powered task burns through dozens of times more tokens than a simple chat message. The price difference is stark: a large general-purpose model costs $15 per million tokens, while smaller specialized models cost five cents per million tokens, according to consultancy Enverso.
The Spending Binge Stops
Some companies went overboard. Meta encouraged employees to maximize token usage as a productivity metric-until executives realized the math didn't work. The company's chief technology officer told staff this month that using AI tools "just for the sake of using them" made no sense.
Uber's chief operating officer raised eyebrows this week by saying the company's AI spending showed no noticeable productivity gains. In some cases, token costs exceeded employee salaries within weeks of deployment, said analyst Jack Gold of J.Gold Associates.
Companies Shift Strategy
To cut costs, organizations are moving to free, open-source AI models that lack the sophistication of ChatGPT or Claude but handle many tasks adequately. Others are switching to smaller models built for specific industries like real estate or finance.
A third approach breaks large tasks into smaller steps and routes each piece to the cheapest model capable of handling it. This strategy can reduce costs by 99 percent compared to using one large model for everything.
Chip shortages and data center constraints add another layer of uncertainty. Demand for computing power outpaces supply, pushing costs higher across the industry. "The cost to use AI for coding has grown exponentially," said Mark Barton of tech consultancy Omniux.
The Market Fragments
AI is becoming a commodity business where price and fit matter more than brand. But the most advanced users will continue paying for state-of-the-art models, said John Belton, a portfolio manager at Gabelli Funds. For AI for Finance and other specialized applications, premium models may remain necessary.
The shift reflects a maturing market. Companies are moving past the initial adoption phase and into the harder work of justifying AI spending with measurable returns.
Your membership also unlocks:
