The era of "spend now, optimize later" AI is officially over. Uber burned through its entire annual AI budget in four months. A 25-person startup now spends more on AI than on its entire payroll. And Gartner is warning that AI coding token costs will rival the average software engineer's monthly salary within two years. CFOs who have been patient observers of the AI boom are now taking the wheel — and the implications for every enterprise are significant.
This is not a slowdown in AI adoption. Enterprise AI has never been more embedded in daily operations. What's changed is the accountability threshold. The question enterprises are now asking is not "Are we using AI?" It's "What are we getting for every dollar we spend on tokens?"
From Tokenmaxxing to Token Governance
For the past two years, a dynamic took hold inside enterprise technology teams that nobody formally endorsed but nearly everyone practiced. Call it tokenmaxxing: the organizational habit of pumping as many tokens as possible into AI models, rewarded by internal leaderboards and fueled by the assumption that more AI usage automatically equaled more productivity.
Companies built cultural norms around heavy usage. Engineering managers tracked token consumption as a proxy for AI engagement. Developers competed to use the most. Nobody asked whether all those tokens were producing proportional output.
The math only worked while the bills were still manageable. They no longer are.
Gartner's June 2026 analysis found that AI coding token costs are on track to meet or exceed the typical software engineer's monthly salary within two years. The baseline they used is a global average salary of $2,000 per month — but Gartner senior principal analyst Nitish Tyagi noted the individual cases are already well beyond that benchmark.
"I have heard scary numbers like 'My developer consumed $20K last month,' or 'A business user consumed $32K,'" Tyagi said. The goal of the warning, he added, is to "alarm the industry about the impact of token cost if it is not governed and controlled."
Those are not outlier anecdotes. They are increasingly representative of what happens when token consumption scales without governance frameworks in place.
The Numbers No CFO Can Ignore
The most cited case study of unchecked AI spend is Uber. In April 2026, the company's CTO Praveen Neppalli Naga disclosed that Uber had exhausted its entire annual AI budget in just four months. The company responded by implementing a formal spending tier system: a base level of $1,500 per employee per month for AI tools, with higher access requiring explicit approval.
That kind of governance didn't exist before the budget ran out. It was reactive. The CFO playbook now requires building these controls before the crisis, not after.
The same dynamic is playing out at smaller companies with even more acute consequences. Flo Crivello, CEO of AI startup Lindy, made a decision this month that would have been unthinkable six months ago: he moved 100% of his company's AI traffic from Anthropic's Claude models to DeepSeek, a Chinese provider of lower-cost open-weight models. The switch will save millions of dollars within months, he told CNBC.
The striking part is what comes after the savings. Even after the switch, Lindy's AI costs will still exceed the company's entire payroll for its roughly 25-person team. "It's a matter of survival for the business," Crivello said.
When a company of 25 people spends more on AI than on human salaries — even after aggressively cutting costs — you have a structural cost problem that requires structural solutions, not just tighter budgets.
The Analyst View: Growth Is Peaking
Wall Street is watching these dynamics closely, particularly because both OpenAI and Anthropic filed confidentially for IPOs in early June 2026. The timing is deliberate.
Gil Luria, equity analyst at D.A. Davidson, put it plainly to CNBC: "Current growth rates for Anthropic and OpenAI are the fastest they will ever be, which is mostly a matter of basic math." The risk he flagged is that large enterprise customers are starting to limit what he called "out-of-control token spend" — and both companies have strong incentives to go public before that correction fully materializes.
The numbers are still impressive. OpenAI is now processing more than 15 billion tokens per minute across its APIs. Anthropic reported an annualized revenue run rate of $47 billion as of May 2026. Enterprise customers now represent more than 40% of OpenAI's total revenue, with that share expected to reach parity with consumer revenue by end of year.
But Gartner's Tyagi identified a key structural problem: "There is no direct relationship between the number of tokens developers consume and their productivity gains." More tokens do not equal more output. They equal a larger bill.
Optimizing token consumption, Tyagi said, is what actually increases quality and productivity — not maximizing consumption. That insight reframes the entire budget conversation.
What CFOs Are Actually Doing Now
The CFO response to AI cost inflation is not uniform, but several patterns are emerging clearly across industries.
Spending tier implementation. Uber's approach — tiered access with defined monthly caps and an approval process for higher usage — is becoming a template. The mechanics vary by company, but the principle is consistent: default access at a reasonable spend level, with exceptions requiring explicit business justification. This creates natural pressure to demonstrate ROI before consuming more tokens.
Model routing by task complexity. One of the most effective cost control strategies emerging in 2026 is matching AI models to task complexity rather than defaulting every query to frontier models. Currently, according to Glean CEO Arvind Jain, roughly 95% of enterprise AI usage still runs on frontier models — meaning the most expensive option for every task, regardless of whether that power is needed.
Model routing addresses this by sending simple, well-defined tasks to cheaper, smaller models and reserving frontier model access for genuinely complex work. The savings potential is significant. The challenge is that the tooling to automate this routing is still maturing.
Open-source and cost-tier alternatives. The DeepSeek migration at Lindy is an aggressive example of a broader trend: enterprises are evaluating non-frontier, open-weight alternatives for workloads where output quality is adequate. Microsoft, Amazon, and Google are all positioning efficiency-focused model offerings to compete on total cost of ownership, not just raw capability.
Formal AI spend governance structures. The Linux Foundation recently announced plans for a Tokenomics Foundation aimed at standardizing how the industry measures and governs AI token spend. That kind of institutional response signals how far the problem has spread. Companies that establish internal token governance ahead of this industry standardization will have a structural advantage.
Shifting the productivity metric. The traditional lines-of-code metric no longer applies when AI can generate entire libraries in seconds. Gartner recommends reframing productivity measurement around value delivered: feature velocity, quality improvements, time-to-feedback reduction. This shift in measurement methodology changes what optimization looks like — and makes token consumption a cost input, not a success metric.
The Technology Vendors Are Responding
OpenAI and Anthropic have both recognized that unrestricted token consumption is not a sustainable position for enterprise relationships. OpenAI launched enterprise spend controls earlier in June, giving administrators the ability to break down credit spend across teams, set usage limits, and give individual employees visibility into their available budgets.
Anthropic has rolled out parallel controls for enterprise accounts. Both moves acknowledge what CFOs have been saying for months: the tools that AI vendors provide for cost management have been far behind the enterprise need.
OpenAI is also making a longer-term bet on an enterprise platform called OpenAI Frontier, designed to let organizations deploy and manage AI agents across their business rather than in isolated pockets. Customers including Oracle, State Farm, and Uber are early participants. The strategy is to move from token-based consumption toward outcome-based enterprise relationships — a model that could change the cost conversation fundamentally if it delivers on its premise.
The CFO Playbook: Four Priorities for H2 2026
Based on what's working across enterprises that have gotten ahead of this cost curve, here is where to focus before year-end.
First, establish baseline visibility. You cannot govern what you cannot see. Before building spending controls, map where AI consumption is occurring: which teams, which tools, which models, and what tasks. Most enterprises discover significant concentration — a small number of teams or use cases driving a disproportionate share of token consumption.
Second, implement tiered access with business justification gates. Following the Uber model is a reasonable starting point. Define a default monthly spend level, require explicit approval for higher tiers, and make the approval process fast enough that it does not create organizational friction that slows legitimate high-value AI work. The goal is friction for waste, not friction for productivity.
Third, pilot model routing on your highest-volume workloads. If 95% of your consumption runs on frontier models, there is almost certainly a meaningful subset that does not require that level of capability. Identify your three or four highest-volume AI workloads and evaluate whether smaller or open-weight models deliver acceptable quality. Even shifting 20-30% of consumption to lower-cost alternatives can materially change the cost trajectory.
Fourth, redefine the productivity measurement framework before you expand AI access. Connecting AI spending to business outcomes is the foundational requirement for sustainable enterprise AI. Without that connection, every budget discussion is a negotiation without data. The teams that establish outcome-based metrics now will have a significant advantage when the next generation of AI capabilities arrives and the pressure to adopt immediately begins again.
The Bigger Picture
The shift happening right now is not a retreat from enterprise AI. OpenAI's CRO Denise Dresser described in April what she sees as two simultaneous trends: rapid adoption and a cost-driven correction. These are not contradictory. They are what maturation looks like.
The companies that will emerge from this inflection point with competitive advantage are the ones that treat AI cost governance as a strategic capability, not an administrative function. CFO involvement in AI strategy is not a slowdown signal. It is the sign that AI has become important enough to manage like every other significant business investment.
The tokenmaxxing era produced a generation of AI-aware organizations. The accountability era that follows it will produce AI-effective ones.
What's your organization's AI spend situation right now? Are you seeing token cost pressure hit your budgets? I'd be interested in what governance approaches are actually working at the enterprise scale you're operating at.
Follow me on LinkedIn or Twitter/X for more enterprise AI analysis twice weekly.
