OpenAI just dropped a fresh set of tools for its ChatGPT Enterprise customers—usage analytics and tighter spend controls—that feel less like a product update and more like a response to a growing headache in the boardroom. If you’ve ever watched an AI bill balloon after a team of engineers started feeding a model endless prompts for code generation, you know the feeling: the technology is amazing, but the invoice can arrive like a surprise party nobody asked for.

The timing isn’t random. Over the past year, generative AI has slipped from experimental side-project to everyday workhorse. Companies are sprinkling GPT-4o, Codex and the like across everything from customer support bots to internal knowledge bases, and the token-based pricing model that powers those services is proving to be a double-edged sword. A token is just a fragment of a word, yet every request—whether it’s a quick summarisation or a marathon session of autonomous agents chewing through logs—eats up a steady stream of them. When usage spikes, the cost can climb without anyone actually deciding to spend more, leaving finance teams scrambling to explain why the AI line item looks more like a luxury car lease than a utility bill.

That disconnect has started to show up in the headlines. Earlier this year, Uber’s CTO told The Information that the company burned through its entire 2026 AI coding budget by April after usage among its 5,000 engineers jumped from roughly a third to over eighty percent in just three months. Microsoft, meanwhile, pulled back its own developers’ AI coding licences after realizing the spending was outpacing the budget. And a Priceline engineer shared with TechCrunch that a routine contract renewal for a coding tool came back four to five times more expensive than anticipated. Stories like these aren’t outliers; they’re becoming the pattern as AI adoption outpaces the traditional guardrails that companies use for cloud spend or software licences.

OpenAI’s answer lands in the shape of a refreshed Global Admin Console. The dashboard now pulls together ChatGPT and Codex credit consumption into a single view, breaking it down by user, product and model. For an admin, that means you can finally see whether a surge in spend is coming from a handful of power users churning out code, a new team experimenting with retrieval-augmented generation, or perhaps an autonomous agent that’s been left running overnight. The console lets you track usage trends over time, spot the top consumers, and drill down into exactly where each credit is going. And because the same data is exposed through a unified Cost API, organizations can pipe the numbers into their own finance or BI tools if they prefer to keep everything in one place.

Beyond visibility, the update gives admins real levers to act on what they see. You can set a default credit limit for the whole workspace—think of it as a baseline budget that everyone starts from. Then you can carve out group-specific caps for teams that need tighter reins, say, the legal department experimenting with contract-review bots versus the sales team using AI for lead enrichment. If a particular engineer genuinely needs more runway to finish a critical project, you can’t increase the limit for everyone else. The employee side gets a mirror of that transparency: users can view their own credit consumption against their allotted budget, request extra credits when they hit a wall, and attach a short note about what they’re working on so the approver can judge whether the ask makes sense.

The idea is to move away from the blunt instrument of “one size fits all” limits and toward something that feels more like the way companies already manage SaaS licences or cloud instances—flexible, observable, and grounded in actual work patterns.

OpenAI isn’t the only one noticing the shift. A recent Gartner projection puts global AI spending at $2.59 trillion for 2026, a nearly 50 % jump from the year before, and much of that isn’t just chips and data centres—it’s the tab that runs up every time an employee shoots a prompt at a model. In parallel, the Linux Foundation announced the Tokenomics Foundation, a nonprofit aimed at creating vendor-neutral standards for measuring and governing AI token spend. Its early backers read like a who’s who of enterprises most exposed to the problem: Accenture, Google Cloud, IBM, JPMorgan Chase, Microsoft, Oracle, Salesforce, SAP and Booking.com, among others. The foundation’s premise is simple: if you can’t trust each provider’s internal meter, you need an independent way to track the flow of tokens the same way you’d watch electricity usage on a smart grid.

What’s interesting is how quickly the conversation has moved from “do we need better models?” to “how do we pay for them sensibly?” Enterprise buyers told Menlo Ventures in their 2024 survey that price barely factored into their tool selection—just 1 % called it a concern—yet they’re now asking for the very controls that would make pricing predictable. The emphasis has swung toward ROI, industry-specific fit and, increasingly, governance.

When you talk to the people actually using these tools day-in, day-out, the story feels less abstract. Ryan Oksenhorn, co-founder of drone-delivery startup Zipline, told OpenAI that his engineering team had gone all-in on Codex since January, and as the broader organization started to adopt it, they realized they needed a way to both spot the hold-outs and keep spend from spiralling. “We asked the team at OpenAI to build usage analytics to help find and train-up folks who haven’t adopted Codex, and for granular usage controls to keep spend predictable,” he said. “These new tools are helping us faster scale productivity of our employees while keeping safeguards in place.”

That sentiment captures the broader mood: enterprises want to keep the momentum of AI adoption humming, but they also want the kind of oversight that stops a fun experiment from turning into a budget-blowing nightmare. By giving admins a clearer picture and more nuanced controls, OpenAI is effectively handing enterprises a set of training wheels—ones that can be tightened or loosened as the organization learns how its teams actually use the technology.

Looking ahead, the real test will be whether these tools become the foundation for a new discipline—AI FinOps, if you will—where spend, usage and impact are monitored in the same breath. The current release doesn’t yet tell you whether a token-heavy workflow is actually delivering value; it just tells you how many tokens you burned. As Forrester analyst Biswajeet Mahapatra warned recently, token consumption measures activity, not impact. But having the data in hand is the first step toward pairing it with outcome metrics—things like time saved, tickets resolved, or revenue generated—to answer the harder question that every CFO eventually asks: “Is this AI spend worth it?”

For now, the rollout is a concrete sign that the providers of large-language models are listening to the very customers that helped push AI from research labs into everyday work. The technology will keep getting smarter, and the bills will keep growing. What’s changed is that enterprises no longer have to guess at where those charges are coming from—they can see them, shape them, and, hopefully, make sure the AI they’re paying for is actually pushing the business forward.