By Arun Shastri, global AI Leader at ZS
Between 2023 and 2026, the price of a single AI token fell by more than 90%. Enterprise AI spending, however, has more than doubled. That isn’t a contradiction. As Apollo chief economist Torsten Slok has observed, when tokens get cheaper, companies don’t spend less —they run more agents, automate more workflows, and generate more code.
AI costs don’t spiral because of a few big decisions. They spiral because of thousands of small ones.
The monthly invoice is where the problem surfaces as token costs are visible and easy to track. But the invoice is only one outcome. The real drivers are harder to see. As organizations embed AI into more workflows, copilots, and agents, costs reflect thousands of everyday decisions about which models to use, how to route work, when to retry a task, and whether AI is the right tool at all.
The risk is concrete. Uber rolled out AI coding tools across its engineering organization in late 2025. Adoption climbed from roughly a third of engineers to more than 80% within a few months, and the company burned through its entire 2026 AI budget in four months. Per-engineer costs typically ran a few hundred dollars a month, but the heaviest users reached $2,000. As Uber’s leadership has acknowledged, the harder problem wasn’t the bill. It was that they couldn’t yet draw a clear line between all that consumption and measurably better products.
Uber’s experience illustrates the broader challenge: AI spending can scale faster than an organization’s ability to measure the value it creates. The goal isn’t just to control costs, but to keep AI investments aligned with business outcomes as adoption evolves.
Success starts by focusing on four things.
1. Know where AI costs really come from
Token consumption is becoming harder to predict. The models people choose, the workflows and agents they deploy, how often they rerun work, and the infrastructure behind it all shape enterprise AI costs. However, much of this activity is invisible. When AI runs on individual desktops, the enterprise can’t see it, version it, or account for it. When it runs in managed environments, that usage becomes visible, and leaders can understand it, govern it, and make better decisions about where AI creates value.
Even the most sophisticated companies are still building this. When Meta moved to rein in its own internal AI spending in 2026, with costs reported to be climbing into the billions, the reason it gave was telling: teams had limited visibility into what they were consuming. You cannot manage what you cannot see.
2. Choose the right model more of the time
If you don’t decide which models are right for which tasks, employees and agents will decide for you. You’ll either overpay by using premium models for routine work or lose control as different teams make different choices across the business.
Once you know where AI costs are coming from, you can begin to monitor and manage which models perform which work. Lower-cost models can often handle routine work, while frontier models may justify more complex or higher-risk tasks. The key is making those choices intentionally, not by default.
3. Treat inefficient AI use as a real cost
How people use AI has become a hidden tax. Many organizations design governance programs to manage access, security, and compliance. Far fewer design them to detect when human behavior makes AI unnecessarily expensive.
It’s not just weak prompts, repeated reruns, and buggy code. It’s also using AI where simpler technologies can do the job more efficiently.
This is where individual behavior compounds. Someone reaches for a frontier model when a smaller one would do the job. They rewrite the prompt several times, generate multiple versions, and keep trying until something works. Repeat that pattern thousands of times across the business and it becomes a meaningful cost driver. Often, a purpose-built tool, or a simpler model chosen deliberately, delivers the same outcome at a fraction of the cost.
4. Don’t leave AI costs for IT to manage alone
AI costs shouldn’t live only with IT. The teams deciding where and how AI is used should also understand what those decisions cost.
When teams can see the costs they create, they make better decisions about when premium models create value, when lower-cost alternatives are sufficient, and where AI belongs at all.
That’s a more durable approach than cutting licenses or imposing usage caps. As AI becomes part of everyday work, leaders have to manage AI costs the same way they manage every other business investment: by making deliberate decisions about where every dollar creates value.
Deliver value from your AI initiatives with ZS.
This sponsored post was supplied by ZS.
