Christian Stegh, CTO and VP of Strategy at eGroup Enabling Technologies.
getty
Just as GenAI begins to show significant ROI, a new layer of complexity enters the equation. After hooking users with unlimited access, AI providers are implementing consumption-based billing for advanced capabilities such as coding assistants and AI agents.
The AI subsidy era has ended, ushering in the age of tokenomics. This poses a new challenge for enterprise IT and finance leaders.
The potential of tools like Cowork, Scout, Claude Code, Codex and GitHub Copilot is undeniable. So too are the costs. After exhausting its annual Claude Code budget in just four months, Uber’s CTO remarked that “I’m back to the drawing board, because the budget I thought I would need is blown away already.”
The cost equation is no longer a simple linear one.
How are enterprises supposed to budget and fund their ongoing AI investments? There are no best practices yet, but historical precedents can help. This article provides some suggestions of how to approach the problem and a potential solution.
Varying Mindsets In The C-Suite
First, who’s primarily funding AI investments?
KPMG’s “Global CEO Outlook 2025” showed that 69% of CEOs planned to spend 10% to 20% of their overall budget on AI in that year. In another 2025 poll from IBM, 18% of leaders surveyed said GenAI funding is coming from net-new spend.
But in a mid-2026 poll, only 6% of organizations are funding GenAI from centralized corporate budgets. Sixty-two percent are paying for AI from IT’s budget, and 31% have no formal GenAI budget at all.
What If The Status Quo Continues?
Even when IT adds a line item, the CIO’s budget can’t keep pace with AI. Gartner’s recent poll showed an overall 14% increase in IT budgets for 2026, but RAM costs are suddenly cutting into that allocation this year.
If IT allocates nothing for AI at all, then the line of business (LOB) will find its way into shadow AI. Said VP of Global Technology Tim Bachta, “If you’re not bringing AI to your staff, they’re bringing it themselves. Why not give them the right tools and visibility?” If not, CISOs will have an eventual governance, risk and compliance nightmare.
Yet tokenomics is an unpredictable equation. IT may decide to wait for predictable pricing that may never come. If they freeze completely and budget for nothing, then LOBs will head back into the shadows.
And what if IT does budget something for associates’ extra use of tokens? Most pay-as-you-go services like Copilot Cowork allow limits to be set on both individual users and the organization as a whole. That type of short-term control is what Uber reportedly invoked, allocating $1,500 per user per month, mainly for coding assistants. And when users exceed their spending limits, will they head back into the shadows?
At best, budgeting for tokens has created an uncomfortable current situation. At worst, it’ll lead to an uncontrollable future.
A Proposed Way Forward
To support the current and future states, CFOs could consider separating the investments into two categories:
1. A baseline of enterprise-class AI tools, which IT would fund.
2. All other consumption-based AI products and features would be charged back to the LOB using the service.
Put differently, IT would fund enterprise AI tools such as M365 Copilot, ChatGPT Enterprise and Claude Enterprise. This is the approach taken by Verdantas, whose results now include shaving cycle time from between 10 and 14 days to hours.
The LOB would fund add-on consumption services, like Microsoft’s Cowork and Scout, OpenAI’s Operator and Codex and Claude Cowork and Code. In addition, departments would fund their own metered AI agents or other specialized AIaaS. Essentially, any tokens used beyond the baseline tools.
IT’s Responsibilities would be to provide base tools like ChatGPT Enterprise, M365 Copilot and Claude Enterprise, which would mean funding the baseline service, securing, training, monitoring and ensuring success.
And in situations where the business requires a unique solution, IT will help facilitate a good, secure decision; implement and monitor the identities, data and APIs; and maintain governance.
LOB’s responsibilities would include identifying business problems, engaging IT for support, collectively seeking options, assessing business value and being data stewards (if not owners).
And in situations where a unique solution was chosen, they would finance it (tokens, SaaS, training), determine spending limits and measure and ensure success.
Other Recommendations To Reduce Token Costs
Leverage Lower-Cost Models
Some of the capabilities of Copilot Cowork, for instance, are certainly impressive. With a little more coaxing, the same results can be achieved with the base Copilot license, simply by using the right prompt-model combinations.
Evaluate Results
Evaluating AI’s efficacy in pilots, estimating costs and then extrapolating those costs at scale can provide a safe on-ramp to evaluating the financial impact.
Enable The Right People
Rather than saying “Yes” to everyone who has an interest or “No” to everyone in the organization, a “Yes if” approach is recommended. This allows particular use cases to be supported. “Yes if” you are a departmental champion. “Yes if” you have taken the survey and training. “Yes if” you have an idea that is useful for multiple people or teams.
Empower The Right Departments
If enough teammates in a specific workgroup are demonstrating viable use cases, consider having the department foot its own AI bill (see above). Consider this nonnegotiable if/when they request a specific Agentic AI solution or third-party service.
Conclusion
Treating every AI expense as an IT cost will eventually constrain adoption. Treating every AI expense as a business expense will create risk in the shadows. A compromise is shared ownership, where IT provides the secure foundation and the business funds the workloads that create measurable value.
When organizations establish clear funding, clear ownership and clear success metrics, they can move beyond debating AI costs and start capturing AI value.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
