Enterprise AI Spending Will Reach $64 Billion: Where the Money Is Actually Going

Gartner published a forecast on July 20 projecting that worldwide spending on AI models and platforms will reach $64.3 billion in 2026, an increase of roughly 63% over 2025. Within that total, spending on generative AI models is expected to grow about 117%, and spending on domain-specific and specialized models is forecast to grow about 210%. Gartner also notes that buyers are increasingly prioritizing cost visibility, model evaluation, reliability and measurable outcomes.

Those are large numbers, and large numbers tend to produce the wrong conversation. The interesting part of the forecast is not the total. It is the composition of the total, and what that composition says about how enterprises now expect to be billed, governed and held accountable for AI.

Specialized models are growing faster than general ones

The fastest-growing category in Gartner's forecast is not the frontier model. It is the narrower model tuned to a domain or a task. That shift has a straightforward economic explanation. A general-purpose model priced for open-ended reasoning is expensive to run against a question that a smaller model answers correctly at a fraction of the cost.

For an IT or HR service function, most inbound volume is repetitive and well-bounded: password and access issues, software requests, onboarding steps, policy questions, device problems. Routing that volume to the largest available model is a defensible engineering decision and an indefensible financial one.

The question for a budget owner is who absorbs that inefficiency. Under a metered contract, the customer does. Under a flat one, the vendor has every incentive to solve it.

Metered AI pricing puts a tax on adoption

Most enterprise AI is now billed by consumption in some form: per interaction, per resolution, per token, per agent action. The model is common enough that buyers have stopped questioning it. They should.

Consumption pricing creates a structural problem specific to employee service: success generates volume. Employees who get fast, accurate answers ask more questions. They bring problems to the service desk that they previously worked around, absorbed, or asked a colleague to solve. That is the entire point of deploying the system, and under a metered contract, it is also the thing that makes the invoice grow.

The consequences are predictable and, in our experience, common:

• IT throttles rollout to the departments already budgeted for, and adoption stalls at a fraction of the organization
• Teams discourage "low-value" queries, which are frequently the ones causing the most cumulative lost time
• Finance discovers the cost curve one quarter late, and the renewal conversation starts from a defensive position
• The business case built on adoption is undermined by the pricing model attached to it

An organization should never have to decide whether an employee's question is worth the marginal cost of answering it.

Rezolve.ai charges the same price every invoice

Our pricing is not metered. It is not per interaction, per resolution, per token or per agent action. You agree a price, and that is the price on every invoice: whether volume holds flat, doubles, or triples as adoption spreads.

That has three practical effects for a finance owner:

The forecast is the actual. The number you put in the budget is the number that arrives. There is no reconciliation exercise, no true-up, no variance to explain.

Adoption is free to grow. Roll out to every department at once. Encourage employees to bring more to the service desk, not less. There is no internal rationing conversation because there is nothing to ration.

Unit economics improve automatically. With a fixed denominator, every additional resolution lowers your cost per resolved request. Under metered pricing, that number is roughly constant no matter how well the deployment goes. Under ours, scale is the return.

We would rather compete on whether the platform resolves work than on how efficiently we can meter it.

The metric that matters is cost per resolution

Gartner's observation that buyers now want measurable outcomes points at a real gap in how AI service tools have been evaluated. Volume metrics (conversations handled, questions answered, tickets touched), measure activity. They do not tell you whether the work was finished.

Cost per resolved request is a harder number to produce and a far more useful one. It forces the platform to distinguish between an answer and a resolution, and it makes the finance conversation concrete. Roughly 70% of requests on Rezolve.ai are resolved before they become tickets, which changes the denominator of that calculation: the spend is measured against work that never entered the queue at all.

That framing makes ROI arguments defensible under scrutiny. Black Angus Steakhouse reduced after-hours IT dependency from 90% to 10%. MyEyeDr resolves issues in roughly 10 minutes, with about 30% of issues never becoming a ticket. JLL runs Ask Ethics across roughly 100,000 employees in more than 80 countries, a scale at which a per-interaction meter would materially shape how the organization chose to use it.

What to ask vendors this budget cycle

If your organization is participating in the 63% increase, the diligence questions have changed:

1. What is on the invoice if usage triples? Show the number, not the rate card.
2. Are there overage charges, consumption tiers or true-ups of any kind?
3. Does anything in the contract create a reason to limit who we roll out to?
4. Which model handles which class of request, and who absorbs the cost of that choice?
5. Which actions can an agent take autonomously, and where does an approval gate sit?

That last question is a financial control as much as a security one. Under any pricing model, an autonomous agent operating without an approval boundary is an unreviewed commitment of company resources.

Predictability and governance are the same discipline

The forecast describes a market buying more models. The organizations that get value from that spend will be the ones that also bought the layer above the models: routing, approval gates, audit trails, outcome measurement, and a commercial model that does not penalize the adoption those investments depend on.

Access to capability is now the easy part. Making it predictable (with autonomy that is approval-gated, audited and explainable, on a price that does not move), is where the budget either compounds or leaks.

See what a flat invoice looks like against your own volume: book a demo or review our pricing.

Figures cited are drawn from Gartner's publicly released July 20, 2026 forecast summary.