Introduction

AI was expected to become cheaper as models improved, but the opposite is happening. Token rates may be falling, yet the complexity of AI tasks has exploded, pushing overall usage – and costs – well beyond projections. Enterprises embedding AI into their products now face invoices that grow faster than their budgets. For lawyers, the challenge is clear: contracts must not only secure functionality but also protect clients from escalating and unpredictable AI charges.

Why AI Costs Keep Climbing

The reason costs rise so quickly lies in how AI systems work. A single inference may be inexpensive, but modern AI rarely involves a single pass. Multi-step reasoning, retrieval of external data, and code execution all magnify token usage, sometimes into the millions. At the same time, vendors are bundling AI into premium tiers or shifting to usage-based pricing that erodes predictability. Introductory discounts may hide the true cost of long-term adoption. Without contractual protections, companies risk entering into agreements that deliver short-term innovation but long-term financial strain.

Key Contractual Protections

Lawyers negotiating AI agreements must focus on pricing and usage terms with the same rigor traditionally reserved for liability and intellectual property. Several strategies are available.

1. One is to insist on fixed pricing for an initial multi-year term, not just twelve months. Better still, clients should negotiate rights to renew at pre-set prices for additional periods, such as two or three years, so they can plan budgets with confidence. To provide flexibility, renewal increases can be capped – linked to CPI or limited to a small percentage.

2. Tiered pricing discounts are also critical. If usage grows, the unit cost should fall, not rise. Contracts can provide that higher volumes trigger progressively lower per-token rates. This encourages scale without punishing success. Similarly, clients can negotiate “all-you-can-eat” models where unlimited usage is allowed within a range of overage. This gives enterprises predictability even when they defined miscalculate demand.

3. Overage charges themselves should be tightly regulated. The agreement should define them clearly and require notice when thresholds approach. Clients should be able to elect a higher plan, rather than being penalized with exorbitant overages. Some contracts allow retroactive conversion—so if a client’s usage unexpectedly doubles, charges are recalculated at the higher-tier rate rather than billed as raw overage.

4. Other protections include most-favored-customer clauses, benchmarking rights, and detailed reporting obligations. Vendors should provide usage dashboards and early alerts, giving customers the ability to control spending before it spirals. Where higher fees are tied to “priority processing” or enterprise features, Service Level Agreements should specify exactly what performance improvements are delivered and what remedies apply if they are not.

5. Finally, lawyers should preserve flexibility. Termination rights for excessive price increases, combined with data portability and migration assistance, prevent lock-in.

Conclusion

AI cost inflation is no longer hypothetical. Vendors are experimenting with aggressive pricing models that shift risk to the customer. Legal counsel must respond by negotiating contracts that secure multi-year stability, introduce predictable pricing tiers, regulate overages, and preserve client flexibility.