Businesses face potentially steep increases in artificial intelligence (AI) costs as major enterprise software providers begin withdrawing the subsidies that helped accelerate corporate adoption, according to an analysis from Harvard Business Review.
Software vendors have so far absorbed much of the cost of graphics processing units, inference and tokens, offering premium AI tools as complimentary, included or effectively unmetered features.
That approach encouraged companies to experiment rapidly and build new processes around the technology. However, some have already exceeded the subsidised usage available to them.
Uber and other businesses reportedly exhausted their entire 2026 AI budgets within months, while one company spent US$500 million in a single month after failing to restrict employee licences.
Vendors are now introducing usage-based models as investors push for returns on the costly infrastructure underpinning AI services.
Oracle includes a base model with subscriptions but charges for premium usage. SAP is preparing a similar structure, retaining free basic queries while charging for more advanced features.
Workday plans to begin charging overages for application programming interface usage after 31 January 2027. Until then, customers have a grace period to investigate and optimise consumption without discouraging AI adoption.
The shift could make companies vulnerable to unpredictable bills. In one illustrative example, a 10,000-person business using a premium agent at a relatively modest rate could see annual costs rise from $3,600 to $27,600 as employee adoption increases from 10% to 50%.
Those estimates cover only one system and exclude infrastructure, energy, technical support, training and change-management expenses. Costs could also double if a vendor raised its unit price from one cent to two cents.
Businesses should therefore measure how much work AI performs, the savings or revenue it generates and the highest per-unit cost that would still deliver an acceptable return, the analysis said.
They should also retain human expertise in essential functions, particularly where processes have been redesigned around a vendor’s technology. Contracts should include spending caps, notice periods for price increases and rights to roll unused credits into future periods.
AI tokens should ultimately be treated as part of workforce planning rather than simply an IT expense, allowing leaders to compare automation costs with employee headcount, reskilling and the risk of losing institutional knowledge.
The benefits of AI may continue to outweigh its costs, but companies were urged to use the remaining subsidised period to prepare for a more uncertain pricing environment.
