How to Respond to the Coming AI Cost Shock

Adapted and summarised for PTCG Consultoria de Gestão from an article by Stacia Garr

Original article published on 17 August 2026

Summary. Artificial intelligence adoption has benefited from heavily subsidised pricing by enterprise software providers, but that period is drawing to a close. As commercial models move towards usage-based charging, organisations are replacing part of their fixed labour costs with variable AI consumption costs. Managing AI therefore ceases to be merely a technology procurement decision and becomes a matter of organisational design, budgeting, talent and risk.

Major enterprise platform providers have absorbed a significant share of the costs of processing, inference and model usage. This policy accelerated experimentation, but it also created an incomplete view of AI’s true economic cost. Once consumption exceeds the included allowances, the initial incentive quickly becomes a real expense.

This transition is already visible. According to Stacia Garr, Oracle includes basic usage while charging for premium models; SAP follows a similar approach; and Workday plans to begin charging for excess usage of certain application programming interfaces from 31 January 2027. The direction is clear: enterprise AI pricing will increasingly depend on actual consumption.

Even at higher prices, AI may continue to generate value. The central question is where, at what scale and with what financial exposure. When a company integrates AI agents into its processes, it shifts resources from capacity it controls – its internal workforce – to capacity it contracts and pays for on a variable basis. This shift requires economic and organisational discipline.

Model the True Elasticity of AI Costs

Setting an overall budget or purchasing a specified number of tokens is not enough. Cost depends on variables that the organisation does not fully control: interactions per user, units consumed per request, models used and price per unit. The first step is to measure the work actually performed by AI, its total associated cost and the value created by each process.

Total cost also includes infrastructure, energy, internal technology teams, training, integration, security, oversight and change management. Organisations should calculate the return on investment for each use case and determine, for every process, the maximum unit cost that remains economically justified.

Protect Critical Functions

Agent-based workflows can become deeply embedded within an organisation. Essential processes are redesigned around the technology, while proprietary data logic may become dependent on a single vendor environment.

Replacing people with AI can also remove technical and institutional knowledge. If a tool becomes too expensive, is discontinued or fails, the company may discover that it no longer has the internal capability required to perform a critical function.

It is therefore essential to identify the functions whose knowledge must remain under the organisation’s control, even when day-to-day execution is automated. In some areas, retaining internal specialists with oversight capability may be sufficient; in others, it will be prudent to secure continued access to external experts.

Contracting arrangements should reflect this exposure. Contracts should include billing caps, consumption alerts, transition periods before price changes, portability rights and clear rules for unused credits. Where demand is uncertain, a more conservative initial purchase reduces the risk of paying for capacity that is never used.

Integrate AI into Workforce Planning

AI costs should not remain isolated within an information technology budget. They should be assessed alongside the costs of personnel, training, process redesign, controls and the potential loss of knowledge. Only this integrated view allows organisations to compare automation, reskilling and recruitment accurately.

Some companies already incorporate this logic into their planning tools, enabling them to compare team performance against planned headcount and AI usage budgets, while assessing alternatives between role automation and workforce reskilling.

AI will probably continue to justify its cost across many functions. However, an organisation should not redesign critical processes around promotional or difficult-to-predict pricing. The current period should be used to measure, test, negotiate and preserve options. Competitive advantage will come not only from adopting AI faster, but from understanding and controlling its economics.

Sources and Notes

Stacia Garr, How to Respond to the Coming AI Cost Shock, Harvard Business Review, 17 August 2026. Accessed 22 September 2026.