This insight summarises Hotels Need an AI Compute Budget Before It Becomes an IT Crisis, by Adam Mogelonsky and Larry Mogelonsky, principals at Hotel Mogel Consulting Ltd., published on Hotel Speak on 11 November 2026.
AI is a variable cost. Unlike a traditional software licence, AI's main operating cost moves with use. In the authors' words, "every prompt, workflow, agent, translation, report and recommendation quietly consumes 'compute'", which is mostly energy plus costs such as datacenter debt service. Few hotels are stress-testing it.
Value still comes first. The authors previously argued that hotel technology should become "a zero-sum exercise", with each new platform replacing an old one. Using an analogy of golden apples that cost twice as much per bushel but carry the potential for net new value, they now argue that technology should be judged by the value it creates, not only by what it costs.
Token creep. AI costs build up quietly through token consumption. With public inflation, providers openly raise token prices, API or MCP rates, or subscription tiers, which procurement can see. With secret shrinkflation, buyers get less compute for the same price: tighter quotas, advanced reasoning models moved into premium add-ons and lower tiers deprioritised at peak times.
The pressures behind the price. Frontier models need GPUs, datacenter capital, cooling and grid power, and better models generally need more computation. Datacenter financing is opaque, and IPOs could force higher pricing. New energy supply takes years, and AI companies could face taxes that are passed on to the businesses using tokens. The authors call these last three factors speculation.
Where hotels will feel it. Many AI capabilities are currently embedded in PMS, CRM, revenue management, booking engine and customer service software at little apparent extra cost. At renewal, vendors may introduce separately priced AI modules, usage caps, rising fees across multi-year contracts or token thresholds.
Compute governance, not a slowdown. The authors do not want adoption slowed. They want it "accompanied by 'compute governance'": modelling token sensitivity over the next three to five years, as finance teams already do for labour, utilities and insurance. Their test scenarios include AI costs rising 20% a year, embedded AI modules becoming separately licensed and staff usage doubling within the same allocation. Token creep, they write, will "emerge gradually with one renewal, one usage cap and one premium feature at a time."
AIHA practical takeaway. List every system in your stack with embedded AI, with its renewal date, usage limits and how AI is priced today. Ask each vendor in writing what will happen to AI features, caps and pricing at renewal. Run the authors' three scenarios against your 2027 technology budget, and give each AI tool a named owner who tracks its usage.
Related AIHA insight: Budget the Meter, Not the Machine: Planning for 2027 When AI Runs on Usage.