U.S. agencies face higher AI costs in 2026-2027

The consultancy said introductory federal agreements in 2025 made some AI tools available for as little as $1, but those deals will not last forever. As agencies move to market rates, McKinsey expects costs to climb even though the price of running a fixed amount of AI capability has fallen roughly tenfold a year and about 1,000-fold over three years. In other words, the technology is getting cheaper, but the way organizations use it is getting dramatically more expensive.

That is the key economic insight. Governments, like corporations, are discovering that AI spending behaves less like a one-time software purchase and more like a usage-based utility bill. McKinsey said federal agencies reported 3,611 AI use cases in their 2025 inventory, up 69% from a year earlier, with about 1,800 already deployed or being tested. It also estimated the potential value of federal AI contracts at $90 billion or more. That is a large enough market to matter for vendors, budget planners and taxpayers alike.

The risk is especially acute as agencies move from pilot projects to broader deployment. McKinsey said AI agents, which can carry out multi-step workflows, often consume five to 30 times more tokens than a simple chatbot query. That means the most ambitious and productive applications are also the ones most likely to blow through budgets. For public institutions under scrutiny from appropriators, that can quickly become a political and operational problem, not just an accounting issue.

The warning extends well beyond Washington. McKinsey said a survey of 120 companies found 93% had exceeded planned AI budgets, underscoring how hard it is for enterprises to contain costs once adoption spreads. That is an important reminder for investors that AI monetization is likely to be uneven: demand is real, but buyers may be slower to realize how much they are spending until usage scales.

For the market, the implications are two-sided. AI infrastructure and software suppliers benefit from rising consumption, but customers face margin pressure and budget discipline. That tension helps explain why investors are watching not only who sells AI, but who can deliver it efficiently. In the public-sector market, that could favor large vendors with trusted procurement relationships, pricing flexibility and broad product portfolios, including Accenture and IBM, while punishing suppliers that depend on short-lived pilot deals.

McKinsey’s advice is practical: forecast costs at post-subsidy prices, set spending limits and alerts, and route work to the cheapest model that still meets quality standards. It also recommended shorter contracts, transparent pricing and regular reviews of whether AI spending is producing measurable gains. That is sound guidance for agencies, but it is also a useful framework for investors trying to separate durable AI demand from hype.

The bigger narrative here is that AI adoption is still in its early innings, and the bill is only now starting to show up. The companies and agencies that treat AI as a strategic, measurable operating expense — rather than an experiment with hidden usage costs — will be the ones best positioned to compound value over the next several years. For long-term investors, that makes AI vendors with real efficiency advantages worth watching, but it also argues for patience and discipline.