US Agencies Face Massive AI Spending Surge as Subsidies End
McKinsey warns US government agencies of a massive AI spending surge as subsidized tool access expires and usage explodes across the federal sector.
United States government agencies are standing on the brink of a massive financial shock regarding their artificial intelligence (AI) expenditures. A new report from McKinsey has issued a stern warning: as the current era of heavily subsidized AI access expires, agencies will likely face a dramatic spike in costs driven by exploding usage and a shift toward commercial market pricing.
The Subsidy Cliff: From $1 Tools to Market Rates
The report reveals a unique economic situation within the federal government. In 2025, introductory federal agreements allowed agencies to access cutting-edge AI tools at nominal prices—some as low as USD 1. However, these arrangements are set to expire throughout 2026 and 2027.
As these deals phase out, agencies will be forced to transition to standard commercial pricing. This shift comes at a time when AI adoption is not just growing, but accelerating. McKinsey notes that while the cost to run a fixed level of AI capability has plummeted nearly 1,000-fold over the last three years, the total enterprise bills are actually climbing because usage is exploding at a much faster rate than the cost of technology is falling.
Exploding Use Cases and the Rise of AI Agents
The scale of AI integration within the US government is immense. According to the 2025 inventory, federal agencies reported 3,611 AI use cases, representing a 69 per cent increase compared to the previous year. The breakdown of these activities includes:
- 1,800 use cases: Already deployed or currently undergoing testing.
- 445 use cases: Classified as high-impact operations.
- $90 billion+: The estimated potential value of federal AI contracts.
A significant driver of these rising costs is the emergence of AI agents. Unlike simple chatbots that respond to single queries, these agents can perform complex, multi-step workflows. These advanced tasks can consume between five to 30 times more tokens than a standard query, rapidly inflating the digital "fuel" required to keep them running.
A Global Trend of Budget Overruns
The struggle to control AI costs is not exclusive to the public sector. McKinsey’s research, which surveyed 120 companies, found that a staggering 93 per cent of them had already exceeded their planned AI budgets. This highlights a universal difficulty in predicting and managing the financial impact of rapid AI adoption.
McKinsey’s Roadmap for Financial Survival
To avoid "surprise bills" and the potential stalling of critical programs, McKinsey urges government leaders to treat AI spending as a distinct budget category. The report suggests several immediate strategic shifts:
- Forecast at Market Rates: Agencies must stop budgeting based on "pilot prices" and instead use post-subsidy commercial rates to plan for the future.
- Implement Cost Controls: Establish strict spending limits and automated alerts to prevent runaway costs.
- Optimize Model Selection: Direct tasks to the least expensive AI model that can still meet the necessary quality and performance requirements.
- Demand Transparency: Pursue shorter, more flexible contracts and insist on transparent pricing models from vendors.
- Focus on Outcomes: Regularly review spending to ensure that AI investments are delivering measurable, tangible benefits.
“Those who wait risk surprise bills, stalled programs, and a much harder conversation with appropriators,” the report concluded, emphasizing that the foundation for tomorrow's massive bills is being laid right now.