Federal AI contracts carried a potential value of $91.8 billion in 2026, according to an analysis the Brookings Institution published May 181. That same measure stood at $4.6 billion two years earlier, and at $355 million back in 20221. The newest jump totals 1,912 percent inside two budget cycles.
Chart 1
Federal AI contract value jumps from $355 million to $91.8 billion in four years
Potential value of federal AI contracts, 2022 to 2026, in billions of dollars
Each dot marks one year Brookings tracked, and the line's steep bend between 2024 and 2026 shows where almost all the growth happened.
Source: Brookings Institution [1]. Chart by The AILately.com Desk.
The numbers behind this chart
| Period | Value | Note |
|---|---|---|
| 2022 | $0.355B | |
| 2024 | $4.6B | |
| 2026 | $91.8B | A 1,912 percent increase from 2024 |
Read quickly. That climb says Washington finally caught up to the industry, chasing agentic tools into every department at once. A narrower number sits underneath it. The Department of Defense holds $90.7 billion of the $91.8 billion, a share of 98.9 percent1.
A separate count comes from the Center for Strategic and International Studies. It puts the department at 64 percent of federal AI contracts and 83 percent of obligations since 2019, and both shares keep rising2.
Kevin Desouza is a senior fellow at Brookings. He also teaches at Queensland University of Technology, and he wrote the plain version with his colleagues, James Denford and Gregory Dawson: "the Defense Department grew its AI investment to such a degree that all other agencies effectively became a rounding error"1. Commerce reported $197 million in AI contract value for 2026. Four more agencies barely register. Health and Human Services reported $138 million, NASA $45 million, the Interior Department $1.5 million, and the Nuclear Regulatory Commission $890,0001.
Chart 2
The Defense Department holds 98.9 percent of the $91.8 billion federal AI total
Share of 2026 federal AI contract value, by department, in billions of dollars
Source: Brookings Institution [1]. Chart by The AILately.com Desk.
The numbers behind this chart
| Item | Value |
|---|---|
| Defense Department | $90.7B |
| Whole | $91.8B |
Those five agencies run Medicare claims, hurricane forecasts and land management. They also issue the licenses that keep nuclear reactors supplying a sixth of the country's electricity. Their combined AI budgets fall short of the rounding error the Pentagon produced inside a single year.
Counted by use cases instead of dollars, civilian agencies rank as the busiest part of government. Agencies logged 3,611 AI use cases in the 2025 inventory the Office of Management and Budget released in April, Nextgov/FCW reported3. That count is up 105 percent from 1,757 the year before, spread across 56 agencies. Health and Human Services led with 447 cases. NASA followed with 425, and the Veterans Affairs Department logged 367, of which 215 carried the formal "high-impact" label3.
Labels decide who watches the tool. The VA's inspector general found in June that more than 15,000 staff were actively using VA GPT or Microsoft 365 Copilot Chat inside clinical settings, Nextgov/FCW reported4. Those tools landed outside high-impact status, and outside the oversight that label requires. Its own language was direct: the gap creates risk to patient safety and limits the department's ability to track AI-related errors4.
Ambient AI Scribe, a narrower tool built only for clinical documentation, carries the high-impact label and the safeguards that come with it. The chat window a clinician reaches for in a hurry keeps the lighter label, and the lighter oversight that comes with it.
Michael Lynch, GSA's deputy administrator, offers the sunnier version of the same math. GSA's Public Buildings Service shrank fast. It fell from more than 5,600 employees to about 3,100 over the course of 2025, a Government Accountability Office review found6. Lynch answered questions about the leaner staff in April by pointing to a "million hours challenge" built on the agency's internal chatbot: "We have about 400,000 hours that are currently identified of ways that we can — not replace people — but remove that non-high-value added time and replace it by putting people on more high-value opportunities within the agency," he said5.
Read the sentence for its order. Lynch names the correction, "not replace people," before anyone raises replacement as the question. The defense arrives three clauses ahead of the attack. Those hours sit on the same ledger as the roughly 2,500 positions his agency's Public Buildings Service cut over the same year.
The Defense Advanced Research Projects Agency financed ARPANET starting in 1969, the predecessor to the modern internet. That network spent two decades moving through military bases and research labs before an ordinary household reached it in the 1990s. The question now is simple. How long do the agencies that process a veteran's claim or a Medicare bill keep waiting? Their share of the money is moving that fast somewhere else.
Rhoden's old question for sports money applies here too, aimed at a federal budget instead of a league. Taxpayers fund both halves of the ledger, the Pentagon's $90.7 billion and the civilian rounding error beside it. Model and cloud vendors profit on both sides at once. The discounted licenses they hand thin-budget agencies build a customer base government will pay full price to keep. A veteran whose medical record passes through an unclassified chatbot carries a risk missing from both numbers.
A $91.8 billion number reads as a government finally moving at the pace of the technology it bought. Set it beside a leaner GSA staff and a VA chat tool that skipped a safety rating. That same number then reads as a defense budget wearing a civilian label. The AILately.com Desk tracks where the dollars and the use cases split inside government, one morning at a time, inside The Signal, its daily email. Sign up, and decide for yourself which number a hundred-billion-dollar AI budget actually describes.
Sources
- James S. Denford, Gregory S. Dawson, and Kevin C. Desouza, "Where does federal AI spending stand in 2026?," Brookings Institution, May 18, 2026, https://www.brookings.edu/articles/where-does-federal-ai-spending-stand-in-2026/
- Yasir Atalan, Erik Tiersten-Nyman, and Benjamin Jensen, "Understanding the U.S. Federal Government's AI Spending," Center for Strategic and International Studies, Sept. 22, 2026, https://www.csis.org/analysis/understanding-us-federal-governments-ai-spending
- Alexandra Kelley, "Agencies report over 3,000 AI use cases in 2025," Nextgov/FCW, April 16, 2026, https://www.nextgov.com/artificial-intelligence/2026/04/agencies-report-over-3000-ai-use-cases-2025/412898/
- Edward Graham, "VA's AI chatbots not designated high-impact, despite clinical use, watchdog says," Nextgov/FCW, June 12, 2026, https://www.nextgov.com/artificial-intelligence/2026/06/vas-ai-chatbots-not-designated-high-impact-despite-clinical-use-watchdog-says/414158/
- Frank Konkel, "GSA No. 2 talks 'million hours challenge,' scaling agency AI efforts," Nextgov/FCW, April 20, 2026, https://www.nextgov.com/artificial-intelligence/2026/04/gsa-no-2-talks-million-hours-challenge-scaling-agency-ai-efforts/412965/
- Sean Michael Newhouse, "Severe staffing cuts at GSA are harming agencies across government, watchdog says," Government Executive, April 17, 2026, https://www.govexec.com/oversight/2026/04/severe-staffing-cuts-gsa-are-harming-agencies-across-government-watchdog-says/412944/