Airys Intelligence Lab · Report
The Untapped Federal AI Market
The conversation around federal AI contracting has been dominated by one headline: the Department of Defense. The numbers are eye-catching — a $13.4 billion standalone AI and autonomous systems budget request for 2026, and DoD AI contract potential value growing 1,605% between 2024 and 2026, now representing 98.9% of all federal AI contract potential value.
Those numbers aren't wrong. They're also misleading for anyone trying to find tractable, near-term, structurally open opportunities in the public market.
Civilian agencies are quietly outpacing defense on growth, even if not on raw dollars:
AI procurement obligation growth · FY2022 → FY2024
More than $3 billion in civilian AI spending flowed through the latest budget cycle. The number of federal agencies holding AI contracts grew from 17 in 2022 to 28 in 2026, with the Department of Health and Human Services holding one of the largest inventories in government.
Reported AI use cases across federal agencies rose 69% in 2025 alone — to 3,611 — more than six times the count from 2023. Far from being a smaller version of the defense market, civilian procurement follows its own logic, priorities, and competitive dynamics with rapid expansion.
Reported federal AI use cases
A more than 6× increase in two years.
Understanding where the real opportunities sit means understanding three layers: what the government wants AI to accomplish, who's already operating in the ecosystem, and who can enter it. Let's walk through each.
The Budget Imperative
The current administration treats AI adoption as a tool for shrinking government, not just modernizing it.
With the federal workforce down more than 317,000 people through layoffs, firings, and buyouts, agencies have an acute, structural need for productivity technology — someone (or something) has to do the work.
The federal government has already signalled:
- The White House's July 2025 AI Action Plan frames AI as key to delivering "the highly responsive government the American people expect and deserve."
- OMB Memoranda M-25-21 and M-25-22 turned that aspiration into procurement law — requiring agencies to track AI performance and cost-effectiveness across the contract lifecycle and treat AI investment data as a core institutional asset.
And it's already paying off. Treasury's Office of Payment Integrity has run ML-based fraud detection since late 2022. The results:
$652.7M
FY2023 recovered
FY2024 prevented or recovered
Nearly a 6× jump
FY2024 breakdown
$4B+
- High-risk transaction prioritization$2.5B
- Check-fraud detection$1B
- Risk-based screening$500M
Competition & the Problem of Concentrated Dependence
The more interesting story isn't how fast government is buying AI — it's how hard it's working to avoid becoming dependent on just a few vendors while doing so.
The GSA OneGov initiative is the clearest example of the tension. Starting in 2025, GSA began offering agencies enterprise AI access — including frontier chatbot licenses — for as little as $1/year. Cheap and easy to adopt. But cheap-to-start doesn't mean cheap-to-leave.
A federal contracting attorney put it plainly to Federal News Network: nominal-rate access isn't really about winning on price — it's about establishing relationships whose value shows up later, in renewals, workflow integration, and switching costs. Once staff are trained and a model is stitched into internal processes, replacing the provider gets harder regardless of what the original contract cost.
Procurement policy is starting to catch up to this risk.
OMB M-25-22 requires agencies to evaluate vendor lock-in at every stage — demos, solicitation language, proposal evaluation, ongoing management — and to negotiate:
- Data portability
- Model portability
- Knowledge-transfer provisions
- Licensing transparency
- Rights to code and models produced during the contract
The memo goes further, directing agencies to actively "adopt procurement practices that encourage competition to sustain a robust Federal AI marketplace," including favoring interoperable products.
Two examples of what happens when it's ignored
April 2026
GAO found FEMA couldn't share certain geospatial model outputs with federal and state partners — it hadn't secured adequate data rights up front, having prioritized speed of acquisition over long-term control.
Pentagon's Maven program
Officials reported ongoing uncertainty about what data/IP rights they'd need to preserve flexibility for future system development.
In response, GAO recommended negotiating data ownership, portability, and intellectual property terms upfront — ensuring agencies can share government-generated outputs, migrate to new vendors, and recompete contracts without excessive switching costs. DOD, DHS, GSA, and VA all concurred with the recommendations.
The government isn't trying to avoid big AI vendors — it's still buying from them at scale. But there's a clear, growing institutional instinct to keep options open: preserve the ability to switch, to benchmark, to negotiate from strength, rather than let early contracting decisions quietly lock in the next decade.
Small Business Participation as Structural Policy
Unlike the informal "we value small business" language common in private-sector procurement, the federal government's commitment is a binding legal requirement.
- The Small Business Act, implemented through FAR Part 19, requires agencies to give small businesses "the maximum practicable opportunity" to compete for federal work.
- This is enforced through contract-level requirements and compliance reporting tied to award and performance.
- Set-aside and sole-source authorities (including SBA programs) let qualifying small firms compete for — or directly receive — contracts outside full-and-open competition, under defined thresholds.
The logic behind it: unchecked concentration among large contractors narrows the vendor base over time and erodes competitive pressure in future cycles. Set-asides exist to keep the supplier base distributed.
It's working, at least by the numbers
of federal AI obligations
captured by small businesses, FY2022–FY2024 — nearly $2 billion total
Small-business AI obligations
- Awards go to systems integrators, consultants, and implementation firms deploying and customizing AI for agency missions, not just developers of frontier AI models.
- Much of this work flows through multi-award contract vehicles (e.g., Alliant 2, IAC MAC, GSA MAS), where firms compete for task orders or participate through teaming arrangements.
The real question isn't whether small businesses are formally included — they are, by law. It's how they actually break into and scale within a market increasingly shaped by complex contracting vehicles, entrenched incumbents, and consolidating AI infrastructure.
The Takeaway
Civilian AI procurement is growing faster than defense.
Backed by real fiscal pressure, real policy infrastructure, and a real (if imperfect) statutory push toward competition and small-business access. For anyone trying to enter the federal AI market, the defense headlines are a distraction from where the doors are actually open.