Industry View · Government & Public Sector
Government is the slowest large adopter of AI and, suddenly, one of the most aggressive. A federal AI Action Plan, $1-per-agency frontier-model deals, and a $10bn Palantir Army contract have collapsed a decade of procurement caution into eighteen months — even as $186bn in annual improper payments shows how much the legacy machine still leaks.
Treasury ML fraud tools recovered $4B+ in FY2024; UK Humphrey ~£20m/yr
FedRAMP-High and IL5 accredited cloud is the gate every government AI workload must pass; cloud held ~51% of the 2025 deployment mix and is the fastest-growing segment.
OpenAI, Anthropic and Google priced Enterprise models at $1 per agency through GSA to win default-tool status across all three branches.
Fusing siloed legacy records into usable, governed data is where most public value — and most failure — lives; bad data is GAO's top barrier to AI in fraud detection.
Targeting/ISR (Maven), payment integrity, benefits triage and citizen chatbots are the live deployments; Treasury credits AI with $4bn+ recovered in FY2024.
Agencies are deploying faster than they can measure outcomes or staff oversight, with DOGE-era errors a cautionary tale; the public sector trails private AI hiring by ~5 years.
01 · The thesis
Public-sector AI is a study in divergence. The addressable market is modest by technology standards — estimates for AI in government and public services cluster between $19.7bn (Global Market Insights) and $26.4bn (Future Market Insights) for 2025, growing 16–19% annually — yet the strategic stakes are outsized. Governments are the largest data holders, the largest payers and, in most economies, the largest employers, so even single-digit efficiency gains are worth tens of billions. The UK has put a number on it: digital and AI could save up to £45bn a year, 4–7% of public-sector spending, though Whitehall's own later estimate halved that to roughly £23.6bn. The 2025 inflection was political, not technological. The Trump administration's July 2025 Winning the Race AI Action Plan and a cluster of executive orders reframed federal AI procurement as a national-priority sprint; the GSA's OneGov deals then handed agencies ChatGPT, Claude and Gemini for $1 or less. The effect is a land grab: frontier labs are buying distribution at a loss, the incumbent integrators are racing to stay relevant, and a $186bn improper-payments problem sits there as the clearest ROI case anyone has ever handed a government CIO.
FedRAMP-High and IL5 accredited cloud is the gate every government AI workload must pass; cloud held ~51% of the 2025 deployment mix and is the fastest-growing segment.
OpenAI, Anthropic and Google priced Enterprise models at $1 per agency through GSA to win default-tool status across all three branches.
Fusing siloed legacy records into usable, governed data is where most public value — and most failure — lives; bad data is GAO's top barrier to AI in fraud detection.
Targeting/ISR (Maven), payment integrity, benefits triage and citizen chatbots are the live deployments; Treasury credits AI with $4bn+ recovered in FY2024.
Agencies are deploying faster than they can measure outcomes or staff oversight, with DOGE-era errors a cautionary tale; the public sector trails private AI hiring by ~5 years.
02 · The two clocks
Procurement is moving fastest. GSA's OneGov initiative compressed a process designed around competition into weeks: ChatGPT and Claude at $1 per agency, Gemini at $0.47, all added to the Multiple Award Schedule in August 2025. The speed has a cost — Ask Sage filed a GAO protest (B-423827.1) arguing the loss-leader pricing routed through reseller Carahsoft circumvents competitive procurement. Deployment is real but uneven. Treasury's machine-learning fraud tools prevented or recovered over $4bn in FY2024; the IRS now uses AI for audit selection, identity-theft filters and taxpayer chatbots; the UK's Humphrey suite is expected to save ~£20m a year analyzing 500 consultations, equal to 75,000 civil-servant workdays. These are pilots scaling, not yet a transformed state. Oversight is the slow clock. GAO warns AI fraud tools are only as good as the data behind them, and the public sector posts roughly one AI job for every 174 in the private sector — a five-year maturity gap. DOGE-era automation has already produced faulty reports from thin human oversight, the cautionary counterweight to the procurement sprint.
03 · Public players & exposure
We plot the listed players on two editorial axes — how exposed each is to AI disruption, against how ready its data, brand and position are to be the answer. The figures in the table are sourced; the placement is our read.
04 · Private flagships
The companies attacking this industry AI-first, with disclosed funding where available:
A 10-year enterprise agreement consolidating ~75 separate contracts, removing pass-through fees and standardizing AI-enabled analytics across Army operations — Palantir's shift from insurgent vendor to incumbent infrastructure.
OpenAI, Anthropic and Google priced Enterprise models at $1 or less per agency across all three branches, with USAi as a no-cost sandbox — a deliberate distribution play to make each lab the default federal tool.
The Pentagon's flagship computer-vision program fuses Anduril's Lattice mesh, Palantir's AIP and (until phase-out) Anthropic's Claude on AWS to move sensor data into analyst workflows.
Machine-learning models screening returns, transactions and check schemes prevented or recovered funds at scale — the clearest dollar-denominated ROI case in federal AI.
A bundle of GenAI tools (Consult, Redbox) for analyzing consultations and briefing, deployed across Whitehall and the Scottish Government as the operational arm of the AI Opportunities Action Plan.
A pre-negotiated Other Transaction Agreement letting DoD components spin up computer-vision and generative-AI decision tools at volume pricing, co-funded by the Chief Digital and AI Office.
05 · Signals
06 · The exposure read
AI rewards clean, structured advantage and punishes friction. The line runs through who owns the data, the brand and the customer — and who is merely a step the technology can route around.
Sources
AI in government & public services — 2025 market size by estimate
Estimates vary by scope and methodology. Sources: Global Market Insights, InsightAce, Future Market Insights (2025-2026 reports). (USD billions, 2025)
AI Action Plan reframes the buyer
"Winning the Race" plus three executive orders make accelerating federal AI procurement national policy, rescind Biden-era guidance, and add an anti-"woke AI" contracting clause for LLM vendors.
The $1 frontier-model deals
GSA OneGov puts ChatGPT and Claude at $1/agency and Gemini at $0.47; Ask Sage's GAO protest argues the loss-leader pricing breaks competitive procurement.
Defense-tech capital surge
Anduril hits a $61bn valuation on a $5bn Series H; defense-related startups raised ~$13.6bn by mid-2026, on pace to double 2025 — VC is now comfortable selling to the state.
Improper payments hit $186bn
GAO pegs FY2025 federal improper payments at $186bn, up 15%, with health programs over half — sharpening the fiscal case for AI payment integrity even as DOGE automation draws scrutiny.
Anthropic designated supply-chain risk
The DoD orders Claude phased out of all military use within six months — a reminder that government distribution wins can reverse on policy, not product.
A 10-year enterprise agreement consolidating ~75 separate contracts, removing pass-through fees and standardizing AI-enabled analytics across Army operations — Palantir's shift from insurgent vendor to incumbent infrastructure.
OpenAI, Anthropic and Google priced Enterprise models at $1 or less per agency across all three branches, with USAi as a no-cost sandbox — a deliberate distribution play to make each lab the default federal tool.
The Pentagon's flagship computer-vision program fuses Anduril's Lattice mesh, Palantir's AIP and (until phase-out) Anthropic's Claude on AWS to move sensor data into analyst workflows.
Machine-learning models screening returns, transactions and check schemes prevented or recovered funds at scale — the clearest dollar-denominated ROI case in federal AI.
A bundle of GenAI tools (Consult, Redbox) for analyzing consultations and briefing, deployed across Whitehall and the Scottish Government as the operational arm of the AI Opportunities Action Plan.
A pre-negotiated Other Transaction Agreement letting DoD components spin up computer-vision and generative-AI decision tools at volume pricing, co-funded by the Chief Digital and AI Office.
| Company | Stance | The sourced fact |
|---|---|---|
| Palantir Technologies PLTR | Embedded incumbent | FY2025 U.S. government revenue grew 55% YoY to $1.855bn; in July 2025 the U.S. Army awarded a 10-year, $10bn enterprise agreement consolidating ~75 contracts (Palantir; FedSavvy). |
| Microsoft MSFT | Platform default | Led the AI-in-government market with over 10% share in 2025 and in Sept 2025 launched a US-government offer bundling Copilot and Azure AI with discounts through Sept 2026 (GMInsights; Microsoft). |
| Booz Allen Hamilton BAH | AI services leader | Ranked the #1 provider of AI services to the U.S. government by GovWin/Deltek (FY2022–24 obligations); ~$600m of annual revenue is AI-attributed, targeting $1bn (Booz Allen; ExecutiveBiz). |
| Anduril Industries PVT | Defense disruptor | Raised a $5bn Series H in 2026 at a $61bn valuation (double its mid-2025 $30.5bn mark), $11.4bn raised to date; its Lattice mesh feeds the Maven Smart System (Crunchbase; Wikipedia). |
| Accenture Federal Services ACN | Consulting in the crosshairs | #11 in the 2025 Washington Technology Top 100 at ~$5.2bn in federal contracts, but absorbed $193m+ in DOGE contract cancellations as the administration targets consulting spend (Wash. Tech; TBRI). |
| Leidos LDOS | Scaled integrator | Topped the 2025 Washington Technology Top 100 at $11.7bn in contracts with a record $46.3bn backlog; largely insulated from DOGE with only ~$56.5m terminated (Wash. Tech; TBRI). |
| Scale AI PVT | Data-layer pick | DoD raised its CDAO Other Transaction Agreement ceiling 5x from $100m to $500m in 2026 as demand outran scope; Meta took a 49% non-voting stake in June 2025 (Washington Technology). |
| OpenAI PVT | Distribution grab | Offered ChatGPT Enterprise to all federal agencies for $1/agency for a year via GSA OneGov (Aug 2025), part of a $200m DoD frontier-AI award shared with Anthropic, Google and xAI (FedScoop). |
| Anthropic PVT | Win then setback | Matched the $1 GSA deal and runs Claude inside Maven, but on 4 Mar 2026 the DoD designated Anthropic a supply-chain risk, ordering its products phased out of military use within six months (FedScoop; Wikipedia). |
| Tyler Technologies TYL | State & local play | The dominant U.S. state-and-local government software vendor, now embedding purpose-built AI across permitting, courts and 311 civic services as agencies modernize legacy systems (Tyler Technologies). |