Public artificial-intelligence contracts awarded by administrations worldwide, focused on the committed amount. Each case is verified against the official source (USAspending, TED, national procurement registries or government announcement) and records the amount in its original currency, without converting between currencies. The value of the record lies in consolidating who buys what AI, from whom and for how much —the money axis— at a global scale, something no procurement portal does cross-nationally. When the amount is not public, the contract is documented and the amount marked as undisclosed, without estimating it.
Statistical readings derived from the attributes of each recorded case. All figures come from the documented events; amounts are computed only over cases with a sum expressed in the indicated currency, without converting between currencies.
Choropleth by number of forensically or judicially documented cases. Countries with no verifiable public cases remain in the base colour — the absence of events does not equal the absence of surveillance. Hover or click a coloured country to see the cases.
Fifty public AI contracts awarded worldwide reveal two clear patterns: the money arrives late and all at once —36 of the 50 contracts are from 2025— and almost half goes to defence, surveillance, immigration or policing, not to public services for citizens.
If digital fines show what states punish, public contracts show what they buy. This tracker documents fifty AI procurement awards by administrations worldwide, each verified against its official source —USAspending in the United States, TED in the European Union, national procurement registries—, with the focus on the amount committed and who receives it.
The first striking data point is temporal: 36 of the 50 contracts were signed in 2025, and five more already in 2026. Public AI spending has not grown: it has exploded. What for years were isolated pilot projects has become, in barely two fiscal years, a wave of large-scale procurement.
Twenty-one of the fifty contracts —more than four in ten— go to defence, surveillance, immigration or policing. The AI that governments buy serves, for the most part, the state's coercive functions rather than healthcare, education or ordinary administration.
The dominant sector is defence, with eleven contracts, followed at a distance by cross-cutting administration and government data. But grouping defence, immigration, policing and surveillance under the umbrella of security and control functions, they add up to twenty-one: the most frequent destination of public AI money. There appear the US Department of Defense contracts with its frontier-AI labs, those of Immigration and Customs Enforcement (ICE), and data platforms for security forces.
The most recurrent supplier is Palantir, with seven contracts spread across the British NHS, the UK Ministry of Defence, NATO, ICE, the US Army and immigration services. It is followed by Clearview AI and the major model labs —OpenAI and Anthropic— which are beginning to sign contracts directly with administrations. The concentration in few suppliers raises, according to transparency organisations cited in the sources, a risk of the public sector's technological dependence on a handful of private companies.
Unlike fines, here the amounts are hard to add up honestly: different currencies coexist (dollars, pounds, euros, Australian dollars, yuan) and, above all, contractual forms that are not money actually spent. Seven contracts are 'ceilings' —an authorised maximum that may never be executed—, several are framework agreements and at least two were awarded for a symbolic price of one dollar. Adding an uncommitted €200 million ceiling to a firm €30 million contract would produce a misleading figure, so this tracker records the amount in its original currency and form and leaves the execution status in plain view.
Thirty-two of the fifty contracts are listed as firm; the rest divide among unexecuted ceilings, still-open tenders and, in some flagged cases, no-bid awards or discrepancies between the public announcement and the official notice. That 'status' field is deliberately the most detailed in the tracker: in public procurement, the gap between what is announced and what is executed is where the story usually hides.
Geographically, the United States leads with seventeen contracts, followed by the United Kingdom with ten and Spain with six; the rest scatter among China, Canada, France and a long tail of countries with one or two cases. That Anglo-Saxon predominance partly reflects where more is spent, but also where public procurement is more transparent and traceable: USAspending and TED publish data many other national systems do not open, which makes documenting those contracts easier and documenting the rest harder.
As in the ecosystem's other trackers, a country's absence does not equal an absence of spending. It means its AI procurement has left no verifiable public trail —and that opacity is, in itself, a relevant data point about the accountability of algorithmic spending.
U.S. Citizenship and Immigration Services (USCIS) began in late October 2025 a contract with Palantir for under $100,000 to implement 'Phase 0' of a 'vetting of wedding-based schemes' (VOWS) platform. Though the amount is small, it illustrates how contracts that may later expand begin, and the pattern of awarding to a vendor already embedded in federal infrastructure.
U.S. Immigration and Customs Enforcement (ICE) awarded Palantir a $30 million sole-source contract (no competitive bidding) to build 'ImmigrationOS', a system that cross-references IRS, Social Security, passport and licence-plate-reader data to identify and track deportation targets. ICE justified the direct award claiming Palantir was the 'only source' able to build it. The prototype was due September 2025 and the contract runs through September 2027. An emblematic case for the absence of competition and its civil-liberties implications.
U.S. Immigration and Customs Enforcement (ICE) awarded Clearview AI a $9.2 million contract in September 2025 for facial-recognition services. Clearview has been fined or banned in Greece, Italy, France, the UK, Canada, Australia and Austria, but its use by US federal agencies continues. It is part of a surge in federal spending on commercial facial recognition.
Within ICE's 2025 surveillance-spending surge, the agency awarded Cellebrite an $11 million contract for tools that unlock phones and extract their data, essential for field operations. It is part of an ICE budget that surged to $28.7 billion that year.
ICE awarded BI2 Technologies a $4.6 million contract for iris-scanning smartphones for its agents, deployed in the field via the Mobile Fortify app, which allows scanning faces and fingerprints on the spot. Part of the agency's biometric-surveillance expansion.
OpenAI agreed with the U.S. General Services Administration (GSA) to offer ChatGPT Enterprise to federal executive-branch workers for a year for just $1. The symbolic price is a deliberate penetration strategy: creating public-sector dependence by offering the tool almost free, anticipating future revenue. Days later, Anthropic responded with an identical $1 offer for Claude. An emblematic case of how the acquisition price may not reflect a contract's real strategic value.
Anthropic offered its Claude model to U.S. Government agencies for just $1 for a year, extending to all three branches (executive, legislative and judicial), shortly after OpenAI's identical offer. AI companies are competing to undercut each other and, presumably, create public-sector dependence. It illustrates the race among vendors to embed themselves in the administration at near-zero cost.
xAI, Elon Musk's company, completed the CDAO's quartet of awardees in July 2025 with a $200 million ceiling agreement, marking its entry into US defense AI procurement.
The US Department of Defense finalised in May 2025 an expansion of the Maven system contract with Palantir up to $1.3 billion. The Maven Smart System integrates AI for intelligence processing and target generation, and has been deployed on classified networks. It is one of the largest individual US defense AI contracts and exemplifies the growth of contracts through successive expansions.
The Department of Defense announced a one-year contract with OpenAI Public Sector LLC for $200 million to develop frontier-AI prototypes in warfighting and enterprise domains. It was the first AI contract publicly listed on the Department of Defense website, weeks before the CDAO's multiple batch.
Within the same CDAO batch, OpenAI received an agreement with a $200 million ceiling to prototype frontier-AI capabilities in warfighting and enterprise domains. OpenAI had already secured a one-year $200 million contract in June 2025 —the first publicly listed on the Department of Defense website—, making this its second major defense deal within weeks.
Google was among the four awardees in the CDAO's July 2025 batch, with a $200 million ceiling agreement to develop AI agents across several Department of Defense mission areas.
The Pentagon's Chief Digital and AI Office (CDAO) awarded Anthropic a two-year prototype other-transaction (OT) agreement with a $200 million ceiling, to develop frontier AI capabilities for national security. It is part of a batch of four simultaneous awards to frontier-AI companies. It is a spending 'ceiling', not a firm amount: actual outlay depends on the task orders issued.
Defense company Anduril secured a $100 million defense contract to deploy advanced AI systems in national-security missions, as part of its collaboration with OpenAI on counter-drone (CUAS) systems. It marked the entry of frontier generative AI into weapons applications.
U.S. Customs and Border Protection (CBP) awarded Clearview AI a $225,000 contract in February 2026 for 'tactical targeting'. In the same period, lawmakers introduced the 'ICE Out of Our Faces Act' to ban ICE and CBP from using facial recognition, with no real prospect of passing. It illustrates the tension between expanding biometric spending and attempts to regulate it.
The US Army granted Palantir an Enterprise Agreement worth up to $10 billion for commercial software, allowing the acquisition of data integration, analytics and AI services. It consolidates 75 prior awards into a single instrument and is one of the US Government's largest software and AI commitments. It is an aggregate ceiling of the contracting vehicle, not a firmly committed outlay.
The US Army, via BAE Systems and INSCOM, awarded a $150,000 subcontract to Clearview AI in December 2024, part of a larger $482 million Army intelligence contract. Special Forces (Green Berets) use Clearview's facial recognition to identify targets. A revealing case of military —not just police— use of commercial biometrics, with no federal law restricting it.
The Police & Crime Commissioner for Northamptonshire awarded a £20 million contract to CDS Support to use AI in examining evidence and police analysis. It is one example of AI adoption in UK police forces at the regional level.
NHS England announced in 2023 a £330 million investment in the Federated Data Platform (FDP), awarded to Palantir with support from Accenture, PwC and others, over seven years. However, the official award notice set the contract at £182.2 million ending February 2027 with extension options. The discrepancy between the announced figure (£330m) and the official-notice figure (£182.2m) is a textbook case on the difference between announced and committed spending. The contract, highly controversial over health-data privacy, has a review break clause in spring 2027.
National Highways, the UK roads agency, contracted Deloitte for £35 million for its 'Chief Data Office', including the development of AI capabilities. It exemplifies the pattern analysts note: much public AI spending goes to consultancies and generalist integrators, not to direct software purchase.
The UK Ministry of Defence's contract with Palantir tripled to £240 million. As debated in Parliament (Hansard), that expansion occurred without a new competitive process, prompting criticism over the lack of competition. It illustrates the pattern of contracts that grow through extensions rather than new tenders.
The UK's Met Office awarded Microsoft a ten-year contract worth £1.03 billion for a supercomputer, the UK's largest AI buyer by contract value since 2018 according to analyst Tussell. The contract underpins compute-intensive climate and predictive modelling.
The largest AI contract awarded by the UK Government in 2025 was a £234 million deal with LGC Group (formerly the Laboratory of the Government Chemist), focused on developing a national measurement system. It tops the £573 million the UK committed in AI contracts up to July 2025.
The UK tax authority (HMRC) signed a ten-year £175 million contract with Quantexa, a London-founded AI company, to modernise its data infrastructure and detect tax fraud. The Government presented it as a deliberate counterpoint to the Palantir model: a British company, allied capital and data that never leaves the HMRC environment. It falls within the Sovereign AI Unit launched in 2025 with £500 million.
The UK Department for Work and Pensions (DWP) contracted UiPath for nearly £10 million over three years to automate administrative tasks. The contract drew controversy: Amnesty International UK criticised the DWP's 'unhealthy obsession' with AI in reforming universal credit and disability benefits, warning it trapped some claimants in 'bureaucratic limbo'. An example of public AI in social benefits with direct impact on vulnerable people.
The UK's Crown Commercial Service expanded its artificial-intelligence Dynamic Purchasing System (DPS), doubling its expected value from £240 million to £480 million (incl. VAT) and extending it to early 2029, in its fifth extension since its 2022 creation. The DPS is one of the UK public sector's principal vehicles for buying AI services from specialist suppliers: unlike a static framework, it allows new providers to be added continuously. Over a third of UK AI contracts since 2018 were channelled through such frameworks. An example of the pattern of aggregate contracting vehicles that grow through expansion.
The University of Bristol awarded Hewlett Packard Enterprise (HPE) a £176 million contract to build Isambard AI, a supercomputer for AI-safety research, aiming to place the UK in a leadership position in the field.
The National Bank of Poland (Narodowy Bank Polski) tendered the purchase of an 'artificial intelligence environment' through the EU procurement portal. The total estimated value was PLN 5,064,359, but the final awarded value was PLN 2,333,840 (excl. VAT) —less than half—. It is a valuable example of the difference between the estimated and the actually contracted amount after competitive tender, an 'outcome' datum rarely consolidated.
NATO awarded Palantir a $30 million contract to deploy its Maven Smart System (MSS NATO), one of the Atlantic alliance's first major AI acquisitions. It marks the expansion of US defense AI systems into multilateral bodies.
Japan's Ministry of Economy, Trade and Industry (METI) allocated ¥72.5 billion (about $470 million) to five companies to develop an AI supercomputer and reduce reliance on US technology. METI funds up to half of the investments: Sakura Internet is the largest beneficiary with ¥50.1 billion, followed by telecoms operator KDDI with ¥10.2 billion. It aims to cut Japan's dependence on US cloud providers like AWS and Azure.
India's Government awarded ₹988.6 crore, within the IndiaAI Mission, to the consortium led by IIT Bombay —with Tech Mahindra, Fractal Analytics and BharatGen— to develop a sovereign 1-trillion-parameter large language model (LLM). It is part of India's strategy to build its own foundation models, within the IndiaAI Mission's overall ₹10,300 crore budget approved in 2024.
Israel's Ministry of Internal Security awarded A2Z Smart Technologies (military division A2Z Military Solutions) a multi-year (four-to-five-year) contract worth over $1.5 million for the service and maintenance of complex systems for the Israeli Prison Service. It is one of the few Israeli public technology contracts with a disclosed figure; much of Israel's surveillance ecosystem is not transparent about its amounts.
France's Ministry of the Armed Forces notified in January 2026 a framework agreement with Mistral AI allowing the armed forces, internal directorates and affiliated public institutions (such as the Atomic Energy Commission, ONERA and the Navy's Hydrographic Service) to access Mistral's AI technology. The technology runs exclusively on French infrastructure, not on foreign or commercial clouds, to ensure control of critical data. The amount was not disclosed. It positions Mistral as the sovereign European alternative to OpenAI and Anthropic, contrasting with the Pentagon controversy.
France's Defence Procurement Agency (DGA) awarded Airbus Defence and Space a framework contract worth up to €50 million to integrate AI components into the weapons, information, communication and cybersecurity systems of the French armed forces. The agreement, also involving Airbus Helicopters, is an execution vehicle for the AMIAD agency (defense AI, created May 2024) and aims to reduce reliance on non-European providers.
The European Commission awarded a contract allowing EU institutions to procure sovereign cloud services for up to €180 million over 6 years. The four awardees are European: Post Telecom (with CleverCloud and OVHcloud), StackIT, Scaleway and Proximus (with S3NS —a Thales–Google Cloud joint venture—, Clarence and Mistral). It was launched as a competition under the Cloud III Dynamic Purchasing System. It reinforces the bet on European digital sovereignty against US hyperscalers.
Spain's public business entity Red.es (state authority) tendered the service for developing, implementing and maintaining artificial-intelligence services and process automation, with an estimated value of €2,331,186 (excl. VAT) and a 36-month duration, via open procedure. At the time of recording it appears as a tender notice, not a final award. Data from the EU's official procurement registry (TED).
The Directorate-General for Telecommunications and Digitalisation of the Government of Navarre awarded, by negotiated procedure without a call for tender, the Axon digital-evidence platform (2025-2028) to 3JUP Security S.L., for an estimated value of €316,815.25 (excl. VAT). The non-competitive procedure for a platform from body-camera and taser maker Axon illustrates how dependence on a specific vendor justifies direct awards. Data from the EU's official procurement registry (TED).
Madrid City Council's Autonomous IT Body awarded Atos, in March 2025, the IT-services contract for the maintenance, evolution and support of its artificial-intelligence platform, for €1,915,598. The contract runs three years (extendable by one) and includes developing a generative-AI agent ecosystem to optimise administrative processes and citizen services. Amount from the procurement portal; awardee and details confirmed by Spanish specialist press.
The Central Procurement Board (Ministry of Finance, for the Ministry of the Interior) awarded the maintenance of the Argos police-query application (2025-2027) to I.C.A. Informática y Comunicaciones Avanzadas S.L. for €509,514.29, in a restricted procedure with 9 bids received. Argos is a query system used by Spanish security forces. Data from the EU's official procurement registry (TED), confirming buyer, winner, exact amount and number of bids.
Spain's National Markets and Competition Commission (CNMC) launched BRAVA, a pioneering AI tool to detect rigged tenders by analysing its database of over 3.5 million public contracts, with real-time access to companies' registry data. The aim is to identify anticompetitive awards before they materialise. The consulted source does not detail the development cost. Another example of public AI serving to monitor the procurement system itself.
Barcelona's Municipal Institute of Innovation and Technology (IMI) tendered a contract to develop three artificial-intelligence pilots for the Municipal Tax Institute, with sustainable-procurement measures, via open procedure (bid deadline 16 February 2026). The TED tender notice consulted does not include the estimated value. An example of AI adoption in local tax administration. Data from the EU's official procurement registry (TED).
Baidu Intelligent Cloud (百度智能云) won the tender for Shenyang's data-circulation and application infrastructure project (沈阳数据流通应用基础设施, lot 1) for ¥364 million in January 2025, per the AI-model tender monitor Smart Hyperparameters. It was one of the three largest AI-model contracts of the month in China. The government sector (政务) has the largest aggregate amount in Chinese AI procurement. Data from specialised Chinese-market monitoring, not a Western official registry.
SenseTime (商汤科技) won the integrated construction-and-operation project for Kunming's AI empowerment centre (昆明人工智能赋能中心建设运营一体化项目) for nearly ¥200 million, one of the largest individual AI-model contracts in China in 2024 per the Smart Hyperparameters monitor. SenseTime is one of the Chinese AI firms sanctioned by the US over its surveillance role.
Fujian Shida Group (福建实达集团) won the tender for the national-level Fuzhou New Area AI scenario-application project (国家级福州新区人工智能场景应用项目) for ¥211 million in January 2025. One of the three largest AI contracts of the month in China per the Smart Hyperparameters monitor, amid the boom in intelligent-computing-centre (智算中心) projects tied to AI models.
Baidu (百度) won the AI-application-support component of the informatisation-construction project for the All-China Federation of Trade Unions' Service Centre (中华全国总工会机关服务中心信息化建设工会数智化建设工程) for ¥31.58 million, Baidu's largest single contract in the first nine months of 2024 per Smart Hyperparameters. In that period Baidu won 36 AI-model projects totalling ¥148.5 million.
ChileCompra published tender ID 869591-2-LP25, 'Document Analysis through Artificial Intelligence Models', to incorporate large language models (LLMs) able to analyse unstructured documents on the Mercado Público portal —which moves over $17 billion a year, 5.3% of Chile's GDP— and detect anomalies pointing to corruption. The contract amount does not appear as a closed figure in the consulted tender notice. It is a notable case of a State using AI to audit its own spending.
Canada's government awarded a CAD 350 million public-service contract to Dayforce to replace the troubled Phoenix pay system, with automation and AI components. It is the largest of Canada's recent individual federal AI and tech outlays.
Canada's federal government invested CAD 240 million in Cohere, the AI company founded by Aidan Gomez —co-author of the seminal transformer-architecture paper—, as a bet on a national language-model champion. It is part of the over 800 million Ottawa has spent on AI technology since 2023, per the government's own data.
Australia's federal government signed in 2018 a five-year, AU$1 billion agreement with IBM as a whole-of-government technology partner, including joint innovation programmes in AI, quantum computing and cybersecurity. It lets all agencies (Tax Office, Home Affairs, Defence, Human Services) access emerging technologies. It is an early historical case of large-scale public procurement with an AI component.
Abu Dhabi's government signed a multi-year agreement with Microsoft and Core42 (a G42 group company) to implement a sovereign cloud system able to process over 11 million daily digital interactions. It falls within the Abu Dhabi Government Digital Strategy 2025-2027, backed by a digital-infrastructure investment of AED 13 billion (about $3.54 billion), aiming to be the world's 'first fully AI-native government' by 2027. The specific contract amount with the providers was not broken out.
Public contracts whose primary object is the acquisition, development, deployment or consultancy of artificial-intelligence systems, awarded by a public administration (national, regional or local) in any country, are recorded. Criteria: (1) a verifiable official source must exist —procurement portal, agency announcement or budget registry—; (2) the amount is stated in the contract's original currency and never converted between currencies, to avoid exchange-rate distortions; (3) when the contract sets a spending ceiling rather than a firm amount, it is flagged as a ceiling; (4) if the amount is not public, the contract is included anyway with the amount marked as undisclosed, without estimating it; (5) the 'outcome' datum —extensions, cuts, differences between announced and official-notice amounts, terminations— is prioritised for its informational value. The record does not judge the appropriateness of the spending; it documents its existence and magnitude.