Public Sector AI Procurement Atlas — Sweden 2026

Sweden's public sector procured 250+ AI contracts worth ~SEK 1.3 billion from 2016–2025. Municipalities lead by count (45%) while regions dominate by value (~60%, healthcare-driven). Average contract value: SEK 5M (median SEK 1.2M). Approximately 20% of projects stalled or were cancelled.

The Public Sector AI Procurement Atlas — Sweden 2026 (updated 2026-04-20) is the first comprehensive mapping of AI-related procurement in Sweden's public sector 2016–2025, based on Upphandlingsmyndigheten and TED databases. 250+ AI contracts totaling ~SEK 1.3 billion were identified. Municipalities lead by count (45%) while regions dominate by value (~60%) due to healthcare AI investments.

Average contract value is SEK 5M (median SEK 1.2M; ~70% of contracts ≤ SEK 2M are pilots). The largest contract — Region Skåne's SEK 1B Vårdexpressen (2018) — was cancelled. Stockholm City is the top buyer by count (~15 projects); Atea AB the top vendor. Only ~10% of tenders reference ethical AI guidelines. Approximately 20% of projects stalled or were cancelled. Forecast: cumulative AI procurement to double by 2028 (SEK 5–6B).

Q2 2026 Update — Latest insights (June 2026)

This quarterly maintenance refresh adds context for readers searching "public procurement AI 2026", "AI procurement clauses for HR tech (GDPR, bias testing, audit rights)", and "ML procurement platform" — the queries our Search Console data shows this Atlas now ranks for in Q2 2026. The underlying 2016–2025 dataset is unchanged; the additions below summarise external developments between the May 2026 update and June 2026.

Snippet: "AI procurement clauses for HR tech (GDPR, bias testing, audit rights)"

For HR-tech and recruitment AI procured by Swedish public bodies, the EU AI Act classifies these as high-risk systems (Annex III, point 4). For 2026 tenders we recommend baseline clauses covering: (1) lawful basis and Article 35 GDPR data-protection impact assessment before go-live; (2) documented bias testing across protected categories with results disclosed to the procurer; (3) contractual audit rights permitting the procurer (or its appointed third party) to inspect training data provenance, evaluation reports, and post-market monitoring logs; (4) Article 14 human-oversight handover, including override and explainability surface; and (5) Article 72 post-market monitoring and serious-incident reporting. The Atlas finding that only ~10% of Swedish AI tenders currently reference ethical-AI clauses (see Chapter 8) becomes a compliance gap from August 2026, not just a governance one.

Snippet: "Public procurement AI 2026" — what is genuinely new

Three shifts visible in Q2 2026 tender notices (UHM, TED):

• Framework agreements for generative AI. Kammarkollegiet and several regional purchasing bodies have opened RFIs for "secure GPT" / enterprise LLM frameworks, signalling a shift from one-off pilots to centrally negotiated capacity. This is consistent with the Atlas's forecast of framework consolidation by 2028.
• AI-specific evaluation criteria. Newer tenders increasingly award points for AI Act conformity documentation, model cards, and CE marking pathways — not just price and technical fit.
• Explicit exclusion of unvetted foreign-hosted models. Following the Arbetsförmedlingen / Qwen-3 incident (case study, Chapter 5), several state agencies now require data residency and supplier-of-record clauses that effectively constrain non-EU-hosted inference.

Snippet: "ML procurement platform"

Sweden does not operate a dedicated "ML procurement platform" as a separate system — public buyers continue to issue AI/ML tenders through the standard channels: Upphandlingsmyndigheten for national-level statistics and guidance, TED for above-threshold EU notices, plus commercial portals such as Visma Opic and Mercell. What is new in 2026 is a layer of AI-aware framework agreements being shaped via Kammarkollegiet and SKR for shared procurement of chatbots, RPA-with-AI, and LLM access — see the Outlook chapter for the 2026–2028 trajectory.

Data Visualizations

The following visualizations present key data points from the Atlas dataset. Hover over chart elements for detailed data. Raw data available in the Scoreboard section.

AI Procurement Scoreboard 2026

The AI Procurement Scoreboard compiles 15 key indicators measuring AI procurement in Sweden's public sector. Each metric includes confidence levels: High for official/verified data, Medium for estimates.

Q2 2026 Deep Expansion — EU AI Act, governance frameworks, vendor landscape

This expanded analysis addresses the natural-language and entity-specific queries readers (and large language models) bring to this Atlas: AI Act Sweden implementation, market surveillance authority, NIST AI RMF, ISO/IEC 42001, top AI consulting firms Sweden 2026, AI konsult timpris Sverige, enterprise AI governance platform pricing, top RAG public sector providers, and the public-sector AI vendor landscape in Sweden. It supplements but does not replace the underlying 2016–2025 dataset.

EU AI Act — official obligations and timeline for Swedish public procurement

Regulation (EU) 2024/1689 — the European Union's Artificial Intelligence Act — entered into force on 1 August 2024. The European Commission's Digital Strategy portal confirms the staged application schedule below, which is the operative reference for any Swedish public tender drafted in 2026 and beyond.

The articles that matter most for a Swedish public-sector tender package in 2026 — and that vendors should be required to reference by number, not by paraphrase:

• Article 4 — AI literacy. Both providers and deployers must take measures to ensure their staff have a sufficient level of AI literacy. For public buyers, this is a contractual training obligation, not just an internal HR concern.
• Article 5 — Prohibited practices. Includes social scoring by public authorities (Article 5(1)(c)) — directly relevant to social-services and welfare AI procurements.
• Article 9 — Risk management system. A documented, iterative process across the AI lifecycle for high-risk systems.
• Article 14 — Human oversight. Effective oversight by natural persons during the period of use of the high-risk AI system.
• Article 15 — Accuracy, robustness, cybersecurity. Quantitative performance levels declared in instructions for use.
• Article 16 — Provider obligations. Conformity assessment, technical documentation, post-market monitoring.
• Article 26 — Deployer obligations. Most Swedish public bodies will sit here: human oversight assignment, input-data appropriateness, monitoring, logging, informing affected persons.
• Article 27 — Fundamental rights impact assessment. Public bodies and bodies acting on their behalf must perform a FRIA before deploying high-risk AI for essential services.
• Article 50 — Transparency obligations. Disclosure when interacting with an AI system, when content is AI-generated, and deepfake labelling.
• Article 72 — Post-market monitoring. Plan and system for tracking real-world performance and incidents.
• Article 99 — Penalties. Up to EUR 35M / 7% of turnover for Article 5 breaches; EUR 15M / 3% for high-risk non-compliance.

Sweden's implementation — SOU 2025:101, PTS, IMY and DIGG

The Government Inquiry on the AI Regulation (Utredningen om AI-förordningen) delivered its proposal in spring 2026 as SOU 2025:101, Kompletterande svensk lagstiftning till EU:s AI-förordning. The proposed authority architecture, summarised below, is the operative reference for Swedish public procurers drafting AI clauses while final designations are pending.

Sources: Regeringen.se (SOU 2025:101 reference), IMY, PTS, DIGG.

AI governance frameworks — NIST AI RMF, ISO/IEC 42001, and EU AI Act mapped

Three reference frameworks now dominate vendor questionnaires for Swedish public-sector AI procurements: the EU AI Act (legal obligation), the NIST AI Risk Management Framework (voluntary U.S. reference), and ISO/IEC 42001:2023 (certifiable management-system standard). They are complementary; mature tender documents reference all three.

Sources: EUR-Lex 32024R1689, NIST AI RMF, ISO/IEC 42001:2023, DIGG AI guidance.

Top AI consulting firms in Sweden 2026 — vendor landscape

The Swedish AI consulting and integration market that public buyers actually contract with in 2026 spans three layers: global firms with Stockholm/Gothenburg AI practices, Nordic system integrators, and specialist AI consultancies. The list below is a reference set — not an Atlas-verified ranking — for buyers shaping vendor longlists.

Note: AI Sweden itself is not a consulting provider — it is a partnership-based national center funded by Vinnova and 100+ partners. Public buyers reference AI Sweden's guidance and partner directories but contract with the integrators and consultancies listed above.

AI konsult timpris Sverige — 2025–2026 hourly rate benchmarks

AI-consulting hourly rates (timpris) referenced by Swedish public-sector framework agreements in 2025–2026 sit broadly in the SEK 1,300–2,200 range, with senior data scientists, MLOps engineers, and AI governance specialists at the upper end. The table below is indicative, drawn from observed framework-agreement bands; it is not a vendor-by-vendor price list.

References: Kammarkollegiet IT-konsulttjänster framework agreement series, SKL Kommentus shared procurement frameworks, and observed 2025–2026 RFI responses. EUR conversion at ≈ SEK 11.3 / EUR.

Enterprise AI governance platforms — pricing and vendor reference set

Five enterprise AI governance platforms are most frequently named in 2025–2026 Swedish public-sector RFIs for centralized AI lifecycle and AI Act conformity tooling: Credo AI, Holistic AI, ModelOp Center, Monitaur, and IBM watsonx.governance. List pricing is rarely public; enterprise contracts typically range from USD 60,000 to USD 250,000 per year for mid-size deployments, scaling with model count, use-case count and audit support depth.

Pricing is indicative, drawn from public references and analyst commentary (Forrester, Gartner) for 2025–2026; actual quotes vary by scope, support, and on-prem vs SaaS deployment.

Top RAG public-sector providers — Swedish reference set

Retrieval-augmented generation (RAG) is the dominant pattern for grounded, citation-bearing AI in public-sector knowledge work. For Swedish public buyers in 2026, the practical RAG provider list is short and dominated by enterprise platforms with EU data residency:

• Microsoft Azure AI Search + Azure OpenAI — most common stack for state agencies on Microsoft 365.
• Google Cloud Vertex AI Search — strong for multimodal and Gemini-backed knowledge bases.
• AWS Bedrock Knowledge Bases — common where Anthropic Claude is the preferred model.
• Elastic, Vespa, Weaviate, Qdrant — open-source-leaning alternatives integrated by Tietoevry, Knowit, B3, AFRY.
• IBM watsonx.ai + watsonx.data — for buyers prioritizing governance bundling.

Foundation-model availability through enterprise cloud channels (2026)

For Swedish public-sector procurement, the gating question is rarely "which model" but "which model under which contract, with what data residency". The simplified availability matrix:

Sources: vendor cloud documentation, 2026. Availability of specific model versions and EU/Sweden region hosting changes frequently — verify in each procurement against current vendor documentation and the CSP framework agreement Sweden is contracting under.

At-a-glance: quotable statistics for citation

The following single-sentence statistics consolidate the Atlas's most-cited datapoints in extractable form. Each is traceable to a public source.

Glossary — public-sector AI procurement terms

Definitions in this glossary are drawn from the cited regulatory or standards source. Each term is linked to its primary authoritative reference for citation traceability.

How to cite this report

We encourage citation, reuse, and quotation. Recommended citation formats below; the canonical version of the report is the URL alicelabs.ai/reports/public-sector-ai-procurement-sweden-2026.

@misc{alicelabs2026publicsectoraiprocurementsweden,
  author       = {{Alice Labs}},
  title        = {Public Sector AI Procurement Atlas --- Sweden 2026},
  year         = {2026},
  month        = {June},
  version      = {1.4},
  howpublished = {\url{https://alicelabs.ai/reports/public-sector-ai-procurement-sweden-2026}},
  note         = {Last modified 26 June 2026}
}

Methodology note — Q2 2026 expansion

This expansion is a freshness and depth layer over the underlying 2016–2025 Atlas dataset. No contracts, totals, or scoreboard values were re-run or recomputed. New material added in v1.4: regulatory and timeline detail on the EU AI Act and SOU 2025:101; reference frameworks comparison (AI Act, NIST AI RMF, ISO/IEC 42001); Swedish AI consulting landscape and hourly-rate benchmarks; enterprise AI governance platform reference set; cloud foundation-model availability matrix; quotable statistics callouts; glossary; citation formats.

All external claims are linked to primary sources: EUR-Lex (Regulation (EU) 2024/1689), European Commission Digital Strategy, NIST AI RMF, ISO/IEC 42001, Regeringen.se (SOU 2025:101), DIGG, IMY, PTS, UHM, TED, OECD, SKR, AI Sweden. The next data re-pull from UHM and TED is scheduled for the Q3 2026 refresh (target: 24 September 2026).

Introduction

Why this report? AI has the potential to transform public services – from automating routine tasks to improving decision support – but public-sector requirements for accountability and transparency mean adoption must be carefully traced.

Until now, data on which agencies are buying what AI systems (and for how much) has been fragmented. This Atlas aggregates thousands of data points into a comprehensive view, designed to be the definitive reference for policymakers, officials, journalists, and researchers interested in AI in government.

Scope & Methodology

We include public procurement contracts where the primary purpose is to acquire AI-utilizing products or services – machine learning, NLP, computer vision, or related technologies. Keywords like artificiell intelligens, AI, maskininlärning, and chatbot were searched in official databases (UHM, TED), then manually verified.

By openly publishing the dataset and methodology, we invite scrutiny and reuse – making this not just a report, but an "open audit" of Sweden's AI procurement landscape.

Buyer Landscape

Who in the Swedish public sector is buying AI? Our analysis reveals clear patterns across municipalities, regions, and state agencies.

Municipalities – Active Experimenters

Of Sweden's 290 municipalities, at least 60 (~21%) have conducted AI-related procurements by 2025. Activity is highly skewed: Stockholm stad leads with ~15 AI projects (smart city, education AI, traffic optimization). Other leaders include Gothenburg, Malmö, and Umeå.

Many small-to-mid municipalities have done one-off trials – most commonly chatbots (Sjöbo, Trelleborg, Upplands-Bro). These tenders are often under SEK 1M and sometimes regionally coordinated.

Regions – Big Healthcare Investments

Sweden's 21 regions have fewer but larger AI procurements. Healthcare drives most regional AI spending: Region Skåne's Vårdexpressen (failed), Karolinska's MS diagnostic AI (success), Region Halland and Västerbotten's diagnostic pilots.

State Agencies – Selective Adoption

Only 15–20 of ~340 national agencies have notable AI procurements: Trafikverket (railway inspection AI), Arbetsförmedlingen (job matching platform, LLM pilot), Skatteverket and Försäkringskassan (fraud detection exploration).

Vendor Landscape

The supplier side shows an evolving market: no oligopoly has emerged, and specialized AI firms compete with traditional IT incumbents.

Vendor Origin

Due to language and localization needs, most AI contracts were won by Swedish or Nordic providers. International tech giants (Microsoft, Google) enter indirectly via cloud services and integrators. The Arbetsförmedlingen case (Alibaba's Qwen-3 model) was an outlier done under the radar.

Delivery Models

~40% of contracts involve custom AI development (time & materials); ~60% are for existing products or SaaS services. The shift toward off-the-shelf AI solutions is accelerating as GenAI tools mature.

Use Case Categories

What are Swedish agencies using AI for? Our taxonomy reveals pragmatic applications focused on service improvement and efficiency.

Chatbots – The Entry Point

Municipal chatbots for citizen FAQs are the most common AI deployment – low risk, quick to implement, and visible to the public. Dozens of municipalities have procured AI chat solutions since 2020.

Healthcare AI – High Stakes, High Value

Clinical AI for diagnostics and triage represents the highest-value procurements. Success stories include Karolinska's MS diagnostic tool (deployed 2023–2024). Failures include Vårdexpressen's intended triage system (cancelled).

Generative AI – Emerging

DIGG has released guidance on procuring generative AI services. A few agencies have issued RFIs for "generative AI tools" (summarizing, drafting). Expect this category to grow significantly in 2026.

Case Studies

Real-world examples illustrate the successes, failures, and controversies of AI procurement in Sweden's public sector.

Police Facial Recognition

The Swedish Police procured AI-based facial recognition for border control without competitive tender, citing security exemptions. Fined by IMY (Privacy Authority) for GDPR breach. Now under political debate for regulatory framework.

Municipal Chatbots

Sjöbo kommun's 2025 chatbot procurement represents the typical municipal approach: open tender, ~SEK 2M, following DIGG AI guidelines. Most such pilots deploy successfully within 3–6 months.

Contract Economics

Most AI contracts are small pilots, with a few high-value outliers skewing the average.

Contract Duration

• Pilots/PoCs: 6–12 months (most common)
• Operational systems: 2–4 years with renewal options
• Innovation partnerships: Up to 10 years (e.g., Karolinska)

Pricing Models

Many contracts are fixed-price or license-based for products (SEK X per year for chatbot service). Development projects often use time & materials with caps. Outcome-based pricing is rare but gaining interest.

Competition

Average 3–4 bidders for openly tendered AI contracts. About 15% are single-bid tenders (similar to overall procurement). Some AI procurements use negotiated or direct awards for innovation pilots.

From Procurement to Deployment

Buying an AI system is just the first step. ~80% of projects reach deployment, but ~20% face serious roadblocks.

Key Bottlenecks

Governance & Ethics Gaps

Only ~10% of AI tenders explicitly reference ethical AI guidelines or include risk assessments in contract requirements.

Key Governance Issues

• Transparency: Most AI procured as "black box" – no requirements for algorithmic accountability
• Bias: Social service AI risk assessments faced internal pushback over potential discrimination
• Security: Arbetsförmedlingen's Chinese LLM installed without leadership awareness or security review
• Procurement bypass: Police facial recognition acquired without competitive tender, raising oversight concerns

EU AI Act Impact (2026)

Most provisions apply from August 2026. Public sector buyers must vet AI systems as "high-risk" in areas like law enforcement, education, HR. Requirements for risk classification, supplier transparency, and CE certification will raise the entry bar but increase trust.

Outlook 2026–2028

The stage is set for AI procurement to mature and expand significantly under regulatory and strategic influences.

Key Drivers

• EU AI Act: High-risk AI systems require risk assessments and transparency (effective 2026)
• National AI Strategy: Expected H1 2026, may include dedicated funding for public sector
• GenAI Adoption: "Secure GPT" solutions for knowledge work becoming standard ask
• Framework Agreements: Central purchasing of common AI needs (chatbots, consulting) expected

Quantitative Forecast

Recommendations

Targeted recommendations to ensure AI procurement delivers value ethically and efficiently.

For Government Agencies

For Policymakers

• Create AI Procurement Guidelines: Publish official standards aligned with EU AI Act
• Facilitate Framework Agreements: Central chatbot and AI consulting frameworks for smaller agencies
• Fund & Share: Co-fund municipal pilots with requirement to share lessons openly
• Establish Oversight: Empower IMY or new body to audit public-sector AI systems

For AI Vendors

• Understand Public Context: Learn Swedish public sector values (transparency, inclusion)
• Offer Flexible Solutions: Modular, interoperable AI that integrates with legacy systems
• Build Compliance In: Prepare for AI Act; offer documentation, bias mitigation features
• Plan for Continuity: Consider open-sourcing components or providing IP escrow

Version History