Where Should Your AI Run? The 2026 European Briefing

1. Capability and origin

Two independent trackers agree on the size of the gap. Epoch AI finds that since January 2026, the best open-weight models have trailed the best closed models by four months on average, slightly more than the three-month average of 2023 to 2025.1 On the Artificial Analysis Intelligence Index, the leading closed model scores 58 and the leading open model 46.2

For a board, the practical reading is this: the open model you can deploy today performs roughly like the closed frontier of a few months ago. For well-defined business tasks such as document processing, search and drafting, that is often sufficient. For the hardest reasoning work, it is not yet.

The second fact is less comfortable. Every open model near the top of the rankings comes from a Chinese lab. Government actions in Italy, the Netherlands, Germany and the Czech Republic have targeted the hosted DeepSeek app and its data transfers. None concerned weights run on an organisation's own servers. Running weights locally removes that transfer. It does not remove behaviour trained into the model: the US standards body NIST found that DeepSeek models echoed state narratives four times as often as US models.3

2. Licences

Open weights come with very different rights. Mistral Medium 3.5 grants no rights to companies with more than $20 million in monthly revenue unless they buy a commercial licence.4 Meta's Llama 4 policy withholds its multimodal rights from companies based in the EU.5 Alibaba moved its flagship Qwen model from Apache 2.0 to a custom licence.6 Permissive licences remain for, among others, Mistral Small and Large, DeepSeek, Gemma, gpt-oss and Granite.

Exhibit 2
Permissive licences remain available in every region, but not at the very top.

Source: model cards and licence texts, accessed 27 September 2026. Hardware is indicative for production inference. Full detail in the Model Index.

3. Economics

2026 moved the economics against self-hosting. Memory shortages pushed an NVIDIA RTX PRO 6000 from its launch price of $8,565 to about $16,000 on NVIDIA's own store.7 The median rental price of an H200 rose by about 25 percent in twelve months.8 Over the same period, the price of a given level of AI capability through an API kept falling steeply.9

Independent research from Carnegie Mellon finds that owning hardware pays back within months for small models, in about two years for medium models and in about five years for large ones. It is mainly viable at high, steady volumes or under strict data-residency requirements.10 Our own break-even analysis reaches the same conclusion: against a mid-priced closed model, an owned eight-GPU server breaks even only at several billion tokens per month, before staff costs.

4. Regulation and control

The AI Omnibus, in force since 27 July 2026, postponed high-risk obligations to 2 December 2027 for stand-alone systems and to 2 August 2028 for AI embedded in regulated products. Transparency duties under Article 50 apply from 2 August 2026.11 A company that simply deploys an open model is in most cases a deployer under the Act.

The more immediate pressure comes from elsewhere:

• US data access. Under the US CLOUD Act, US providers can be compelled to hand over data they control wherever it is stored. Microsoft France told the French Senate under oath that it could not guarantee this would never happen.12
• DORA. Financial entities must manage ICT third-party risk and plan exits. In November 2025 the European supervisors designated 19 critical ICT providers, including AWS, Google Cloud and Microsoft.13
• NIS2. The Dutch Cyberbeveiligingswet has been in force since 15 August 2026 and brings supply-chain security duties to more than 8,000 organisations.14
• Transfers. The EU-US Data Privacy Framework stands, but an appeal is pending at the Court of Justice. The accurate description is legal uncertainty.15

5. Security

Self-hosting removes third-party access to prompts and outputs, and gives you audit and exit control. It does not make a model safe. Prompt injection tops the OWASP list of LLM risks,16 and the UK National Cyber Security Centre warns it may never be fully mitigated.17 Research by Anthropic and the UK AI Security Institute showed that around 250 poisoned documents can plant a backdoor in models of very different sizes.18 When you run the model, these risks become yours to manage.

6. What leaders should do

1. Classify workloads by data sensitivity. Most AI use can stay on APIs. Reserve private deployment for the data you cannot afford to share.
2. Set a model-origin and licence policy. Decide at board level which origins and licence types are acceptable, before engineering teams choose for you.
3. Start in a European cloud. Rent capacity from an EU-owned provider to prove value before buying hardware at 2026 prices.
4. Treat models as supply chain. Verify weights, prefer safe file formats, red-team before production, and plan for monthly model updates.

Organisations are moving in this direction: Accenture reports that 62 percent of European organisations are seeking sovereign AI solutions,19 and HSBC chose to run Mistral models on its own infrastructure.20

Sources

1. Epoch AI, Open-closed ECI gap, 29 May 2026. epoch.ai/data-insights/open-closed-eci-gap
2. Artificial Analysis, Intelligence Index v4.3 changelog and model pages, September 2026. artificialanalysis.ai/changelog
3. NIST CAISI, Evaluation of DeepSeek AI models, 30 September 2025. www.nist.gov/news-events/news/2025/09/caisi-evaluation-deepseek-ai-models-finds-shortcomings-and-risks
4. Hugging Face, Mistral-Medium-3.5-128B licence. huggingface.co/mistralai/Mistral-Medium-3.5-128B
5. Meta, Llama 4 Acceptable Use Policy. www.llama.com/llama4/use-policy
6. Qwen3.8-2.4T-A95B licence, Hugging Face. huggingface.co/Qwen/Qwen3.8-2.4T-A95B
7. Thunder Compute, NVIDIA RTX PRO 6000 pricing, September 2026. www.thundercompute.com/blog/nvidia-rtx-pro-6000-pricing
8. getdeploying.com, NVIDIA H200 rental prices, accessed 27 September 2026. getdeploying.com/gpus/nvidia-h200
9. Epoch AI, LLM inference price trends, 12 March 2025. epoch.ai/data-insights/llm-inference-price-trends
10. Carnegie Mellon University, cost-benefit analysis of on-premise LLM deployment, arXiv 2509.18101. arxiv.org/abs/2509.18101
11. Lewis Silkin, The Digital Omnibus on AI enters into force, 27 July 2026. www.lewissilkin.com/insights/2026/07/27/the-digital-omnibus-on-ai-enters-into-force-today-102nedo
12. The Register, Microsoft exec admits it cannot guarantee data sovereignty, 25 July 2025. www.theregister.com/off-prem/2025/07/25/microsoft-exec-admits-it-cannot-guarantee-data-sovereignty/458553
13. EIOPA, ESAs designate critical ICT third-party providers under DORA, 18 November 2025. www.eiopa.europa.eu/european-supervisory-authorities-designate-critical-ict-third-party-providers-under-digital-2025-11-18_en
14. Rijksoverheid, Cyberbeveiligingswet in force from 15 August 2026. www.rijksoverheid.nl/actueel/nieuws/2026/07/07/cyberbeveiligingswet-en-wet-weerbaarheid-kritieke-entiteiten-vanaf-15-augustus-2026-van-kracht
15. WilmerHale, Court of Justice to review challenge to EU-US Data Privacy Framework, 1 December 2025. www.wilmerhale.com/en/insights/blogs/wilmerhale-privacy-and-cybersecurity-law/20251201-european-court-of-justice-to-review-challenge-to-eu-us-data-privacy-framework
16. OWASP, Top 10 for LLM Applications 2025. genai.owasp.org/resource/owasp-top-10-for-llm-applications-2025/
17. UK NCSC, on prompt injection, December 2025. www.ncsc.gov.uk/news/mistaking-ai-vulnerability-could-lead-to-large-scale-breaches
18. Anthropic, UK AI Security Institute and Alan Turing Institute, A small number of samples can poison LLMs of any size, October 2025. www.anthropic.com/research/small-samples-poison
19. Accenture, Europe seeking greater AI sovereignty, November 2025. newsroom.accenture.com/news/2025/europe-seeking-greater-ai-sovereignty-accenture-report-finds
20. HSBC and Mistral AI partnership, 1 December 2025. www.hsbc.com/news-and-views/news/media-releases/2025/hsbc-and-mistral-ai-join-forces-to-accelerate-ai-adoption-across-global-bank

Method: desk research of public sources, accessed 27 September 2026. Vendor-sponsored data is labelled as such. This briefing contains no paid content. Corrections: hello@lindstead.com.