TL;DR
Thorsten Meyer AI has narrowed its earlier case for owning AI infrastructure, arguing that most organizations gain more from leading models and vendor-routing tools. The analysis still supports sovereign systems where law, classified work or sensitive regulated data leaves no viable alternative.
Thorsten Meyer AI published an analysis on July 16, 2026, arguing that most organizations should use the best-performing AI model available instead of paying for sovereign infrastructure unless legal or data restrictions require it. The article narrows the publication’s earlier support for model ownership and says many buyers may be overpricing geopolitical risk while underpricing weaker performance, qualification delays and infrastructure costs.
The analysis says the strongest case against sovereignty is the capability gap. It cites vendor-reported results showing Inkling at 77.6% on SWE-bench, compared with 95% for Fable 5, and 63.8% versus 89.5% on Terminal-Bench. The publication cautions that these results are self-reported and await independent replication.
It also argues that sovereign deployment carries a measurable cost through security qualification, specialist staffing, idle computing capacity and slower product releases. Citing its earlier reporting, the publication points to an estimated $75,000 to $100,000 annual staffing cost, a roughly tenfold idle-capacity penalty and large European infrastructure commitments. Those figures come from different sources and do not establish a single cost applicable to every buyer.
The recommendation is divided by legal exposure. Organizations handling classified material, national health data or finance covered by rules such as DORA may face binding restrictions that make sovereign deployment necessary. For most other companies, the article recommends a vendor-neutral routing layer that can redirect requests when a model becomes unavailable.
Against sovereignty: the strongest case for just using the best model
This publication has spent five weeks arguing one thing — and every piece converged. That should bother you. It bothers me. When eight analyses reach the same verdict, you’re not running an analysis. You’re running a thesis, and the evidence has started arriving pre-sorted.
So here’s the case against — argued properly, with the same evidence, turned around. Not a strawman erected to be knocked down. The version a smart CTO would put to me across a table, and which I have not yet answered in public. The claim: for almost everyone, sovereignty is an expensive hedge against a risk they’ve mispriced — and the rational move is to use the best model and get on with it.
Defence · classified · national health data · DORA-bound finance. The foreign-legal-order risk isn’t theoretical and isn’t insurable by other means — it’s a legal gate. No benchmark opens it. Your alternative isn’t a worse model; it’s no deployment at all.
Statistically, you are. You have a reasonable, politically legible, entirely unbudgeted feeling — and an industry built to monetize it. The capability compounds, the tax is real, the opportunity cost is brutal, and 18 days is survivable.
I’ve spent five weeks arguing you should own your stack. The strongest case against says: for most of you, that’s an expensive way to be worse, sold by people whose real product is a feeling. And that case is mostly right. What survives is smaller and sharper — everything above the router line (the qualification programme, the owned cluster, the custom pre-training run, the €11B data centre) you should buy only if a law requires it, never because a narrative does. A router is the sovereignty most people actually need. 90% of the resilience for ~2% of the cost — and it would have made 12 June a non-event. So run the honest test: are you bound, or are you performing?
Capability Gains Challenge Sovereignty Costs
The argument matters because AI procurement can affect both product quality and speed to market. A company choosing a weaker model for political or reputational reasons may accept more failed tasks, longer development cycles and higher operating costs without removing its main exposure to outages or data breaches.
The analysis also draws a line between mandatory sovereignty and voluntary preference. If buyers without binding restrictions dominate demand, suppliers may focus on labels and ownership structures instead of air-gapped systems, exportable weights and other capabilities needed by organizations that cannot legally use foreign-controlled services.
Co-Intelligence: Living and Working with AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Five Weeks of Sovereignty Advocacy
Thorsten Meyer AI said its previous eight analyses had repeatedly favored owning models and infrastructure over depending on an API. Those reports examined ownership, computing capacity, foreign legal authority and the possibility that a vendor or government could cut off access.
The July 16 article was presented as a correction to that pattern. Its central example was a Commerce directive that, according to the publication, removed Fable 5 and Mythos 5 on June 12 before access returned July 1. The publication characterized the episode as an 18-day service degradation with alternative vendors available, rather than proof that every buyer needed a fully owned stack.
Benchmarks and Savings Need Verification
It is not yet clear whether the cited benchmark gaps would persist under independent testing or reflect performance in each buyer’s production environment. The cost estimates also combine different providers, certifications and infrastructure projects, limiting direct comparisons.
The publication’s claim that routing offers 90% of the resilience for about 2% of the cost was not accompanied in the supplied material by a common methodology or buyer-level calculation. The legal boundary can also vary by country, contract, data category and regulator, so organizations may not fit cleanly into either category.
Legal Tests Will Shape Procurement
AI buyers will need to document whether law, regulation or data classification blocks the use of foreign-controlled models before funding owned infrastructure. Those without such restrictions can test multi-provider routing, outage procedures and model portability against their actual continuity requirements.
Independent replication of the cited benchmarks and comparable cost studies would show whether the proposed capability-first approach delivers the performance and savings claimed.
Key Questions
Has Thorsten Meyer AI abandoned AI sovereignty?
No. The publication has narrowed its position, supporting sovereign systems for organizations facing binding legal or security restrictions while questioning their value for other buyers.
Which organizations may still need sovereign AI?
The analysis identifies defence, classified work, national health data and some regulated finance as areas where foreign control may create a legal barrier to deployment.
What does an AI routing layer do?
A routing layer can send requests among multiple model providers, allowing a service to use a fallback when its preferred model is unavailable. It provides operational resilience but does not remove every legal or data-governance risk.
Are the model performance figures independently confirmed?
No. The publication describes the cited results as vendor-reported benchmarks that are awaiting independent replication.
What decision test does the analysis propose?
Buyers should first determine whether sovereignty is required by law or data rules. If it is not, the article argues for comparing the performance, cost and resilience of leading hosted models against an owned deployment.
Source: Thorsten Meyer AI
