TL;DR
Thorsten Meyer AI has challenged its earlier support for owning AI infrastructure, arguing that most companies gain more from stronger models and multi-provider routing. It says sovereign systems remain justified for organizations facing legal, classified-data or national-security restrictions, but many cost and performance claims still lack independent verification.
Thorsten Meyer AI reversed the broad thrust of five weeks of pro-sovereignty commentary on July 16, arguing that most companies should use the strongest available AI models instead of owning their entire technology stack. The publication said sovereign infrastructure remains justified for defense, classified data, national health systems and legally restricted finance, but described it as an expensive performance penalty for many other buyers.
The analysis divides organizations into two groups: those bound by law or data restrictions and those choosing sovereignty as a precaution. For the first group, it says foreign ownership, jurisdiction or export controls can block deployment regardless of model quality. For the second, it argues that multi-provider routing and business-continuity planning can address outages and vendor dependence at far lower cost.
Thorsten Meyer AI cited reported benchmark gaps between systems it called Inkling and Fable 5, including 77.6% versus 95.0% on SWE-bench and 63.8% versus 89.5% on Terminal-Bench. The publication cautioned that the figures came from Artificial Analysis and vendor tables, were partly self-reported and were awaiting independent replication. They should be read as evidence cited by the author, not settled comparisons.
The publication also cited higher qualification, staffing and idle-compute costs for sovereign systems. Its central operational example was an alleged restriction affecting two named models from June 12 to July 1. It characterized the episode as an 18-day service degradation with fallbacks available, though the supplied material does not provide the underlying directive or enough documentation to verify the event independently.
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?
Model Quality Versus Legal Control
The argument matters because companies are deciding whether to spend on owned clusters, local hosting and qualification programs or direct that money toward products and customers. If the cited capability gaps hold, choosing a weaker sovereign model could reduce software-agent success rates and delay deployment. If legal exposure is binding, however, a higher-performing foreign model may be unavailable regardless of its benchmark score.
The analysis also draws a distinction between operational resilience and sovereign control. Routing requests among providers may limit the effect of an outage, price increase or model withdrawal, but it cannot resolve every jurisdictional, secrecy or government-access concern. Readers should treat the proposed router as a continuity measure rather than full legal independence.
ACADEMY Models USS Enterprise CV-6 Battle Midway 1/700, Model Kits
- Assembly required: Requires assembly and painting
- Scale: 1/700 scale model
- Dimensions: 14.9 inches long, 2 inches wide
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Five Weeks of Pro-Sovereignty Reporting
Thorsten Meyer AI said its previous eight analyses had repeatedly supported the principle “own the model, not the API”, examining ownership, compute capacity, foreign control and the ability of suppliers to withdraw access. The July 16 article was presented as a deliberate challenge to that editorial thesis, not as a new law, regulatory decision or independently commissioned market study.
The publication referenced European cloud-certification costs, infrastructure spending, ownership limits and survey data indicating broad interest in sovereignty. It contrasted a reported 72% sovereignty figure attributed to CISPE with a Gartner finding that deployment decisions were concentrated in three regulated sectors. Those references support the author’s interpretation, but the supplied material does not include the full surveys, methodologies or comparison periods.
“Your alternative isn’t a worse model; it’s no deployment at all.”
— Thorsten Meyer AI, on regulated users
Benchmarks and Costs Need Verification
It remains unclear whether the cited benchmark differences persist in real company workloads, since the publication acknowledges that some results are self-reported. The supplied source also does not document how its cost comparisons were normalized across hardware, utilization, staffing, security and contract terms.
The claim that routing delivers 90% of resilience for about 2% of the cost is presented without a calculation that readers can inspect here. The scope of the reported June model restriction, the availability of substitutes and the effects on customers also remain unconfirmed from the supplied material.
Legal Tests Before Infrastructure Spending
Organizations following the analysis would next need to determine whether law, contract or data classification prevents the use of foreign-controlled services. Those without such restrictions could test multi-model routing, document fallback performance and compare the result with the cost of self-hosting. Regulated buyers would still need legal review, security validation and sector-specific approval before deployment.
The wider argument will depend on independent benchmark replication, transparent total-cost comparisons and evidence from future supplier restrictions. Those findings will show whether the publication’s narrowed position holds beyond the examples cited in its July 16 article.
Key Questions
Is Thorsten Meyer AI saying sovereignty never matters?
No. It supports sovereign deployment for legally bound organizations, including classified defense, national health and some financial workloads. Its objection concerns companies choosing costly ownership without a specific legal or security requirement.
What does a model router do?
A router can send requests to more than one AI provider and redirect traffic when a service fails or becomes unavailable. It can improve continuity, but it does not create full control over model weights, infrastructure or jurisdiction.
Are the reported model comparisons confirmed?
Not independently in the supplied material. The publication says the benchmark figures include vendor-reported results and await replication. Performance may also differ across private datasets and production tasks.
How should a company decide between sovereignty and model performance?
The first test is whether law, classification rules or contracts restrict foreign-controlled models. If they do not, the analysis recommends comparing model performance, routing options and full ownership costs rather than treating sovereignty as an automatic requirement.
Source: Thorsten Meyer AI
