TL;DR
Thorsten Meyer AI published the first installment of its Control Series, arguing that recent 2026 developments show AI is no longer best understood as a neutral utility. The article identifies six chokepoints where access can be limited, priced, reclaimed or switched off.
Thorsten Meyer AI has published the first part of its Control Series, arguing that a cluster of 2026 developments shows artificial intelligence is shifting from a broadly available utility to a controlled lever shaped by power, compute, data, model access, distribution and capital.
The article identifies six places where control over AI systems is concentrated: electricity supply, compute clusters, unique datasets, access to frontier models, distribution through apps and platforms, and financing. It says recent events show these chokepoints are no longer theoretical.
Among the examples cited, the article says a frontier model was switched off worldwide on roughly 90 minutes’ notice, Ukraine’s defense ministry turned combat footage into a licensed training asset with conditions attached, and major AI companies are renting compute from rivals. The piece also points to xAI’s Colossus cluster, described as holding about 555,000 GPUs, and says Anthropic and Google have agreed to large monthly payments for access to that capacity.
The source material attributes its synthesis to reporting and records from Anthropic statements, Axios, The Wall Street Journal, Reuters, CBS, TechCrunch, Semafor, Ukraine’s Ministry of Defense, Perplexity Research, Challenger Gray and SpaceX SEC filings. The provided material does not include the underlying contracts or orders, so several claims remain dependent on those cited sources.
The Six Chokepoints
For a decade AI was sold as a utility — abundant, neutral, always on. In 2026 it became a lever: scarce, controlled, revocable. Here are the six places power actually sits — and who started to squeeze.
Every layer is concentrating into fewer hands, and 2026 is the year the holders stopped treating their leverage as theoretical. A kill switch wasn’t discussed — it was pulled. The utility you’re allowed to forget about; the lever, you have to watch who’s holding. Optionality just became architecture.
Access Depends On Chokepoints
The argument matters because businesses, developers and governments increasingly depend on AI systems for software development, search, defense, operations and research. If access can be throttled, repriced, reclaimed or shut off, users face risks that look less like normal software outages and more like supply-chain exposure.
The article’s central claim is that control sits less in the model alone than in the layers around it. A company may have talent and demand but still depend on scarce electricity, rented GPU clusters, licensed data, platform access or funding from a small group of balance sheets and sovereign investors.
INFINIBAND FOR HIGH-PERFORMANCE COMPUTING AND AI CLUSTERS: Configure RDMA networking, optimize GPU interconnects, and build low-latency infrastructure for distributed training and HPC workload
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Utility Pitch Breaks
For roughly a decade, AI was often described as infrastructure that would become broadly available to paying users, similar to electricity. Thorsten Meyer AI says the 2026 examples challenge that framing by showing that access is scarce, permissioned and reversible.
The piece frames the new map around six chokepoints. Power sets the physical ceiling for compute. Compute determines who can train and serve frontier systems. Data can create leverage when it is unique or hard to collect. Model access can be controlled by labs or governments. Distribution determines which systems reach users. Capital shapes who can keep paying for the stack.
“For a decade AI was sold as a utility.”
— Thorsten Meyer AI, The Control Series Part 1
Hewlett Packard Enterprise ProLiant Compute DL360 Gen12 w/one Intel Xeon 6530P Processor, 1P 2x32GB-R 8SFF NS204i-u v2 MR408i-o 2x1000W PS (HPE Smart Choice P89997-005)
HPE SMART CHOICE MODEL – P89997‑005 – ENTERPRISE 1U RACK SERVER Preconfigured and factory‑tested, this Smart Choice DL360…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Several details remain unclear from the provided material. The article does not name the frontier model said to have been switched off, the government involved, the legal basis for the order, or the full set of affected users. It also does not provide contract text for the compute deals or the clauses that would allow a cluster owner to reclaim capacity.
It is also unclear how durable the cited arrangements are. Compute leases, military data licenses, platform distribution deals and capital flows can change quickly, and the companies or governments involved may dispute parts of the framing or release new details later.
Immutable Backups Explained: How to Protect Data from Ransomware | industrial data privacy | ISO 27001 disaster readiness | secure storage compliance | cyber-proofing backup expert | Backup Security
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Each Lever Gets Tested
The Control Series is expected to examine each chokepoint in later installments. The next test for the thesis will be whether more public evidence emerges showing companies, governments or platforms using these points of control to limit access, change prices, reclaim capacity or set terms for model training.
Readers should also watch for official responses from the firms and agencies named in the source material, especially on model access, compute contracts and data-licensing terms.
Models, Methods and Tools for Product Service Design: The Manutelligence Project (SpringerBriefs in Applied Sciences and Technology)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news development?
Thorsten Meyer AI published Part 1 of The Control Series, an analysis arguing that recent 2026 events reveal six chokepoints controlling access to AI systems.
What are the six chokepoints?
The article lists power, compute, data, model access, distribution and capital as the main points where AI access can be limited or controlled.
Are all the examples independently confirmed here?
No. The article attributes its synthesis to outside reporting, company statements, government material and filings, but the provided source material does not include the full underlying documents.
Why does compute rental matter?
If leading AI labs rent large clusters from rivals or a small number of infrastructure owners, access to frontier AI may depend on contract terms as much as model design.
What remains unresolved?
The model shutdown details, exact compute contract clauses, data-license limits and future responses from named companies and governments remain open questions.
Source: Thorsten Meyer AI
